<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0" xmlns:cc="http://cyber.law.harvard.edu/rss/creativeCommonsRssModule.html">
    <channel>
        <title><![CDATA[Stories by Marco Suma on Medium]]></title>
        <description><![CDATA[Stories by Marco Suma on Medium]]></description>
        <link>https://medium.com/@marcsuma?source=rss-ea35b10f9152------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*p6ULgvMJcFL9yvuiDOmxzA.jpeg</url>
            <title>Stories by Marco Suma on Medium</title>
            <link>https://medium.com/@marcsuma?source=rss-ea35b10f9152------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Sun, 22 Nov 2026 08:56:27 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@marcsuma/feed" rel="self" type="application/rss+xml"/>
        <webMaster><![CDATA[yourfriends@medium.com]]></webMaster>
        <atom:link href="http://medium.superfeedr.com/" rel="hub"/>
        <item>
            <title><![CDATA[Come ho costruito il mio piano di investimento (e perché non vado mai nel panico)]]></title>
            <link>https://marcsuma.medium.com/come-ho-costruito-il-mio-piano-di-investimento-e-perch%C3%A9-non-vado-mai-nel-panico-654dbcccf824?source=rss-ea35b10f9152------2</link>
            <guid isPermaLink="false">https://medium.com/p/654dbcccf824</guid>
            <category><![CDATA[personal-finance]]></category>
            <category><![CDATA[italiano]]></category>
            <category><![CDATA[stock-market]]></category>
            <category><![CDATA[financial-planning]]></category>
            <category><![CDATA[investing]]></category>
            <dc:creator><![CDATA[Marco Suma]]></dc:creator>
            <pubDate>Sun, 25 Oct 2026 10:16:38 GMT</pubDate>
            <atom:updated>2026-10-25T10:16:38.814Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="Illustrazione di un piano di investimento: un obiettivo, un nucleo nell’S&amp;P 500 e quattro regole, accanto a un grafico in cui gli acquisti mensili raddoppiano dopo un crollo" src="https://cdn-images-1.medium.com/max/1024/1*pnywg8zEltCh-3qGnkX5TQ.jpeg" /></figure><p><a href="https://medium.com/p/ab9ec44bdf0f"><em>This article is also available in English.</em></a></p><p>Nota: quella che segue è la mia esperienza personale e non costituisce una consulenza finanziaria. Non sono un consulente abilitato e ciò che ha funzionato per la mia situazione, il mio orizzonte temporale e la mia tolleranza al rischio potrebbe non funzionare per la Sua.</p><h3>In sintesi</h3><p>Sono nato in una famiglia umile e ho costruito un patrimonio nel lungo periodo grazie a un piano, non a operazioni fortunate. Il piano ha un unico obiettivo (la pensione e il futuro dei miei figli), un nucleo (la maggior parte del portafoglio investita nell’S&amp;P 500) e alcune regole attorno a esso: acquistare settori silenziosi supportati dalle mie ricerche, investire ogni mese e di più dopo un crollo, non vendere mai opzioni che non posso coprire e vendere solo quando cambia la mia convinzione su un’azienda. L’orizzonte lungo è il motivo per cui non vado nel panico. Le regole sono il motivo per cui questa calma non è fede cieca.</p><h3>Da dove sono partito</h3><p>È stato mio padre a ispirarmi a investire. Lo faceva in modo molto semplice, acquistando soltanto azioni italiane, e nessun altro intorno a me investiva. Non è insolito: l’Italia non è mai stata un Paese con una forte cultura dell’investimento, e il denaro era qualcosa da risparmiare, non da far fruttare.</p><p>Ho iniziato a fare trading nel 2012 e i miei primi errori sono stati quelli tipici. Acquistavo azioni perché se ne parlava nelle notizie, reagivo in modo eccessivo a ogni movimento e controllavo i prezzi ogni giorno. Quando leggevo una notizia, il mercato l’aveva già incorporata nei prezzi, e il controllo quotidiano trasformava la normale volatilità in uno stress costante.</p><p>Ciò che ha cambiato tutto è stato fissare un unico obiettivo: costruire un patrimonio per la mia pensione e per il futuro dei miei figli. Un obiettivo distante 10 o più anni trasforma la maggior parte delle notizie di mercato in rumore. Ogni regola che segue nasce da quell’obiettivo.</p><h3>Come ho imparato</h3><p>Non avevo alcuna formazione finanziaria, quindi me la sono costruita. Queste sono le fonti che hanno dato forma al mio piano:</p><ul><li><strong>The Intelligent Investor (libro):</strong> acquistare con un margine di sicurezza e considerare gli sbalzi d’umore del mercato come un’opportunità, non come un segnale. Ci tornerei? Sì.</li><li><strong>Business Adventures (libro):</strong> aziende e mercati falliscono e si riprendono secondo schemi ricorrenti, perché il comportamento umano si ripete. Ci tornerei? Sì.</li><li><a href="https://www.youtube.com/@AndreiJikh"><strong>Andrei Jikh</strong></a><strong> (YouTube):</strong> l’interesse composto, i dividendi e l’investimento mensile sono idee semplici che funzionano, se si ha la costanza di seguirle. Ci tornerei? Sì.</li><li><strong>All-In (podcast):</strong> come gli investitori esperti del settore tecnologico interpretano i trend macroeconomici e i nuovi settori prima che diventino mainstream. Ci tornerei? Sì.</li><li><strong>Rich Habits (podcast):</strong> la ricchezza si costruisce un passo alla volta: risparmiare, investire, evitare che il tenore di vita cresca insieme al reddito. Ci tornerei? Sì.</li><li><strong>My First Million (podcast):</strong> esistono molti modi per guadagnare oltre allo stipendio, e le nicchie silenziose sono spesso le migliori. Ci tornerei? Sì.</li><li><a href="https://seekingalpha.com/"><strong>Seeking Alpha</strong></a><strong> (articoli/podcast):</strong> leggere sia la tesi rialzista sia quella ribassista su un’azienda prima di acquistarla. Ci tornerei? Sì.</li><li><a href="https://www.ft.com/"><strong>Financial Times</strong></a><strong> (sito web):</strong> il contesto globale conta più dei titoli del giorno. Ci tornerei? Sì.</li></ul><p>Che cosa eviterei, col senno di poi: le notizie di mercato quotidiane, i consigli sui social media, qualsiasi fonte che venda certezze e, soprattutto, gli articoli acchiappaclic del tipo “5 azioni da comprare subito”. Quando un titolo finisce in quella lista, la notizia è di solito già incorporata nel prezzo.</p><p>L’abitudine più utile non è stata una fonte. È stata fare le mie ricerche prima di ogni acquisto, invece di comprare ciò che qualcun altro consigliava.</p><h3>Il metodo</h3><p>Il mio piano si basa su un nucleo e quattro regole. Ognuna esiste per impedirmi di prendere decisioni emotive.</p><h4>Il nucleo: diversificazione attorno all’S&amp;P 500</h4><p>La maggior parte del mio portafoglio è investita nell’S&amp;P 500. È la base, ed è noiosa di proposito. Tutto il resto di questo articolo (settori silenziosi, big tech, criptovalute, oro, opzioni) è diversificazione attorno a quel nucleo, dimensionata in modo che nessuna singola scommessa possa danneggiare il piano se dovesse fallire.</p><h4>Regola 1: acquistare ciò che è silenzioso, ma solo se la ricerca lo conferma</h4><p>Cerco settori di cui i media non parlano e nei quali credo nella tendenza di lungo periodo. Il silenzio, da solo, non è un segnale. Alcuni settori sono silenziosi perché stanno morendo, quindi ogni candidato deve prima superare la mia analisi.</p><ul><li><strong>Come individuo il silenzio:</strong> procedo dall’alto verso il basso. Prima studio il contesto macroeconomico e identifico le aree con le maggiori probabilità di successo nel lungo periodo. Poi, all’interno di quelle aree, scelgo le aziende di cui ancora nessuno parla.</li><li><strong>Che cosa deve confermare la ricerca:</strong> ricavi in crescita, margini sani, debito sostenibile, una tendenza strutturale che il mercato ignora.</li><li><strong>Esempio:</strong> Twist Bioscience (TWST), nel settore delle bioscienze, è rimasta a lungo ignorata. L’ho acquistata durante la pandemia. Oggi segna +378% nel mio portafoglio. Intel (INTC) segue uno schema diverso: un operatore solido e affermato da tempo, sul quale il mercato ha più volte reagito in modo eccessivo. Oggi segna +241% nel mio portafoglio. L’ho acquistata per la prima volta anni fa e ho continuato ad aumentare la posizione ogni volta che il mercato reagiva in modo eccessivo. Nota di trasparenza: detengo tuttora entrambi i titoli.</li></ul><h4>Regola 2: investire ogni mese, e di più dopo un crollo</h4><p>Utilizzo il piano di accumulo del capitale (PAC, in inglese dollar cost averaging): un importo fisso ogni mese, qualunque cosa faccia il mercato. Dopo un crollo, ne aumento il peso.</p><ul><li><strong>La mia regola:</strong> quando un titolo scende molto più di quanto meriti dopo un crollo di mercato, e la mia ricerca della Regola 1 conferma ancora la tesi, lo considero una chiara opportunità di acquisto sul ribasso e raddoppio l’importo investito. A essere sincero, non ho un criterio misurabile per stabilire che cosa sia “eccessivo”. Si tratta di una valutazione basata su anni di osservazione dei mercati, e funziona solo perché la ricerca deve prima confermarla.</li><li><strong>Da dove proviene la liquidità aggiuntiva:</strong> dal mio reddito da lavoro. Risparmio ogni mese e mantengo una parte dei risparmi come liquidità pronta per opportunità come queste, così un crollo non mi costringe mai a vendere qualcos’altro per poter acquistare.</li><li><strong>Dove la applico:</strong> soprattutto agli ETF. Anche i grandi titoli consolidati, come gli hyperscaler, sono buoni candidati. Sui singoli titoli la tesi deve comunque reggere, altrimenti mediare al ribasso significa afferrare un coltello che cade.</li></ul><p>Una precisazione onesta: secondo una ricerca di Vanguard, l’investimento in un’unica soluzione batte il PAC circa due volte su tre, perché i mercati salgono più spesso di quanto scendano. Continuo comunque a utilizzare il PAC, perché il suo valore è comportamentale. Mi ha permesso di continuare ad acquistare quando l’istinto mi diceva di vendere.</p><h4>Regola 3: vendere opzioni solo quando posso coprirle</h4><p>Vendo put garantite da liquidità (cash-secured put) e call coperte (covered call) per ottenere un reddito modesto ma costante.</p><ul><li><strong>Put:</strong> vendere una put significa essere pagati per inserire un ordine di acquisto al di sotto del prezzo attuale. Vendo put solo su attività che desidero possedere a quello strike, con la liquidità già accantonata. In un vero crollo verrò assegnato a un prezzo superiore a quello di mercato, e lo accetto in anticipo. Funziona come una versione retribuita della Regola 2.</li><li><strong>Call:</strong> una call coperta limita il mio potenziale di guadagno, e i miei risultati migliori sono arrivati da settori silenziosi che in seguito sono saliti con forza. Per questo vendo call solo con strike molto al di sopra del prezzo corrente. Il premio è inferiore, ma conservo la maggior parte del potenziale rialzo e, se un titolo arriva a quel livello, sono lieto di realizzare i profitti.</li><li><strong>Mai allo scoperto:</strong> ogni opzione è coperta da azioni o da liquidità. La vendita di opzioni fa guadagnare poco e spesso, ma occasionalmente può far perdere molto. La disciplina di avere sempre pronti il sottostante o la liquidità è la vera lezione.</li></ul><h4>Regola 4 (la lezione che ho imparato tardi): sapere perché si possiede un titolo e che cosa porterebbe a venderlo</h4><p>Una convinzione senza una regola di vendita è il modo in cui si accompagna un titolo fino allo zero. Non ho mai scritto una tesi formale. Le mie ricerche erano nella mia testa e hanno funzionato, ma hanno reso più difficile valutare gli insuccessi. Se ricominciassi, questa sarebbe la prima abitudine che aggiungerei: due righe scritte prima di ogni acquisto, sul perché compro e su che cosa dimostrerebbe che ho torto. La mia regola di vendita è semplice: vendo solo quando cambia la mia convinzione sull’azienda, mai perché il prezzo è sceso. E succede. Quando accade, scelgo il momento della vendita in modo che la perdita riduca le imposte sulle plusvalenze realizzate altrove, ove consentito dalla legge.</p><h4>Tutto ciò che ho provato</h4><ul><li><strong>S&amp;P 500:</strong> esito positivo. Per me è un nucleo irrinunciabile: le 500 maggiori aziende statunitensi. Fallisce solo se fallisce l’America, e in quel caso avremmo tutti problemi ben più grandi.</li><li><strong>Grandi aziende tecnologiche:</strong> esito positivo. Il mondo corre veloce e tutto può succedere, ma queste aziende sono enormi e tecnologicamente all’avanguardia. Per me sono una delle scommesse più solide.</li><li><strong>Vendita di put e call sulle big tech:</strong> esito positivo. Si colloca sul versante rischioso e può sembrare una scommessa. Per me funziona solo perché sono disposto a detenere queste aziende se vengo assegnato, oppure a venderle a un determinato prezzo.</li><li><strong>Oro (tramite ETF):</strong> esito positivo. La mia copertura per quando tutto il resto va storto. Attualmente +100% nel mio portafoglio.</li><li><strong>ETF su singoli Paesi:</strong> esito negativo. ETF sull’India (IMVP): la prospettiva di crescita era reale, ma mi sono convinto che la corruzione e la scarsa organizzazione potessero rappresentare un freno permanente. La sola crescita macroeconomica non basta.</li><li><strong>Singoli titoli in settori silenziosi:</strong> esito positivo. Twist Bioscience: avevo intuito presto l’ascesa delle biotecnologie e delle bioscienze, quindi era solo questione di tempo (+378%).</li><li><strong>Criptovalute:</strong> esito positivo. Investo dal 2016 e ho mantenuto le posizioni attraverso diversi cicli profondi.</li><li><strong>Trading algoritmico (forex):</strong> esito negativo. Soprattutto per mancanza di risorse: richiede molto più tempo e capitale per tentativi ed errori di quanto ne avessi, e il vantaggio competitivo continua a ridursi con l’aumentare della concorrenza. Non escludo di riprenderlo.</li><li><strong>Obbligazioni:</strong> esito positivo. In gran parte le evito: a 10–15 anni dalla pensione preferisco ancora la crescita alla stabilità.</li><li><strong>Titoli di moda:</strong> esito negativo. Virgin Galactic (SPCE): l’ho acquistata a un prezzo molto alto durante l’euforia e non si è mai ripresa. Il settore spaziale potrebbe ancora cambiarne le sorti, ma detenerla oggi è più una scommessa che un investimento. L’esatto opposto della Regola 1.</li></ul><h3>Che cosa è andato storto e come ho venduto</h3><p>Non tutti gli investimenti hanno funzionato. Alcuni sono falliti e li ho venduti.</p><ul><li><strong>L’insuccesso:</strong> la cannabis, tramite l’ETF YOLO. Era un settore di moda, ma la mia ricerca indicava che il mercato era ancora lontano dal pieno sviluppo, con ulteriori margini di crescita. A decretarne la fine è stata la regolamentazione: i governi sono rimasti in silenzio, la legalizzazione si è arenata e il settore è morto, almeno per ora.</li><li><strong>La vendita:</strong> ho mantenuto la posizione per cinque anni. Poi il mondo è cambiato intorno a essa: guerre, prezzi del petrolio e governi con priorità ben più importanti. Era evidente che la cannabis non sarebbe stata tra queste, quindi la mia convinzione è cambiata e ho venduto.</li><li><strong>La lezione:</strong> i titoli e i settori di moda sono <strong>quasi sempre una pessima idea</strong>. Quando tutti ne parlano, il prezzo incorpora già la notizia, ed è solo questione di tempo prima che scenda. Ecco perché la Regola 1 parte dal silenzio. La seconda lezione riguarda il rischio normativo: quando un settore dipende dalle decisioni dei governi, nessuna ricerca può incorporarlo completamente nel prezzo.</li></ul><p>Ho utilizzato le minusvalenze per compensare le imposte sulle plusvalenze, ove consentito dalla legge. L’ordine conta: ho venduto perché la mia convinzione era cambiata, non per risparmiare sulle imposte. Il risparmio fiscale è stato una conseguenza.</p><h3>Perché vendo raramente e perché non vado nel panico</h3><p>Vendo raramente perché il mio obiettivo è lontano 10–15 anni. Un calo del 30% è doloroso da osservare, ma non cambia la possibilità di raggiungere la pensione o di garantire il futuro dei miei figli. Cambia soltanto il prezzo di ciò che acquisterò il mese successivo.</p><p>Questa è la vera risposta alla domanda “come fa a non andare mai nel panico?”. Non è un tratto caratteriale. È il risultato di tre decisioni prese in anticipo:</p><ol><li><strong>Un obiettivo abbastanza lontano</strong> da far sì che un crollo sia uno sconto, non una minaccia.</li><li><strong>Regole per acquistare durante un crollo</strong> (Regola 2 e Regola 3), così ho qualcosa da fare invece di qualcosa da temere.</li><li><strong>Sapere che cosa mi porterebbe a vendere</strong> (Regola 4), così mantenere una posizione è una decisione, non una negazione.</li></ol><p>La prova più dura è stata il COVID. Ogni titolo che possedevo è sceso, ed è stato davvero doloroso da vedere. Non ho venduto. Al contrario, il mio trasferimento a Singapore mi ha fornito nuova liquidità, che ho investito sul mercato mentre i prezzi erano ancora relativamente bassi. Quel periodo è la mia risposta a chi pensa che questo piano funzioni solo in un mercato rialzista: le regole hanno retto quando il mercato non l’ha fatto. E il COVID è stato breve. Il mercato ribassista del 2022 è stato la prova più lunga: quasi un anno di prezzi in calo, senza alcuna rapida ripresa all’orizzonte. Ho resistito anche allora e ho continuato a investire ogni mese.</p><p>Un portafoglio di lungo periodo è utile anche al di fuori del mercato. Rafforza la propria posizione quando si negozia con una banca, perché dimostra anni di disciplina, non soltanto uno stipendio.</p><h3>Gli strumenti che ho sviluppato</h3><p>Sono un ingegnere, quindi cerco continuamente di trasformare queste regole in codice. La maggior parte è ancora un prototipo e un progetto è fallito, ma svilupparli mi costringe a rendere esplicito il mio giudizio.</p><ul><li><a href="https://github.com/marcosuma/mok-algo-trader"><strong>mok-algo-trader</strong></a><strong>:</strong> crea strategie sul forex con analisi tecnica o machine learning, ne esegue il backtest e poi le applica in tempo reale con denaro virtuale. Stato: pubblico, in pausa (finora senza successo).</li><li><strong>Backtest del PAC ponderato:</strong> confronta su dati reali l’investimento in un’unica soluzione, il PAC semplice e il PAC rafforzato dopo i crolli. Stato: privato, in fase di prototipo.</li><li><strong>Screener dei settori silenziosi:</strong> assegna un punteggio ai settori confrontando l’attenzione mediatica con i fondamentali. Stato: privato, in fase di prototipo.</li><li><strong>Tracker delle opzioni:</strong> monitora i premi incassati, il rischio di assegnazione e la liquidità accantonata per ciascuna put aperta. Stato: privato, in fase di prototipo.</li></ul><h3>Conclusioni</h3><ol><li>Parta da un unico obiettivo di lungo periodo. È la fonte della Sua calma.</li><li>Trasformi il Suo istinto in regole scritte: quando acquistare, quando acquistare di più, quando vendere.</li><li>Consideri gli insuccessi come dati e si informi sulle norme fiscali del Suo Paese.</li><li>Se sa programmare, provi a trasformare le Sue regole in strumenti. Anche quando lo strumento non funziona, scoprirà dove le Sue regole sono vaghe.</li></ol><h3>Restiamo in contatto</h3><p>Se ha trovato utile questo articolo, mi aggiunga su <a href="https://www.linkedin.com/in/marcosuma">LinkedIn</a>. Se desidera approfondire, può prenotare una sessione con me su <a href="https://mentorcruise.com/mentor/marcosuma/">MentorCruise</a>.</p><p>Possiamo parlare di crescita professionale nel settore tecnologico o di come imposto un piano di lungo periodo. Tenga presente, tuttavia, che non si tratta di una consulenza personalizzata sugli investimenti.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=654dbcccf824" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[How I Built My Investment Plan (and Why I Never Panic)]]></title>
            <link>https://marcsuma.medium.com/how-i-built-my-investment-plan-and-why-i-never-panic-ab9ec44bdf0f?source=rss-ea35b10f9152------2</link>
            <guid isPermaLink="false">https://medium.com/p/ab9ec44bdf0f</guid>
            <category><![CDATA[stock-market]]></category>
            <category><![CDATA[financial-planning]]></category>
            <category><![CDATA[personal-finance]]></category>
            <category><![CDATA[investing]]></category>
            <category><![CDATA[algorithmic-trading]]></category>
            <dc:creator><![CDATA[Marco Suma]]></dc:creator>
            <pubDate>Sun, 25 Oct 2026 10:16:25 GMT</pubDate>
            <atom:updated>2026-10-25T10:16:25.382Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="Illustration of an investment plan: one goal, an S&amp;P 500 core and four rules, next to a chart where monthly buys double after a crash" src="https://cdn-images-1.medium.com/max/1024/1*ph3XCYa0MWFLALthOq1TNA.jpeg" /></figure><p><a href="https://medium.com/p/654dbcccf824"><em>Questo articolo è disponibile anche in italiano.</em></a></p><p>Note: this is my personal experience, not financial advice. I am not a licensed advisor, and what worked for my situation, horizon and risk tolerance may not work for yours.</p><h3>tl;dr</h3><p>I was born in a humble family and built long-term wealth with a plan, not with lucky trades. The plan has one goal (retirement and my kids&#39; future), a core (most of my portfolio in the S&amp;P 500) and a few rules around it: buy quiet sectors my research believes in, invest monthly and more after crashes, never sell options I can&#39;t cover, and sell only when my belief in a company changes. The long horizon is the reason I don&#39;t panic. The rules are the reason that calm is not blind faith.</p><h3>Where I started</h3><p>My dad is the one who inspired me to invest. He did it in a very basic way, buying only Italian stocks, and nobody else around me invested at all. That&#39;s not unusual: Italy has never been a country with a strong investing culture, and money was something you saved, not something you put to work.</p><p>I started trading in 2012, and my first mistakes were the usual ones. I bought stocks because they were in the news, I overreacted to every move, and I checked prices every day. The news had already been priced in by the time I read it, and checking daily turned normal volatility into constant stress.</p><p>What changed everything was setting one goal: build wealth for my retirement and my kids&#39; future. A goal 10 or more years away turns most market news into noise. Every rule below comes from that goal.</p><h3>How I learned</h3><p>I had no financial education, so I built one. These are the sources that shaped my plan:</p><ul><li><strong>The Intelligent Investor (book):</strong> Buy with a margin of safety, and treat the market&#39;s mood swings as an opportunity, not a signal. Would I read it again? Yes.</li><li><strong>Business Adventures (book):</strong> Companies and markets fail and recover in patterns, because human behaviour repeats. Would I read it again? Yes.</li><li><a href="https://www.youtube.com/@AndreiJikh"><strong>Andrei Jikh</strong></a><strong> (YouTube):</strong> Compounding, dividends and investing every month are simple ideas that work if you stick to them. Would I read it again? Yes.</li><li><strong>All-In (podcast):</strong> How experienced tech investors read macro trends and new sectors before they go mainstream. Would I read it again? Yes.</li><li><strong>Rich Habits (podcast):</strong> Wealth is built step by step: save, invest, avoid lifestyle creep. Would I read it again? Yes.</li><li><strong>My First Million (podcast):</strong> There are many ways to make money beyond a salary, and quiet niches are often the best ones. Would I read it again? Yes.</li><li><a href="https://seekingalpha.com/"><strong>Seeking Alpha</strong></a><strong> (articles/podcast):</strong> Read both the bull and the bear case on a company before buying. Would I read it again? Yes.</li><li><a href="https://www.ft.com/"><strong>Financial Times</strong></a><strong> (website):</strong> Global context matters more than daily headlines. Would I read it again? Yes.</li></ul><p>What I would skip in hindsight: daily market news, social media tips, any source selling certainty, and above all the clickbait &quot;5 stocks to buy now&quot; articles. By the time a stock makes that list, the news is usually already priced in.</p><p>The most useful habit was not a source. It was doing my own research before every purchase, instead of buying what someone else recommended.</p><h3>The framework</h3><p>My plan is a core and four rules. Each one exists to stop me from making an emotional decision.</p><h4>The core: diversification around the S&amp;P 500</h4><p>Most of my portfolio sits in the S&amp;P 500. It&#39;s the foundation, and it&#39;s boring on purpose. Everything else in this post (quiet sectors, big tech, crypto, gold, options) is diversification around that core, sized so that no single bet can hurt the plan if it fails.</p><h4>Rule 1: buy what is quiet, but only if the research agrees</h4><p>I look for sectors the media is not talking about, where I believe in the long-term trend. Silence alone is not a signal. Some sectors are quiet because they are dying, so every candidate has to pass my research first.</p><ul><li><strong>How I spot silence:</strong> I work top-down. First I study the macro environment and identify the areas most likely to succeed in the long run. Then, inside those areas, I pick the companies nobody is talking about yet.</li><li><strong>What the research must confirm:</strong> growing revenue, healthy margins, manageable debt, a structural trend the market ignores.</li><li><strong>Example:</strong> Twist Bioscience (TWST), in the bioscience sector, sat ignored for a while. I bought it during the pandemic. It is now +378% in my portfolio. Intel (INTC) is a different pattern: a strong, long-established player that the market has repeatedly overreacted on. It is now +241% in my portfolio. I first bought it years ago and kept adding to the position every time the market overreacted. Disclosure: I still hold both.</li></ul><h4>Rule 2: invest every month, and more after a crash</h4><p>I use dollar cost averaging (DCA): a fixed amount every month, whatever the market does. After a crash, I increase the weight.</p><ul><li><strong>My rule:</strong> when a stock drops far more than it deserves after a market crash, and my research from Rule 1 still confirms the thesis, I treat it as a clear buy-the-dip opportunity and double the amount I invest. To be honest, I don&#39;t have a measurable trigger for &quot;exaggerated&quot;. It&#39;s a judgment call built on years of watching markets, and it only works because the research has to agree first.</li><li><strong>Where the extra cash comes from:</strong> my job income. I save every month and keep part of it as liquidity ready for opportunities like these, so a crash never forces me to sell something else to buy.</li><li><strong>Where I apply it:</strong> mostly ETFs. Large, established stocks like the hyperscalers are good candidates too. On single stocks, the thesis must still hold, otherwise averaging down is catching a falling knife.</li></ul><p>An honest caveat: Vanguard&#39;s research found lump-sum investing beats DCA about two thirds of the time, because markets go up more often than down. I still use DCA, because its value is behavioural. It kept me buying when my instinct said sell.</p><h4>Rule 3: sell options only when I can cover them</h4><p>I sell cash-secured puts and covered calls for a small, steady income.</p><ul><li><strong>Puts:</strong> selling a put is getting paid to place a buy order below today&#39;s price. I only sell puts on assets I want to own at that strike, with the cash set aside. In a real crash I will be assigned above the market price, and I accept that in advance. It works as a paid version of Rule 2.</li><li><strong>Calls:</strong> a covered call caps my upside, and my best winners came from quiet sectors that later ran. So I only sell calls at strikes far above the current price. The premium is smaller, but I keep most of the upside, and if a stock gets there, I&#39;m happy to take profits.</li><li><strong>Never naked:</strong> every option is covered by shares or cash. Option selling earns a little often and can lose a lot occasionally. The discipline of always having the asset or the liquidity ready is the real lesson.</li></ul><h4>Rule 4 (the lesson I learned late): know why you own it, and what would make you sell</h4><p>Conviction without a sell rule is how people ride a stock to zero. I never wrote a formal thesis. My research lived in my head, and it worked, but it made the failures harder to judge. If I started again, this is the first habit I&#39;d add: two written lines before every purchase, why I buy and what would prove me wrong. My sell rule is simple: I sell only when my belief in the company changes, never because the price dropped. It does happen. When it does, I time the sale so the loss reduces the taxes on gains I&#39;ve realized elsewhere whenever legally possible.</p><h4>Everything I tried</h4><ul><li><strong>S&amp;P 500:</strong> Worked. For me it&#39;s a must-have core: the 500 largest US companies. It fails only if America fails, and then we&#39;d all have much bigger problems.</li><li><strong>Big tech companies:</strong> Worked. The world moves fast and anything can happen, but these companies are huge and technologically ahead. To me they are one of the most solid bets.</li><li><strong>Selling puts and calls on big tech:</strong> Worked. It sits on the dangerous side and can look like a bet. It works for me only because I&#39;m happy to hold these companies if I get assigned or happy to sell at a certain price.</li><li><strong>Gold (via ETF):</strong> Worked. My hedge for when everything else goes wrong. Currently +100% in my portfolio.</li><li><strong>Country ETFs:</strong> Failed. India ETF (IMVP): the growth story was real, but I came to believe corruption and weak organisation could be a permanent drag. Macro growth alone is not enough.</li><li><strong>Single stocks in quiet sectors:</strong> Worked. Twist Bioscience: I saw the rise of biotech and bioscience early, so it was just a matter of time (+378%).</li><li><strong>Crypto:</strong> Worked. Investing since 2016, holding through several deep cycles.</li><li><strong>Algo trading (forex):</strong> Failed. Mostly a lack of resources: it needs far more time and capital for trial and error than I had, and the edge keeps shrinking as competition grows. I don&#39;t rule out going back to it.</li><li><strong>Bonds:</strong> Worked. I mostly stay away: with 10 to 15 years to retirement, I still prefer growth over stability.</li><li><strong>Hyped stocks:</strong> Failed. Virgin Galactic (SPCE): I bought at a very high price during the hype, and it never recovered. Space could still change its course, but holding it today is more a bet than an investment. The opposite of Rule 1.</li></ul><h3>What failed, and how I sold</h3><p>Not every investment worked. Some failed, and I sold them.</p><ul><li><strong>The failure:</strong> cannabis, through the YOLO ETF. It was a hyped sector, but my research told me the market was still far from fully developed, with more growth ahead. What killed it was regulation: governments stayed silent, legalization stalled, and the sector died, at least for now.</li><li><strong>The sell:</strong> I held for five years. Then the world changed around it: wars, oil prices, and governments with much bigger priorities. Cannabis was clearly not going to be one of them, so my belief changed, and I sold.</li><li><strong>The lesson:</strong> hyped stocks and sectors are <strong>almost always a bad idea</strong>. By the time everyone talks about them, the price already includes the news, and it’s only a matter of time before it drops. That’s why Rule 1 starts from silence. The second lesson is regulatory risk: when a sector depends on government decisions, no amount of research can fully price it in.</li></ul><p>I used losses to offset taxes on my gains whenever legally possible. The order matters: I sold because my belief changed, not to save taxes. The tax saving was a consequence.</p><h3>Why I rarely sell, and why I don’t panic</h3><p>I rarely sell because my goal is 10 to 15 years away. A 30% drawdown is painful to watch, but it doesn’t change whether I reach retirement or pay for my kids’ future. It only changes the price of what I buy next month.</p><p>This is the real answer to “how do you never panic?” It’s not a personality trait. It’s the result of three things decided in advance:</p><ol><li><strong>A goal far enough away</strong> that a crash is a discount, not a threat.</li><li><strong>Rules for buying in a crash</strong> (Rule 2 and Rule 3), so I have something to do instead of something to fear.</li><li><strong>Knowing what would make me sell</strong> (Rule 4), so holding is a decision, not denial.</li></ol><p>The hardest test was COVID. Every stock I owned dropped, and it was really painful to watch. I didn’t sell. Instead, my relocation to Singapore gave me fresh liquidity, and I put it into the market while prices were still relatively low. That period is my answer to anyone who thinks this plan only works in a bull market: the rules held when the market didn’t. And COVID was short. The 2022 bear market was the longer test: most of a year of falling prices, with no quick rebound in sight. I held through that too, and kept investing every month.</p><p>A long-term portfolio also works outside the market. It strengthens your position when you negotiate with a bank, because it shows years of discipline, not only a salary.</p><h3>The tools I built</h3><p>I’m an engineer, so I keep trying to turn these rules into code. Most of it is still a prototype, and one project failed, but building it forces me to make my judgment explicit.</p><ul><li><a href="https://github.com/marcosuma/mok-algo-trader"><strong>mok-algo-trader</strong></a><strong>:</strong> Builds forex strategies with technical analysis or machine learning, backtests them, then trades them in real time with paper money. Status: public, paused (unsuccessful so far).</li><li><strong>Weighted DCA backtest:</strong> Compares lump sum, plain DCA and crash-weighted DCA on real data. Status: private, prototyping.</li><li><strong>Quiet-sector screener:</strong> Scores sectors on media attention vs fundamentals. Status: private, prototyping.</li><li><strong>Options tracker:</strong> Tracks premiums collected, assignment risk and cash reserved per open put. Status: private, prototyping.</li></ul><h3>Takeaways</h3><ol><li>Start from one long-term goal. It is the source of your calm.</li><li>Turn your instincts into written rules: when to buy, when to buy more, when to sell.</li><li>Treat failures as data, and learn the tax rules of your country.</li><li>If you can code, try to turn your rules into tools. Even when the tool fails, you learn where your rules are vague.</li></ol><h3>Let’s connect</h3><p>If this was useful, add me on <a href="https://www.linkedin.com/in/marcosuma">LinkedIn</a>. If you’re interested in talking more, book a session with me on <a href="https://mentorcruise.com/mentor/marcosuma/">MentorCruise</a>.</p><p>We can talk about career growth in tech, or how I think about building a long-term plan. Just note that it’s not personal investment advice.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=ab9ec44bdf0f" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[AI: The Next Abstraction Layer]]></title>
            <link>https://marcsuma.medium.com/ai-the-next-abstraction-layer-64906ca8befe?source=rss-ea35b10f9152------2</link>
            <guid isPermaLink="false">https://medium.com/p/64906ca8befe</guid>
            <category><![CDATA[technology]]></category>
            <category><![CDATA[future-of-work]]></category>
            <category><![CDATA[software-engineering]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <dc:creator><![CDATA[Marco Suma]]></dc:creator>
            <pubDate>Sun, 04 Oct 2026 08:11:14 GMT</pubDate>
            <atom:updated>2026-10-04T08:11:14.568Z</atom:updated>
            <content:encoded><![CDATA[<h4>How AI is changing the way we work, build, and interact with computers</h4><figure><img alt="Illustration of computing’s abstraction layers, from machine code up to operating systems, with a new AI layer on top marked probabilistic and leaking into the layers below" src="https://cdn-images-1.medium.com/max/1024/1*91YWy0ycryVVY0cCqsd14A.jpeg" /></figure><p><strong>tl;dr:</strong> Today’s “AI revolution” bundles several technologies, but the deeper change is a new abstraction layer: we describe what we want and machines handle more of the how. Unlike every layer before it, this one is probabilistic, so it leaks. That makes verification and accountability the scarce skills, and it threatens the entry-level work where people learn them. The technology can be transformative while the economics built around it still contain a bubble.</p><p>Every layer of computing we have built so far kept one promise: give it the same input and you get the same output. That promise is why we trust compilers and operating systems enough to stop thinking about what sits underneath them.</p><p>AI is the first layer that breaks it. It can write excellent code and miss an obvious edge case. It can summarize hundreds of pages and confidently misstate the one detail that matters.</p><p>And yet it may be the most important new layer since the programming language itself. That tension is what this article is about: a layer that lets humans move higher, built on a foundation we cannot fully trust.</p><h3>AI is the next abstraction layer</h3><p>AI is not new. Alan Turing asked whether machines can think in 1950, and the term “artificial intelligence” was coined for the Dartmouth workshop in 1956. What is new is the scale, the accessibility and the noise.</p><p>Today almost anything involving machine intelligence gets called AI: large language models, coding agents, image generators, robotaxis, humanoid robots, recommendation systems. These are related, but they did not arrive on one timeline. Industrial robots have worked on assembly lines since the 1960s, and self-driving cars trace back to the Defense Advanced Research Projects Agency (DARPA) Grand Challenges of 2004 and 2005.</p><p>So the useful question is not “what will AI do?” It is: <strong>what is changing underneath the noise?</strong></p><p>My answer is <strong>abstraction</strong>. Computing history is the story of humans moving further away from the machinery:</p><ul><li>Machine code became assembly, and assembly became high-level languages</li><li>Languages became frameworks, libraries and platforms</li><li>Operating systems hid the hardware, and graphical interfaces hid the command line</li><li>The smartphone hid the computer behind icons and gestures</li></ul><p>At every step the human moved one level up. A Python developer doesn’t think about CPU registers. Someone sending a message doesn’t think about packets.</p><p>AI adds another layer. Programming has always meant translating an idea into instructions a machine can execute. Now we describe the outcome and let the system handle more of the translation. Andrej Karpathy joked in 2023 that the hottest new programming language is English. I would push it further: <strong>the next level of programming may be writing design documents.</strong> You describe what the system should do, its constraints and the outcome you want, and the machine proposes the implementation.</p><p>The same applies beyond code. Today we open email to deal with email, a calendar for meetings, a spreadsheet for data. An AI-native computer inverts that relationship. You don’t ask “which application should I open?” You ask “prepare everything I need for tomorrow’s customer meeting”, and the system works out which data, tools and actions it needs.</p><p>This is no longer speculative. Microsoft began describing Windows as an <a href="https://venturebeat.com/ai/microsoft-remakes-windows-for-an-era-of-autonomous-ai-agents">“agentic OS”</a> in late 2025, and at Build 2026 it <a href="https://dev.to/maxmendes91/microsoft-build-2026-windows-wants-to-be-the-operating-system-for-ai-agents-59kl">repositioned Windows as a runtime for AI agents</a>. The first announcement also drew <a href="https://www.techradar.com/pro/security/no-one-wants-this-microsoft-exec-hit-with-barrage-of-abuse-after-celebrating-windows-evolution-to-become-an-agentic-os">loud user backlash</a>, a reminder that a new layer has to be wanted, not only built.</p><p>The abstraction is no longer hiding the code. <strong>It is hiding the software.</strong></p><h3>Unlike every layer before it, this one leaks</h3><p>Here the analogy breaks, and this is the most important part of the argument.</p><p>Every previous layer was <strong>deterministic</strong>. A compiler turns the same code into the same machine instructions every time. That reliability is what let programmers stop looking underneath: you trust the layer, so you stop thinking about it.</p><p>AI is <strong>probabilistic</strong>. The same request can produce different results, and a model that solves a hard problem can fail an easy one. A calculator never decides that 17 × 24 is a good opportunity to improvise. A language model sometimes does.</p><p>Joel Spolsky wrote in 2002 that <a href="https://www.joelonsoftware.com/2002/11/11/the-law-of-leaky-abstractions/">all non-trivial abstractions leak</a>. This one leaks by design, and the evidence is consistent:</p><ul><li><strong>The frontier is jagged.</strong> In a <a href="https://pubsonline.informs.org/doi/full/10.1287/orsc.2025.21838">field experiment with 758 Boston Consulting Group consultants</a>, GPT-4 users completed 12.2% more tasks, about 25% faster and at more than 30% higher quality, on tasks inside the model’s capabilities. On a task just outside them, AI users were 19 percentage points less likely to get the right answer, even though their answers read as more persuasive.</li><li><strong>We misjudge it.</strong> In a <a href="https://metr.org/blog/2025-07-10-early-2025-ai-experienced-os-dev-study/">2025 randomized trial by METR</a> (Model Evaluation and Threat Research), experienced open-source developers expected AI to make them 24% faster. They were measured 19% slower, and afterwards still believed they had been 20% faster. METR’s <a href="https://metr.org/blog/2026-02-24-uplift-update/">February 2026 update</a> suggests newer tools now give a speedup, but its confidence intervals include zero. The durable finding is the perception gap: we are bad at judging whether the layer below us is working.</li></ul><p>Edsger Dijkstra argued in the late 1970s that natural language is a poor way to program, because its ambiguity is exactly what formal languages exist to remove. He was right about the ambiguity. What is changing is where the precision lives.</p><p>When the implementation is generated, the formal layer moves to <strong>verification</strong>: tests, acceptance criteria, evaluations and monitoring. In a world of design documents, a specification you can check is worth more than code you can write.</p><p>You can only move up a layer if you can trust the one below it. With AI, that trust is earned task by task, not granted once.</p><h3>So the bottleneck moves from creating to judging</h3><p>If trust is earned task by task, someone has to do the checking. That changes where the real work sits.</p><p>An AI coding agent can build an impressive feature in an afternoon. That is extraordinary, and it is not the same as shipping it.</p><p>Production software has to fit existing systems. It needs security, tests, observability and backwards compatibility. Someone has to own it six months later, support it when it breaks and explain it to a customer. Someone has to decide whether the requirement was right in the first place.</p><p>The same holds outside software. AI can draft a business plan; that doesn’t make the business viable. It can draft a contract; that doesn’t make the contract correct. This leads to the idea I think is central to the whole transition:</p><blockquote><strong><em>AI dramatically reduces the cost of generating possibilities. It does not reduce the cost of deciding which possibilities deserve to exist.</em></strong></blockquote><p>It can even increase it. The curl project, one of the most widely used pieces of open-source software, <a href="https://www.bleepingcomputer.com/news/security/curl-ending-bug-bounty-program-after-flood-of-ai-slop-reports/">ended its bug bounty in January 2026</a> after a flood of low-quality, AI-generated vulnerability reports. Each report took minutes to generate and far longer for the project’s small security team to disprove.</p><p>When producing five hundred ideas takes minutes, generation stops being the bottleneck. <strong>Judgment</strong> becomes the bottleneck.</p><h3>Work moves up, one task at a time</h3><p>That shift is already reshaping work, though not the way the headlines frame it.</p><p>The common question is “which jobs will AI replace?” A better one: what happens when AI moves one layer into a workflow?</p><p>Economists have used a task-based lens for two decades, since David Autor, Frank Levy and Richard Murnane showed that computers replace specific tasks rather than whole occupations. It fits AI well. <strong>The unit of change is the task, not the job.</strong> Three kinds of work show how.</p><h4>The software engineer</h4><p>Software is almost entirely digital, so it moves first. The chain runs from requirements to design, code, tests, deployment and operations, and AI now participates in every link: writing code, generating tests, reading logs, explaining unfamiliar systems and investigating bugs.</p><p>The engineer doesn’t disappear, but their time shifts toward requirements, architecture, trade-offs and validation. The engineer becomes less a translator between specification and code and more a designer and reviewer of systems.</p><h4>The electrician, the driver, the factory operator</h4><p>A language model in a data center cannot wire a building. The physical world has weight, friction, unpredictable environments and imperfect sensors. Moravec’s paradox still applies: what is easy for humans is often hard for machines.</p><p>But physical automation is no longer a separate story. Robotics and autonomous driving grew independently of chatbots, yet they increasingly run on the same kind of foundation models. Vision-language-action models now combine perception, language and motor control in a single system.</p><p>The results are leaving the lab. Waymo <a href="https://techcrunch.com/2026/03/27/waymo-skyrocketing-ridership-in-one-chart/">crossed 500,000 paid driverless rides a week in March 2026</a>, ten times its volume two years earlier. Two Figure humanoid robots <a href="https://kraneshares.com/humanoid-robotics-in-2026-the-race-from-pilot-to-platform/">ran 10-hour shifts at BMW’s Spartanburg plant for 11 months</a>. The trajectories started apart and are converging, partly on the same technology. The human role shifts toward supervising machines, handling exceptions and troubleshooting.</p><h4>The office worker</h4><p>Much administrative work is information transformation. Read a document, extract the data, check it against a rule, move it to another system, write a standard reply, repeat. Generative AI and traditional automation are both good at this.</p><p>So here is the uncomfortable question: <strong>if AI performs 80% of someone’s workflow, what is the remaining 20%?</strong> It can be the valuable part: exceptions, judgment, human contact, accountability and process improvement. But the job title can stay the same while the work underneath changes completely.</p><h3>Moving up needs a ladder, and the bottom rung is thinning</h3><p>In all three examples, the human moves up a layer. That assumes there is a ladder to climb. The data says the bottom of the ladder is already thinning.</p><p>Stanford’s Digital Economy Lab tracks payroll records from ADP, the largest US payroll provider. In its <a href="https://digitaleconomy.stanford.edu/news/canariesaug26/">August 2026 update</a>, employment for 22-to-25-year-olds in the most AI-exposed occupations, such as software development and customer service, is 19% below where it would be had it kept pace with peers in less-exposed jobs. A year earlier the gap was 15%.</p><p>Experienced workers show no comparable gap, and the authors see no widespread, economy-wide displacement. The adjustment runs mostly through fewer junior hires, not layoffs.</p><p>This exposes a gap in the abstraction argument. Seniors earn their judgment by doing the lower-layer work for years: writing the code, debugging the outage, reading the contract line by line. If AI does that work and companies stop hiring juniors to do it, where do the next seniors come from? Previous abstractions let people skip a layer. This one may remove the layer people learn on.</p><p>It also forces a sharper answer about what stays human. “Judgment” is a weak answer, because models keep getting better at evaluation: AI already reviews code and grades other AI. A better answer is older. A training slide <a href="https://simonwillison.net/2025/Feb/3/a-computer-can-never-be-held-accountable/">attributed to IBM in 1979</a> reads: “A computer can never be held accountable, therefore a computer must never make a management decision.”</p><p>Judgment can be partly automated. <strong>Accountability</strong>, meaning someone who owns the decision and its consequences, cannot. That is the layer humans keep, and organizations need people who are prepared to hold it.</p><h3>Not everyone climbs at the same speed</h3><p>Even with a ladder in place, people, companies and countries won’t climb it at the same speed. Technology runs on three clocks: <strong>individual, organization, society.</strong></p><p>An individual can find a tool tonight and work differently tomorrow. A company has legacy systems, security policies, procurement, legal requirements, incentives and thousands of people. A society has to change education, regulation, labor markets and infrastructure. The gap between the clocks is large, and it shows in the data:</p><ul><li><strong>Adoption is near-universal.</strong> <a href="https://www.mckinsey.com/capabilities/operations/our-insights/the-state-of-ai">McKinsey’s 2025 global survey</a> found 88% of organizations use AI in at least one business function.</li><li><strong>Transformation is rare.</strong> In the same survey only 39% report any impact on EBIT (earnings before interest and taxes), and about 6% attribute more than 5% of EBIT to AI. Those high performers are far more likely to have redesigned their workflows. MIT’s Project NANDA <a href="https://virtualizationreview.com/articles/2025/08/19/mit-report-finds-most-ai-business-investments-fail-reveals-genai-divide.aspx">reported a starker number</a>: 95% of the enterprise pilots it studied showed no measurable profit-and-loss impact. Treat that figure as directional, since it came from a preliminary report on a limited sample.</li><li><strong>Countries are diverging.</strong> Microsoft’s <a href="https://blogs.microsoft.com/on-the-issues/2026/09/21/the-continued-state-of-global-ai-diffusion-in-2026/">Global AI Diffusion Report</a> puts generative AI use at 28.8% of the working-age population in the Global North and 16.2% in the Global South in Q2 2026, with the gap still widening. <a href="https://www.anthropic.com/research/anthropic-economic-index-january-2026-report">Anthropic’s Economic Index</a> finds usage strongly correlated with GDP per capita.</li></ul><p>Imagine two companies with the same frontier models. Company A gives every employee an AI assistant. Company B redesigns how it builds products, runs support, writes software and makes decisions. They have access to the same technology. They are not using the same technology.</p><p>That is the paradox. AI may be one of the most democratizing technologies ever built, because powerful intellectual capability is getting cheap. It may also widen gaps, because the benefits compound for those who reorganize around it. <strong>The scarce resource is no longer intelligence. It is the ability to adapt.</strong></p><h3>The layer gets cheaper while the bets get bigger</h3><p>Adaptation decides who benefits from AI. It doesn’t decide who gets paid for building it.</p><p>The technology can be real while the economics around it are wrong. The internet was transformative; many internet companies were not. The dot-com crash didn’t disprove the internet. It exposed the gap between a technology’s long-term importance and the short-term expectations priced into it.</p><p>The abstraction lens adds something the usual dot-com comparison misses. <strong>Abstraction layers tend to become commodities.</strong> Compilers were once expensive products and are now free. TCP/IP made fortunes for the companies built on top of it, not for the protocol. The exceptions are layers with strong lock-in, like Windows in the 1990s or today’s cloud platforms.</p><p>AI is moving down the commodity curve fast. <a href="https://epoch.ai/publications/the-plunging-price-of-thought">Epoch AI estimates</a> that the price of reaching a fixed level of AI performance has fallen about 13 times per year since 2023. Meanwhile the largest cloud companies plan to spend <a href="https://finance.yahoo.com/sectors/technology/articles/hyperscalers-hit-700-billion-2026-111243744.html">more than $700 billion on capital expenditure in 2026</a>, most of it on AI infrastructure.</p><p>So the trillion-dollar question is not whether AI is useful. It is whether the model layer becomes <strong>the next Windows or the next compiler</strong>: a toll booth, or a free utility that others build fortunes on.</p><p>Some analysts already think the market is late in the cycle. In September 2026 Capital Economics <a href="https://www.cnbc.com/2026/09/10/wall-street-firm-believes-the-ai-stock-market-boom-is-nearing-an-end-heres-why.html">forecast the S&amp;P 500 would rally into year-end and then fall back in 2027</a>. A <a href="https://www.ecb.europa.eu/press/blog/date/2026/html/ecb.blog20260817~754a8a4418.en.html">European Central Bank blog post in August</a> argued a correction is likely even if today’s valuations are rational. The counterargument is real too: unlike in 2000, the biggest spenders are profitable incumbents, although they <a href="https://www.forbes.com/sites/jasonkirsch/2026/06/02/the-ai-capex-to-revenue-gap-is-widening---and-markets-are-starting-to-notice/">increasingly fund the build-out with debt</a>.</p><p>Either way, the end state is familiar. John McCarthy is often credited with the complaint that once something works, nobody calls it AI anymore. If AI succeeds it becomes software, infrastructure, a feature. Successful technology disappears into the background, and that is exactly when it stops commanding a premium.</p><h3>The proof: I wrote this article with AI</h3><p>This article was a small test of its own argument.</p><p>I didn’t ask a model what to think. The ideas started as conversations: observations, questions and half-formed arguments. Some sounded good and turned out weak; others only became clear once I tried to explain them.</p><p>What I stopped caring about was the mechanics of writing: finding the right words, rearranging paragraphs, rewriting the introduction because the conclusion moved. I spent my time on the thoughts and let the model handle the expression. I moved one layer up.</p><p>The layer also leaked, exactly as described above. The model fact-checked me and not every claim survived: my first draft called the AI-native operating system a future possibility, when Microsoft had been building it into Windows since 2025. And every fact it brought back still needed checking. The thinking stayed mine, the wording moved down a layer, and the verification ran both ways.</p><h3>One layer higher, if we keep the ladder</h3><p>Put together, the argument runs like this. AI is the next abstraction layer: we describe intent, and machines handle more of the implementation. Unlike every layer before it, this one leaks, so the bottleneck moves from creating to judging.</p><p>Work moves up one task at a time, but moving up needs a ladder, and its bottom rung is thinning. Not everyone climbs at the same speed. Meanwhile the layer gets cheaper while the bets on it get bigger.</p><p>If that holds, the skills that matter shift with it:</p><ul><li>When creation gets cheaper, <strong>verification</strong> becomes the craft</li><li>When implementation gets easier, <strong>deciding what to build</strong> matters more</li><li>When machines can judge, <strong>accountability</strong> stays human</li><li>When juniors stop doing the lower layer, <strong>training them</strong> has to become a deliberate investment instead of a side effect</li></ul><h4>What I don’t know yet</h4><p>I would rather end with the questions I can’t answer than pretend the argument is settled:</p><ul><li><strong>Will the leak close?</strong> If models become reliable enough to use without checking, AI becomes an ordinary abstraction and most of the caveats here expire. If the frontier stays jagged, verification stays the core skill.</li><li><strong>Where will the next seniors come from?</strong> If companies keep hiring fewer juniors, someone still has to build the judgment that accountability depends on, and someone has to pay for it.</li><li><strong>Can accountability scale?</strong> When agents make thousands of decisions a day, “a human is accountable” risks becoming a signature on decisions nobody read.</li><li><strong>Will the divide close or compound?</strong> Cheap and open-weight models could let late adopters leapfrog, or the gap between Company A and Company B could widen every year.</li><li><strong>Is the model layer the next Windows or the next compiler?</strong> The answer decides whether more than $700 billion of capital spending this year pays off.</li></ul><p>We spent decades making computers better at producing answers. The harder problem now is deciding which questions are worth asking, and making sure there are still people who can check the answers.</p><p>Not machines replacing humans. Not humans competing with machines. <strong>Humans moving one abstraction layer higher, as long as we keep the ladder.</strong></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=64906ca8befe" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Start with tenets: revisiting the why-first approach when making a decision]]></title>
            <link>https://marcsuma.medium.com/start-with-tenets-revisiting-the-why-first-approach-when-making-a-decision-eee11ca30459?source=rss-ea35b10f9152------2</link>
            <guid isPermaLink="false">https://medium.com/p/eee11ca30459</guid>
            <category><![CDATA[software-engineering]]></category>
            <category><![CDATA[engineering-management]]></category>
            <category><![CDATA[decision-making]]></category>
            <category><![CDATA[technical-writing]]></category>
            <category><![CDATA[leadership]]></category>
            <dc:creator><![CDATA[Marco Suma]]></dc:creator>
            <pubDate>Tue, 29 Sep 2026 17:07:17 GMT</pubDate>
            <atom:updated>2026-09-29T17:07:17.586Z</atom:updated>
            <content:encoded><![CDATA[<h4><em>Why engineering debates stall, and the section I now put at the top of every design doc.</em></h4><figure><img alt="Illustration of a design doc where a Tenets section, in priority order, sits between the problem statement and the options" src="https://cdn-images-1.medium.com/max/1024/1*rxw5ejEZFDRcq3cvmZIohQ.jpeg" /></figure><p><em>Note: these are lessons from my experience as an engineer and manager at AWS, Google, Meta and MongoDB. Take what works for you.</em></p><p><strong>TL;DR:</strong> Most engineering debates don’t stall because people disagree on the solution. They stall because nobody agreed on what matters most. Starting with why tells everyone the problem is worth solving. Tenets, a short and ordered list of principles placed right after the problem statement, tell them which trade-offs to take. Agree on the tenets first and the design debate gets shorter. Disagree on the tenets and you have a small, high-level question that leadership can settle quickly.</p><h3>The debate that never ends</h3><p>Many times in my career I had to make a decision: which design to choose, which process works better, which option to ship first. And many times a healthy discussion turned into a very long one. The review would loop on “why not X?” and “have you considered Y?”, and an hour later we were back where we started.</p><p>It took me a while to see the pattern. People weren’t disagreeing on the options. They were disagreeing on the principles behind them, without ever saying so. One person was optimising for speed, another for safety, a third for reuse. Every option looked wrong to someone, because each of them was scoring it against a different, unwritten rubric.</p><h3>Start with why is necessary, but not sufficient</h3><p>Simon Sinek popularised the idea in <a href="https://simonsinek.com/books/start-with-why">Start with Why</a> (2009). His Golden Circle puts the why (the purpose) at the center, then the how, then the what. People commit to something when they understand why it matters. The tech industry has absorbed this well: good design docs open with a problem statement, and good product reviews start from the customer.</p><p>In engineering decisions, though, agreeing on the why is only the first step. Once we all agree the problem is worth solving, we usually face several reasonable ways to solve it, and several ways to build each one. The why tells you where to go. It doesn’t tell you which trade-off to take when two good options pull in different directions.</p><p>That’s the gap tenets fill.</p><h3>What a tenet is (and what it isn’t)</h3><p>I first came across tenets at Amazon, where they are part of the writing culture. A tenet is a short principle that states what the team believes and, more importantly, what it prioritises. Phil Le-Brun, Executive in Residence at AWS, describes a tenet as “an organisational belief that accelerates decision-making” (<a href="https://aws.amazon.com/blogs/enterprise-strategy/tenets-supercharging-decision-making/">AWS, 2023</a>): the tie-breaker when two good options conflict.</p><p>Good tenets share a few properties:</p><ul><li><strong>They are ordered.</strong> When two tenets conflict, the higher one wins. That’s the whole point. Le-Brun suggests no more than seven, in descending order of priority. For a single design doc I rarely need more than three to five.</li><li><strong>They take a side.</strong> A tenet needs a reasonable opposite. “We value quality” helps nobody, because nobody argues for the opposite. “We prefer shipping a narrower feature on time over a complete feature late” is a tenet: someone could sensibly argue the reverse.</li><li><strong>They are specific to the decision.</strong> Company values are too generic to settle a design debate. Charlie Kindel, who wrote extensively about tenets from his time at Amazon, calls this being “endeavor-specific” (<a href="https://blog.kindel.com/2020/02/10/tenets/">Tenets, 2020</a>).</li><li><strong>They are open to challenge.</strong> Amazon tenets traditionally end with “unless you know better ones”. As Vishal Bardoloi puts it, that phrase is an invitation to think, speak up and improve them (<a href="https://bardoloi.com/blog/2018/12/05/unless-you-know-better-ones/">2018</a>).</li></ul><p>In a document, tenets sit right after the introduction and the problem statement (the why), before any option is discussed. Readers learn how the author is reasoning before they see what the author proposes.</p><p>If this sounds familiar, it’s because other companies reached a similar idea by different routes. Google design docs have a <a href="https://www.industrialempathy.com/posts/design-docs-at-google/">goals and non-goals</a> section, where non-goals are things that could reasonably be goals but are deliberately excluded. The <a href="https://ozimmer.ch/practices/2022/11/22/MADRTemplatePrimer.html">MADR template</a> for architecture decision records has a “decision drivers” section. Tenets add one thing both often lack: an explicit order.</p><h3>Revisiting the example: two teams, both right</h3><p>Picture two teams. Team A owns the data layer and cares about durability: no data loss, careful migrations, backward compatibility. Team B owns the product and cares about user experience and delivery dates. A new feature needs a change to the data model, and a customer launch is six weeks away.</p><p>Team A proposes a new versioned schema with a background migration: safe, but eight weeks of work. Team B proposes adding the new fields to the existing model and cleaning up later: four weeks, with some debt. The review goes in circles, because both teams are right. Each one is optimising for what it is responsible for.</p><p>Now put three tenets at the top of the doc, before either proposal:</p><ol><li>Existing customers never lose data or need to change their integration.</li><li>We ship to our launch customers by the launch date, even if the first version covers only their use case.</li><li>We minimise work we know we will have to redo within a year.</li></ol><p>The conversation changes. Team B’s proposal passes tenets 1 and 2 but fails 3. Team A’s proposal passes 1 and 3 but fails 2. Because the tenets are ordered, the teams now look for an option that satisfies 1 and 2 and does as well as possible on 3: for example, the new fields exposed behind a versioned API, with the full migration scheduled right after the launch. Nobody had to win the argument. The tenets did the prioritising.</p><p>And if Team A believes tenet 3 should come before tenet 2, that’s a much better argument to have. It’s one sentence, not two designs.</p><h3>From labels to tenets</h3><p>My first tenet lists looked like this:</p><ul><li>Time to market</li><li>Prioritise use cases A, B and C</li><li>Reusability</li><li>No breaking changes</li><li>No migrations required</li></ul><p>These are labels, not tenets. Everyone agrees with them, so they settle nothing. Rewritten with an opinion and an order, they become useful:</p><ol><li><strong>No breaking changes.</strong> Existing users keep working without touching their code, configuration or data. No migrations required.</li><li><strong>Use cases A and B first.</strong> C waits for the next iteration if supporting it slows us down.</li><li><strong>Time to market over completeness.</strong> We ship to design partners this quarter, even with a narrower scope.</li><li><strong>Reuse when it’s cheap.</strong> We build for reuse only when it doesn’t delay tenets 1 to 3.</li></ol><p>The difference: each line now tells you what to give up.</p><h3>A real example: auto-generating the MongoDB Atlas Terraform provider</h3><p>At MongoDB, my team builds the infrastructure-as-code experience for MongoDB Atlas. This year we <a href="https://www.mongodb.com/company/blog/technical/how-auto-generating-terraform-improved-our-apis-customer-experience">wrote about</a> how we auto-generate Terraform resources from the Atlas API specification. Before that work, a new Atlas API capability took four to ten weeks to reach the Terraform provider, and a typical resource took one to one and a half weeks of development. With the generator, it takes one to two days.</p><p>We didn’t call them tenets in that post, but reading it back, the decisions follow a few clear principles:</p><ul><li><strong>Customer experience first.</strong> Generated resources must be indistinguishable from hand-written ones: same schema shape, same plan output, same errors.</li><li><strong>Nobody gets stranded.</strong> Each generated resource is pinned to an Atlas API version, so customers who can’t upgrade immediately stay unblocked.</li><li><strong>Fix upstream.</strong> When the generator hits an ambiguous API, we fix the API and its standards once, instead of patching around it in every tool.</li><li><strong>Adopt incrementally.</strong> Generated resources live alongside hand-written ones, so we could ship without a big-bang migration.</li></ul><p>The hard trade-offs in that project, from where custom code lives to how versioning works, kept coming back to these principles. Once they were clear, most design discussions became a quick check against them.</p><h3>When the tenets themselves become the debate</h3><p>There’s a catch: tenets are debatable too. Why should “prioritise AI agent clients” rank above “reusability” in this document? Someone will ask, and they should.</p><p>That isn’t a flaw. It’s the main benefit. A debate about tenets is small, explicit and high level. Unlike a debate about implementation details, it’s one you can escalate.</p><p>Leadership usually can’t help you choose between two schema designs: that conversation needs a different audience and a lot of context. But leaders are very good at deciding whether time to market matters more than reuse this quarter, because that’s a business priority, and business priorities are their job. So when the tenets are contested, I take them to leadership first, get a decision, and then go back to the design with the right people in the room.</p><p>Two habits keep this healthy:</p><ul><li><strong>Review the tenets before the solution.</strong> Share them early, or spend the first minutes of the design review agreeing on them. If they change, the rest of the discussion changes with them.</li><li><strong>Once agreed, commit.</strong> Amazon calls this “disagree and commit”: you can argue for a different order, but once the order is set, you build to it. “Unless you know better ones” keeps the door open for new data, not for relitigating every review.</li></ul><h3>How I write tenets today</h3><ul><li>Three to five per document, ordered by priority.</li><li>One idea per tenet, written as a sentence, not a label.</li><li>Every tenet has a reasonable opposite. If nobody could disagree with it, delete it.</li><li>Specific to this decision, not a copy of the company values.</li><li>Placed right after the problem statement and reviewed before any option is discussed.</li><li>Revisited when the context changes, not when the debate gets uncomfortable.</li></ul><p>Next time a design review loops on “why not X?”, stop and ask a different question: which tenet does X serve better? If nobody can answer, you don’t need a better design yet. You need tenets.</p><p><em>I mentor engineers preparing for Big Tech interviews and tech-lead roles. If you’d like help, you can find me on </em><a href="https://mentorcruise.com/mentor/marcosuma/"><em>MentorCruise</em></a><em>.</em></p><h4>Further reading</h4><ul><li>Phil Le-Brun, <a href="https://aws.amazon.com/blogs/enterprise-strategy/tenets-supercharging-decision-making/">Tenets: supercharging decision-making</a> (AWS Executive in Residence blog, 2023)</li><li>Charlie Kindel, <a href="https://blog.kindel.com/2020/02/10/tenets/">Tenets</a> (2020)</li><li>Vishal Bardoloi, <a href="https://bardoloi.com/blog/2018/12/05/unless-you-know-better-ones/">(Unless you know better ones)</a> (2018)</li><li>Malte Ubl, <a href="https://www.industrialempathy.com/posts/design-docs-at-google/">Design Docs at Google</a> (2020)</li><li>Olaf Zimmermann, <a href="https://ozimmer.ch/practices/2022/11/22/MADRTemplatePrimer.html">The Markdown ADR (MADR) Template Explained and Distilled</a></li><li>Agustin Bettati and Marco Suma, <a href="https://www.mongodb.com/company/blog/technical/how-auto-generating-terraform-improved-our-apis-customer-experience">How Auto-Generating Terraform Improved Our APIs and Customer Experience</a> (MongoDB, 2026)</li><li>Simon Sinek, <a href="https://simonsinek.com/books/start-with-why">Start with Why</a> (Portfolio, 2009)</li></ul><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=eee11ca30459" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Working in engineering teams: tips for healthy habits]]></title>
            <link>https://marcsuma.medium.com/working-in-engineering-teams-tips-for-healthy-habits-61a0bb82428e?source=rss-ea35b10f9152------2</link>
            <guid isPermaLink="false">https://medium.com/p/61a0bb82428e</guid>
            <category><![CDATA[team]]></category>
            <category><![CDATA[habits]]></category>
            <category><![CDATA[teamwork]]></category>
            <category><![CDATA[software-engineering]]></category>
            <category><![CDATA[engineering]]></category>
            <dc:creator><![CDATA[Marco Suma]]></dc:creator>
            <pubDate>Fri, 01 Dec 2023 16:21:49 GMT</pubDate>
            <atom:updated>2023-12-01T16:21:49.804Z</atom:updated>
            <content:encoded><![CDATA[<p><em>Note #1: these are</em><strong><em> </em></strong><em>tips based on my </em><strong><em>experience</em></strong><em>. You are free to use them if you find them useful.</em></p><p><em>Note #2: they </em><strong><em>are not in order</em></strong><em>. They are just being written as they come.</em></p><p><em>Note #3: they </em><strong><em>are not complete. </em></strong><em>I intend to add new ones when they pop-up in my mind.</em></p><h3>#1</h3><p>A <strong>great team</strong> is based on a <strong>respectful,</strong> <strong>inclusive and diversified</strong> <strong>environment. </strong>In other words, <strong>psychologically safe</strong>.</p><h3>#2</h3><p><strong>Empower</strong> each other in making decisions. Build things based on <strong>trust</strong>. Have <strong>fun</strong>.</p><h3>#3</h3><p>Use <strong>effective communication. </strong>Be <strong>as accurate as possible with your language </strong>— <strong>precision</strong> is the key especially when not being physically together <em>(note: tl;dr: are highly recommended when there are long conversations and context is hard to gather)</em>.</p><ul><li><strong>Avoid</strong> usage of <strong>acronyms</strong> as much as possible (unless explicitly defined and fully agreed on)</li><li>Be <strong>concise</strong> and <strong>predictable</strong><em>. </em>Concise because you want to encourage people to interact with what you are saying and optimise their time. Predictable because creating healthy routines makes everyone more comfortable</li></ul><h3>#4</h3><p>Be <a href="https://marcsuma.medium.com/have-strategic-patience-and-be-right-a-lot-afa24d20c967"><strong>strategic patient</strong></a><strong>.</strong></p><h3>#5</h3><p>Use <strong>calendar for everything</strong>: your OOOs, your tasks and of course your meetings. Splitting your day in time-boxed events helps a lot with keeping up with different tasks of different nature.</p><h3>#6</h3><p>Use a <strong>journal </strong>where you keep track of your work (including achievements, no matter how big they are) — <em>(I know someone in my team does this already)</em></p><h3>#7</h3><p>Use a <strong>personal to-do list. </strong><em>I use Google Docs but there are plenty of tools online.</em></p><ul><li>I use my to-do list for literally everything that I can’t do right away: sending an e-mail, writing down an idea, following up with a colleague, creating a follow-up task and so on…</li><li>The goal is to minimise interruptions on what you are doing and free up your mind from things you have to remember</li></ul><h3>#8</h3><p><strong>Measure everything with data. </strong>Quantify progress, leverage telemetry and try to prioritise <em>“move-a-metric”</em> projects</p><ul><li>Use <strong>data</strong> also to <strong>measure the impact.</strong></li></ul><h3>#9</h3><p>Learn how to ask <strong>the right questions.</strong> Great engineers can solve very complex problems or spot unseen risks by just asking the right question.</p><h3>#10</h3><p><em>(Something I re-heard recently in my team)</em> implement <strong>the boy-scouting rule: </strong><em>“Always leave the code you are working on a little bit better than you found it”</em></p><h3>#11</h3><p>Learn how to<strong> disagree and commit</strong> by unblocking your peer</p><ul><li><em>Example: </em>if to achieve something you have <em>option A</em> and <em>option B</em> and they are both equally good and bad, <strong>be open</strong> and give your peer the opportunity to choose theirs. There is no race within the team on who wins more ideas.</li><li><strong>Embrace the change</strong>: <strong>be courageous </strong>of trying <strong>something new</strong>.</li><li>We should aim for an environment where <strong><em>“let’s try it out”</em></strong><em> </em>is preferred at <em>“here we don’t do it like that”.</em></li></ul><p><strong>Note: </strong>This is all true <strong>unless</strong> a decision poses a threat to our availability, security and overall team health.</p><h3>#12</h3><p><strong>Team health </strong>is the priority. <strong>For real</strong>. But don’t forget that company goals are the priority too.</p><ul><li>As a team we should aim at 20% sprints and 80% marathons. We need to have a <strong>long-term mindset, </strong>but still keep a closed-eye to <strong>low-hanging fruit wins</strong> or anything that can help us <strong>unblock </strong>and <strong>move faster</strong>.</li></ul><h3>#13</h3><p><strong>Constructive feedback is a gift for everyone</strong>.</p><ul><li><strong>The receiver:</strong> be open and believe in the good intentions. Most of the times feedback is constructive and comes from past experience in order to avoid same mistakes. I was told once that when you receive a gift you can actually decide to do whatever you want. You can even discard it if you don’t consider it valuable. But it is highly recommended to use it and eventually discuss it with your manager.</li><li><strong>The giver: </strong>too much feedback is risky, unless there is a mentoring or coaching in progress. Even <strong>managers </strong>should be very<strong> predictable</strong> when they provide their feedback. <em>Why? </em>Too much feedback can create an <em>adversarial environment</em>, where people can feel monitored and micromanaged. Always make sure your feedback is welcomed. Use the <strong>two-datapoint rule</strong>. One single datapoint to provide a feedback might be too little (we can all have a bad day). Too many datapoint of the same event might be too late to suggest a different behaviour. <em>When you think you want to provide feedback, write it down for yourself and keep it there for a bit. If a similar situation happens again, then it’s a good time to follow-up.</em></li><li><strong>Not all feedback is relevant.</strong> Something might just be a matter of different working style or cultural background. <strong>Respect, inclusivity </strong>and <strong>diversity</strong> are the <strong>keys for an healthy team. </strong><em>Ask yourself how important this topic is? Is it really relevant to mention it? Does it have a good impact? Does the behavior that deserves feedback repeated multiple times in the past? Or is it just one case?</em></li></ul><h3>#14</h3><p><strong>Personal growth</strong> generates <strong>company growth</strong>. In order for us to focus on the right things, not only we have to follow our <strong>north star </strong>and choose <strong>the right projects</strong>, but we also have to match them based on <strong>individual’s career expectations, </strong>so that engineers can grow.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=61a0bb82428e" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[I got laid off…]]></title>
            <link>https://marcsuma.medium.com/i-got-laid-off-6d11e04db2c6?source=rss-ea35b10f9152------2</link>
            <guid isPermaLink="false">https://medium.com/p/6d11e04db2c6</guid>
            <category><![CDATA[google]]></category>
            <category><![CDATA[layoffs]]></category>
            <category><![CDATA[engineering]]></category>
            <category><![CDATA[meta]]></category>
            <category><![CDATA[tech]]></category>
            <dc:creator><![CDATA[Marco Suma]]></dc:creator>
            <pubDate>Sat, 28 Jan 2023 02:42:01 GMT</pubDate>
            <atom:updated>2023-02-01T01:08:10.414Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*N26NeCQWfD9yRpjTfpNj4Q.jpeg" /></figure><p>..and I never thought it could happen to me. So I want to tell you my experience as being part of the massive layoffs happening these days in tech companies.</p><p>During my last 7 years I worked in 3 of the most prestigious tech companies for a software engineer: <strong>Amazon</strong>, <strong>Google</strong> and <strong>Meta</strong>. Every time someone would hear my story they were impressed and admired the career path. At some point even I looked at myself and built (<strong>wrongly</strong>) some sense of <em>invincibility</em> towards certain real life events such as not having a job: “I am a senior software engineer, I manage a team and have a lot of work responsibilities, my performance review is great, I never had HR issues, no teammates issues, …”.</p><p><em>10 November 2022</em>: you can imagine my reaction when at 6.00am of just a normal day like others I woke up here in Singapore ready to start my new working day and suddenly two minutes after I received the email that I was part of the lay-offs.</p><p><strong>I initially felt lost and disoriented</strong>. “And now?” I asked myself. I immediately woke up my partner as if she could do something to fix it: I just needed someone next to me at that moment. I have to say thanks to her. She quickly cheered me up and managed to give me motivation and stay positive: the fact that I always kept my mental health is surely a lot thanks to her. Because of that, I feel so grateful I had someone next to me to share this moment with and I didn’t feel alone. I know for a fact that people move to countries for work, they are on a work visa, they don’t have many familiar friends and in these situations they are in an extremely weak condition.</p><h3><strong>I wish I had…</strong></h3><p>One thing I experienced is that I started going into a spiral of negativity made of “if only I had..” or “I wish I..” — I am sure many people (if not all) have gone through this:</p><ul><li>I wish I saw this coming…</li><li>I wish I didn’t leave Google for Meta…</li><li>I wish I didn’t move to Singapore…</li><li>I wish I had moved to a different org before this happened…</li></ul><p>Quite often we wish to change our past when we’re not happy with our present. But life is all about that, it’s about choices that we make and that will have consequences in our future (positive or negative). We have to be ready to <em>marry </em>those consequences, whatever they will be. You can’t predict all the consequences, it’s impossible. You can estimate them, but you’ll never have 100% confidence.</p><p>What my partner suggested to me was: “Let’s put a timer of 60 seconds in which you are allowed to say all the negativity you want. After those 60 seconds, we move on and we plan the action items.” — I felt it made so much sense, so I recommend it to you.</p><h3><strong>It’s not you against them</strong></h3><p>It’s not you against them because in reality <em>there is no “them”</em>.</p><p>I read so many posts and comments on different social networks that were mostly sharing the same pattern: “They (the company) were so brutal in this layoff / They don’t deserve us as employees / This is a demonstration that they don’t care about you / They only care about business growth” — this tendency of <em>humanising </em>things and creating an “<em>us against them” </em>environment is very frequent in the human psychology.</p><p>Who is “they” actually? Where are these cruel people that are doing all of this?</p><p><strong>Those people are us!</strong> We are those same people that make the decision to lay off thousands of employees. We are those same people that try our best to get the promotion and get more money “whatever it takes”. We are those same people that exploit the growth wave for our own interests by knowing that there will always be a recession waiting for us.</p><p>There is no “you against them” and nobody hates you. This is part of the game that we have designed for ourselves. It’s how our society works and you can decide how to play. Funny enough: when we are proud of something our company did we tend to say “We”, when we are angry and in disagreement we tend to say “They”.</p><p>If one day you will be a VP, you will have to make those decisions. You will likely make the same decisions that are being made these days. There is a good chance that among those people who had to make a decision of laying off employees there were some who got that position thanks to the overhiring that led us into this: it sucks, I know, but it’s all related to our human nature of seeking for unstoppable growth.</p><p>To be clear: I am not happy that I was laid off, I surely didn’t deserve it and I felt betrayed. But the reality is that nobody deserves to be part of a massive lay-off.</p><h3><strong>We all risk to become like a dinosaur</strong></h3><p>People who spend years and years in the same company have higher likelihood to become dinosaurs: being a dinosaur fundamentally means that you are outdated for the current market and less and less companies will be willing to take you. A dinosaur is extremely good at the routine/tools/mechanics of their current company but eventually bad at what the market needs.</p><p>In fact, interviewing for a new company is a full-time job itself. Once I will get a new job I will share my full experience mentioning all the companies I interviewed for. It’s a serious job, it takes a lot of effort and discipline. Interviewing for senior roles is even tougher: the margin for error is extremely small. Not only do you actually have to know things but you need to be mentally fresh, confident and <em>lucky</em>.</p><p>Becoming a dinosaur it’s not something you must avoid because it’s not acceptable that being in a company for many years is a bad thing, it should be the opposite. But you need to face the reality that nobody is safe: Google recently laid-off so many people that had been working there for 10+ years.</p><h3><strong>Work hard or do the minimum required</strong></h3><p>It’s your choice, nobody will force you. I have always preferred to work hard and I will likely continue to do so as long as I love my job. But this didn’t save me from being laid off. Someone would tell me: “See? It’s not worth it, they don’t deserve this” <em>(as I was mentioning before in the previous section…)</em>.</p><p>The reality is that I’ve always worked hard because I wanted to do so, I liked it and I did it for myself so that I could learn as fast as I could.</p><p>Chances are that in most places if you do the minimum required you will survive, get your salary at the end of the month, eventually get promoted (at a slower pace, but who cares?) and you’ll still be happy. But the problem is that some people are just not fine with that, they are happy doing more. If you decide to work hard and go beyond what is expected from you, just make sure you are doing it because you like it and you are passionate about it. Not because of mine or to be liked by someone else.</p><h3><strong>How close are you to the core business?</strong></h3><p>I think that one thing that contributed to the layoff in my case is that our team in Singapore was surely not the most important in our organisation. We (60+ people) were part of an organisation in Meta that “protects users and businesses from harm” — these organisations (typically called “Business/Community Integrity” or “Trust and Safety”) are basically measured on how “clean” social networks are from violence, pornography, scam and all other things that are not compliant to the company policy.</p><p>Even though the organisation was extremely important, it became more and more evident that us in Singapore were being left behind and most of the important things were being managed between North America and EMEA. As an organisation that ultimately does not generate “direct revenue” and needs to shrink, the first impacted employees are the ones that are part of less important or redundant teams, this is obvious. For american tech companies and for engineers Singapore is also geographically in disadvantage from this point of view because it represents a minority and a third geographical area in addition to the predominance of NA and EMEA.</p><p>I am telling you this because I want you to keep this in mind in the future for yourself: choose a place that you like (first) but also keep in mind that the probability to be considered redundant in a tech company is inversely proportional to the importance of your team in the organisation/company mission.</p><p><strong>Conclusions</strong></p><p>Recently I learned the <a href="http://www.drmarlo.com/dr-marlo-speaks/maybe-so-maybe-not-well-see/"><em>parable of the Chinese farmer</em></a><em> </em>(this link is just the first one I got from a Google search — or maybe you can ask <strong>ChatGPT</strong>?) and I loved it. You must read it if you don’t know the story. I loved it because it’s fundamentally true and it manifests it in every single day in the world: a bad event in your life might not still be perceived as such in the future — it might actually turn into something to be thankful for (and same story for the opposite).</p><p>This is how I hope these huge historical lay-offs will be remembered for all of us: “bad luck, good luck, we’ll see…”.</p><p><a href="https://marcosuma.com/">Marco Suma</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=6d11e04db2c6" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[How I approach being a Senior Tech Lead]]></title>
            <link>https://marcsuma.medium.com/how-i-approach-being-a-senior-tech-lead-36e41956a86a?source=rss-ea35b10f9152------2</link>
            <guid isPermaLink="false">https://medium.com/p/36e41956a86a</guid>
            <category><![CDATA[meta]]></category>
            <category><![CDATA[leadership]]></category>
            <category><![CDATA[engineering-mangement]]></category>
            <category><![CDATA[google]]></category>
            <category><![CDATA[software-engineer]]></category>
            <dc:creator><![CDATA[Marco Suma]]></dc:creator>
            <pubDate>Sat, 08 Oct 2022 03:05:35 GMT</pubDate>
            <atom:updated>2026-09-25T13:30:21.066Z</atom:updated>
            <content:encoded><![CDATA[<p>After 3 years of being a <strong>TL in tech companies</strong>, I wanted to share <em>a few personal suggestions</em> to who is new to this role.</p><p>I chose this time because now that my team is made of 10 engineers my responsibilities and ways of working are evolving. This team size is making my job all about <strong>management, leadership and designing</strong> aspects and little or nothing related to coding. Because of that, I wanted to pause a bit and share with you <strong>9 points on how I approached being a TL</strong> so far.</p><ul><li><strong>A TL is a role of high responsibility</strong>. To incentivize your team to trust you, you should: (a) Share credits with all your team when things succeed; (b) Be the first one who takes responsibility for failures.</li><li><strong>Be bold and opinionated</strong>: it’s great to always be available to every idea but also it’s okay to say no to someone’s request. If you are 100% certain that an idea is not good, you should feel ok to reject it and explain why. In most cases though it is better if you take some time to think about it. It’s also important for you to be bold and call out when something is not going well. Being the one who brings awareness of a problem will pay back in the long term to your team’s growth.</li><li><strong>Don’t hesitate to take ownership of product strategies</strong>, even if your team already has a PM. You should not completely defer the work to them, but you should at a minimum collaborate. Your collaboration with PMs and EMs is essential. Although boundaries among these three roles need to be set, there’s no strict rules on what these are. I personally believe these boundaries should be purposely vague. These three roles should be able to help each other and wear different hats whenever needed. Dynamics in teams vary but in theory none of these three roles should strictly prevail on the other. I strongly advocate for an active presence of TLs in product strategy decisions. TLs are the people who have better chances to know how systems work and how feasible certain ideas are. Not only that, they are the ones that might know from the inside what’s the right thing to do to improve their product. Make sure you take part in updating team projects at an organizational level. You should also encourage your team members to do the same, by consistently sharing updates and what are the project strategies.</li><li><strong>If needed, be ready to make decisions on your own first and then socialize the idea</strong> and seek approval later. It might lead to stagnation and slowness if a TL seeks for full consensus or waits for other people’s feedback every time a decision has to be made. If there’s no agreement, you should be that person that at the end can decide for everyone. You need to take that responsibility.</li><li><strong>Use words of encouragement</strong> and be ready to publicly celebrate your team’s success. At the same time, you should not shy away from <strong>providing growth feedback</strong> to your colleagues on their project. Most of the time, I believe that we should recognize success in front of others while giving growth feedback privately. Other times providing growth feedback in front of the whole team can be useful for the others so that they can avoid making similar mistakes.</li><li><strong>Expect your team members will look for your help when needed</strong>. At the same time, lead by example and <strong>be the first one who seeks help</strong> from others if you feel the team needs it.</li><li><strong>Don’t stress yourself in guaranteeing your ideas are always the best and you are never wrong</strong>. Being a TL means you might have more experience than others in the team, but it doesn’t mean you have to be the technically strongest on each specific domain. You should always focus on leveraging expertise to indicate benefits and risks, bring different perspectives, but when you are not perfectly confident of which option is best, leaving a decision to someone in your team might be the best action to take. Your role as a TL is also to increase trust and confidence in the team, not showing you are the one who is always right. Try to avoid creating a dependency towards you. If your team is always seeking approval from you, you are creating weakness. Ultimately, you want to let your team grow without creating a big knowledge and experience gap between you and your team</li><li><strong>Sometimes you need to be the first person to talk, sometimes you let others talk</strong>. The balance is important because: (a) It increases the level of confidence and maturity in your team; (b) You don’t let others think that you are the only one who knows everything</li><li><strong>Wear multiple hats and understand your audience</strong>: having a meeting with a software engineer is not the same as having one with a lawyer or a marketing team. The language and interests are completely different. You need to have chameleonic skills in order to take the best out of it. Try to understand what the audience is interested in. In my opinion, you should be part of those meetings. It’s not just “the PM’s (or EM’s) responsibility” — it’s also your business because it’ll help you understand the diversity of point of views.</li></ul><p>I hope this was useful. And feel free to let me know your thoughts!</p><p><em>I mentor engineers preparing for Big Tech interviews and tech-lead roles. If you’d like help, you can find me on </em><a href="https://mentorcruise.com/mentor/marcosuma/"><em>MentorCruise</em></a><em>.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=36e41956a86a" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[About last time I didn’t set-up a dashboard and alarms for my project]]></title>
            <link>https://marcsuma.medium.com/about-last-time-i-didnt-set-up-a-dashboard-and-alarms-for-my-project-250cb60718e6?source=rss-ea35b10f9152------2</link>
            <guid isPermaLink="false">https://medium.com/p/250cb60718e6</guid>
            <category><![CDATA[monitoring]]></category>
            <category><![CDATA[software-engineering]]></category>
            <category><![CDATA[engineering]]></category>
            <category><![CDATA[software-engineer]]></category>
            <dc:creator><![CDATA[Marco Suma]]></dc:creator>
            <pubDate>Sun, 31 Jul 2022 11:27:59 GMT</pubDate>
            <atom:updated>2022-07-31T11:27:59.808Z</atom:updated>
            <content:encoded><![CDATA[<p>Last time I didn’t set-up a dashboard and alarms for my project I was working in Amazon Web Services, for a product called <a href="https://aws.amazon.com/cloudwatch/">CloudWatch</a>. CloudWatch offers a cloud-based monitoring solution to customers so that they can easily set-up a monitoring solution (either custom or off-the-shelf) in a timely manner.</p><p>Working in CloudWatch meant that you had to have monitoring in your DNA.</p><p>One night I was oncall and one of the so-called company-wide “Game Day” was planned, where one entire AZ was planned to purposely shut-down to test robustness of our services.</p><p><em>To give some more context: engineering model in AWS works a bit differently compared to other companies like Google or Meta. As a SWE (Software Engineer), you actually have to take care of the whole lifecycle of your service (including re-purposing of the hosts where your service is running — you also had to build your own CD pipeline to release software in your servers).</em></p><p>My team was new and did not have a good dashboard to measure our CPU and memory consumption at host-level. As the oncall engineer, I was asked if “I could ack and confirm our service was ready to lose one entire AZ”. I was unprepared to answer and I did not have a quick place to look at it; <strong>moved by faith and laziness</strong> I answered “yes, we are ready”.</p><p>You can imagine we were not… [<a href="https://en.wikipedia.org/wiki/Murphy%27s_law">Murphy’s Law</a>]</p><p>In the middle of the night, my pager went off. One alarm fired because of an elevated number of failures while serving customer requests.</p><p>By doing a quick RCA I noticed that the CPU of our servers were spiking up to 100% and this was obviously causing failures due to machine overload. Thankfully, we had this <em>unrelated</em> alarm that fired but the truth is that we were completely blind on capturing these type of events. In the middle of the night I had to re-purpose new servers and wait for the termination of the “Game Day”.</p><p>That lazy “yes, we are ready” answer costed me one sleep-less night, an high-impact event (somehow known as SEV) and most importantly it caused customer impact.</p><p>As an engineer that learns from mistake, that was my last time that I took those questions lightly but most importantly it was the last time I was unprepared without a proper dashboard and a set of strong alarms.</p><p>So here’s my questions to you as a software engineer: <strong>when was your last time you didn’t build a dashboard and set-up proper alarms for your service?</strong> Hopefully it’s a long time ago, but maybe it’s not. So here I am to give some high-level suggestions on setting-up metrics / dashboards and alerts to guarantee high availability of your service (even if they are internal):</p><ul><li>Make sure your service / product is publishing metrics somewhere</li><li>Be <em>organized </em>with your metrics, it’s important they follow a common and standard naming convention across your entire org (possibly)</li><li>Do an analysis and try to understand what things can go wrong with your service. After that, group the metrics that are able to help you answer if your service is healthy or not</li><li>Create a solid amount of alarms that are looking at those metrics</li><li><strong>Be prepared to learn from mistakes</strong></li></ul><p>Monitoring is like gardening: you have to constantly take care of it, prune the old alarms, add new ones in parallel with the evolution of your product, update dahsboards and so on.</p><p>Overall, monitoring does not always prevent a problem, but it definitely helps minimizing it. We can talk about testing another time, that certainly helps with the prevention.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=250cb60718e6" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Have strategic patience and “be right a lot”]]></title>
            <link>https://marcsuma.medium.com/have-strategic-patience-and-be-right-a-lot-afa24d20c967?source=rss-ea35b10f9152------2</link>
            <guid isPermaLink="false">https://medium.com/p/afa24d20c967</guid>
            <category><![CDATA[software-engineering]]></category>
            <category><![CDATA[management-and-leadership]]></category>
            <category><![CDATA[engineering]]></category>
            <category><![CDATA[management]]></category>
            <category><![CDATA[leadership]]></category>
            <dc:creator><![CDATA[Marco Suma]]></dc:creator>
            <pubDate>Sun, 31 Jul 2022 11:17:00 GMT</pubDate>
            <atom:updated>2022-07-31T11:17:00.887Z</atom:updated>
            <content:encoded><![CDATA[<p>These days, I am spending my time in Spain with my girlfriend’s family. Her father is an Admiral of the Spanish Navy. We were having a totally random conversation and he asked me: “Do you want to know something I always tell in my direct reports? I tell them they have to learn the art of Strategic Patience.”</p><p>He explained to me what he meant and that caught my attention so much because I was in the middle of preparing a note about long term impact and the two concepts were incredibly similar, just put with different words. So I decided to scratch that note and rewrite it by using this concept: strategic patience. It perfectly applies with what I meant.</p><h3>What’s strategic patience?</h3><p>Strategic patience means “to remain vigilant while waiting with patience for the expected results to come, as a result of your implemented action plan”. There are three important concept in this:</p><ul><li>Implement Action plan <strong>(strategy)</strong></li><li>Remain vigilant <strong>(observe)</strong></li><li>Wait for the results <strong>(patience)</strong></li></ul><h3>Strategy</h3><p>A <strong>strategy is a plan </strong>that you make by aggregating and organizing informed decisions you’ve made about solving a business, product, org, company or market problem.</p><p>You need to make sure strategies are a full-time commitment, and as such you don’t scratch them every time to recreate new ones — instead, make sure you update them when needed by moving fast and stay on top of the market.</p><p><strong>When you choose a strategy, you need to make sure you are right. </strong>You need to do all your best and think thoroughly before agreeing on something, and back-up your decisions based on the best available data.</p><p><strong>Continuously changing strategy risk to make an organization weak, because it ends up focusing on the short term aspects and never on the long term part</strong>.</p><h3>Observe</h3><p>You need to be a great observer. When you are a data driven company and try to make decisions based on data analysis, you are on the right path for success. Little or something is left to the gut feeling.</p><p>Observing allows you to make sure your strategy is working. It’s somehow something similar to a feedback loop: you <strong>send</strong> <strong>an input</strong> to the system, you <strong>get an output</strong>, you <strong>observe it</strong>, and finally you decide if your “system” needs to be changed and how.</p><p>To be clear: <strong>everybody can still make mistakes after observing, but if you don’t observe you will make a mistake for sure.</strong> That’s why I’ve always advocated for relentless monitoring, logging and alarming. There’s nothing you should do without these three things.</p><p>Remember: <strong>observe your results and be ready to act.</strong></p><h3>Patience</h3><p>Here’s my main solution to applying a long term plan: you must be patient. You can’t build everything in one day. You can do certain things in one day, but you certainly can’t go from 0 to 1 in one day. 0 to 1 requires patience (together with what we’ve just discussed so far).</p><p>Let me give you an example: let’s say you’re creating Twitter. You had this amazing idea and you’re implementing it. Your goal is to let the whole world, as many people as possible, use your product. How long do you think it takes to reach that goal? If for 1 year you haven’t seen much growth, do you think you should change the target so that it’s better achievable? <strong>You need to be patient</strong>. <strong>If you have done the right strategy, you are observing and you have a customer oriented mentality there’s nothing to panic</strong>.</p><p>My final motto is: <strong>be patient while observing your strategy being implemented</strong>. That’s another way to think that you need to focus on long term plans.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=afa24d20c967" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[My biggest failure was the start of the greatest journey]]></title>
            <link>https://marcsuma.medium.com/my-biggest-failure-was-the-start-of-the-greatest-journey-4c4099443357?source=rss-ea35b10f9152------2</link>
            <guid isPermaLink="false">https://medium.com/p/4c4099443357</guid>
            <category><![CDATA[failure]]></category>
            <category><![CDATA[leadership]]></category>
            <category><![CDATA[think-big]]></category>
            <category><![CDATA[software-engineering]]></category>
            <category><![CDATA[failure-to-success]]></category>
            <dc:creator><![CDATA[Marco Suma]]></dc:creator>
            <pubDate>Sat, 15 Jan 2022 15:20:44 GMT</pubDate>
            <atom:updated>2022-01-15T15:20:44.380Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/600/1*mRauU5GP3kqpQLuVH0CBoA.jpeg" /></figure><p>I perfectly remember the day of my biggest failure. I remember it because it’s the day where everything started. Let me tell you a story.</p><p>It was 2018 and I was working at Amazon. It was almost my second year in the company and I was also in my second team in the same AWS org, the Cloudwatch team in Dublin. In those days I just had an idea of implementing what is now called <a href="https://docs.aws.amazon.com/AmazonCloudWatch/latest/monitoring/CloudWatch-Service-Quota-Integration.html">AWS Usage Metrics</a>, a service officially released in 2019. I led that project until the last day I worked in AWS. After that, other brilliant engineers kept working on it.</p><p>At that time, I thought it was a quick win for the company and a great win for our customer because it was a solution offered for prevention from getting throttled when using AWS services: in other words, I wanted to offer them the possibility of having monitoring metrics to track their resources usage; with that, they could use alarms to be informed when reaching thresholds and therefore prevent being throttled and ask for a limit increase.</p><p>In fairness, that is a more mature idea of the one I had. When I proposed it to my manager, it was more like “we could offer a service for customers to show them their rate of usage of our APIs”.</p><p>My manager still liked the idea (btw I only had great managers in Amazon, and he was one of them) and so I started socialising it even more. Also one of my colleagues, who is now a great friend, more senior than me and in the team for many years, seemed excited about it also because he had that in mind for a long time but never managed to prioritise it.</p><p>Everything looked great.</p><p>I was excited by the idea that I could deliver something proposed by me supported by the excitement of most people in the team (this is the most ideal scenario for an engineer looking for promotion: you create scope for you and the team, you deliver it and create value for your customers). So we planned the project, estimated the capacity and added it in the roadmap of the following quarter.</p><p>When the new quarter came, this project was not the highest priority therefore it took a bit of time for me to start making some progress. When the time came, my manager and the senior engineer in the team decided to present a design document of what we wanted to do and how. It was my first big, important, design doc from scratch to be presented alone to my director and other principal engineers. I was nervous and actually with no experience of what I had to write in that doc. What was the information I had to focus on? How to write it? I actually had to kind of <em>sell the idea</em> as well.</p><p>Of course, the fact that it was the first time was actually quite evident: it went to a disaster up to a point that my director called my manager after the meeting to ask for explanations on how there was so little progress and many unanswered questions after many weeks from the start of the quarter.</p><p>That moment represented a turning point in my career. It almost feels like to me that to become a senior engineer I had to go to that sort of a failure. Weirdly, I almost felt embarrassed but at the same time happy that happened. But what exactly was the failure? And why was it so important?</p><p>Let’s start with a fact: one of the biggest lessons I learned in software engineering is that you must know the limits of your system. You must know what is the tipping point at which it does not scale anymore. If you know that, you’ve done an awesome job. This was a frequent exercise we were running with the team in Amazon: we were running these so-called <strong>Game Day</strong> events where the intention was to <em>test</em> the production environment by simulating outages or reduced capacity.</p><p>In other words, what you do is to increasingly remove capacity up to when you start seeing failures. As soon as that happens, you record that moment and you restore everything. That exercise is going to give you information like how much unit of capacity is needed for each amount of requests you handle. Similarly, you can emulate more and more incoming concurrent requests (stress test) and you verify up to where you arrive without failures. All of these are very valid techniques you need to know and do.</p><p>But the point of this note is actually different. Based on the same principle I’ve just explained, <strong>in life you must know your limits in order to succeed</strong>: you must fail in order to be able to make progress. That is why I was happy that this episode happened to me: I knew that from that day on, I would have become better and I would have learned.</p><h3>Be a master in writing design docs</h3><p>For a software engineer, writing design docs should not be a hobby or a “I’ll do it in my free-time” thing. For what it matters, when running for promotions things like your ability to <strong>bring people to an agreement</strong>, your <strong>clarity when writing and expressing concepts</strong> are very much considered.</p><p>With a well-thought design doc, implementing something becomes ridiculously easy. It’s something that you can end-up doing with peace of mind and maybe delegate to your intern or junior engineer.</p><p>Here’s some suggestions when you have to write a design doc.</p><h4>Know your tenets</h4><p>The first time a principal engineer told me “I don’t even know why do we need this, what are our tenets?” I was shocked. I didn’t even know what tenets were and I underestimated the importance of <strong>giving context</strong> and making sure<strong> everyone is on the same page</strong>. On a big scale, tenets are usually supported by a team mission. Practically, is a list of principles for which you are coming up with the idea of solving a specific problem in a specific way. Some examples of tenets are:</p><ul><li>Time-to-market: you’ve done a user research analysis and you’ve realised that you’re way behind your competitors on a specific functionality. Whatever the functionality is, you want to implement a solution that is mindful of the time needed to implement</li><li>Cost: your team is running short of engineering capacity or your company cannot afford as many servers as you need. You need to prefer a solution that has a lower cost than the other</li><li>Availability: having a system that is always available requires some mechanisms like redundancy or replication. This has a cost. If availability is not the most important thing, you should call that out. If it is, you should still call that out too.</li><li>Flexibility: let’s say you’re building a new feature that consists of an insights dashboard. This can contain many types of insights, and there may be new insights for the future. You may need to enlist in your principles the need of designing the system in a flexible way, that allows future customisation without having to rewrite the code.</li></ul><h4>Every problem has more than one solution</h4><p>Every problem has more than one solution, but which one is the best? To know that, that is why the tenets I’ve just explained are important. But one essential thing to do when mastering a design doc is to present in your doc all <strong>the most important potential solutions</strong> that you’ve found out. And most importantly, you need to be able to define the <strong>pros </strong>and <strong>cons </strong>of these solutions. With that, you should also express which one is your <strong>recommended</strong> one. This approach works very well! It’s an incredible booster for a constructive conversation around the design doc. Once, I presented a doc with three solutions, and I proposed one of them. By driving the discussion with the team, one of my colleagues raised a valid point / idea that she came up with while reading one of the non-recommended solutions. Without writing it, probably she would have never realised it.</p><h4>Distinguish high-level details from detailed design</h4><p>I’ve seen so many meetings with managers or directors driven by engineers where they would talk about specific chosen programming languages, or class names or chosen frameworks. If you do so, you’re a bit out of context. Please, know your audience when you’re presenting something, adapt your language to their needs. If you’re talking to a director, you probably want to focus on business values and high-level details. If you’re talking with the engineering team, they’ll probably want to know the detailed design you’ve defined.</p><h3>Think Big!</h3><p>When I approached presenting this project, I had honestly thought 10% of what could have been actually done. The problem is that I had not asked myself the right questions. For example, I didn’t think of how many services we could have integrated with; and I did not do an exhaustive market research (which eventually led us to find the perfect team in AWS to collaborate with).</p><p>The secret of delivering the right value for customers is to <strong>think big</strong>. I like to mention Henry Ford: “<strong>If I had asked people what they wanted, they would have said faster horses</strong>.” Thinking big is the key to success. If you are planning to do A, leave yourself some headroom and plan also for B and C. Of course, you don’t have to think big just for the sake of it. Most everything you do should make sense and bring value for your customers. And please, <a href="https://levelup.gitconnected.com/the-5-most-important-principles-when-working-in-a-big-tech-b5ae2c72d388">know who your customers are</a>.</p><p>Thinking big actually takes time, that’s why most of the time we tend to skip that part. It takes effort, it’s like writing a novel. But you can only gain from it: you’ll create more opportunities for yourself, your team and maybe, who knows, your org.</p><p>There are multiple ways you can think big. My favourite one is: <strong>socialise your ideas</strong>. Every time I talked informally with my colleagues I always realised something I was missing. I genuinely think that I abused those chats to guarantee a successful project. There’s not only that. If you want to think big, you must have uncomfortable conversations, with people that have a completely different background than yours. I can prove you this works with <em>reductio ad absurdum</em>: if you were about to always talk with a copy of yourself, you wouldn’t learn anything new. Different point of views are crucial.</p><p>I hope you enjoyed this small note. I write as a hobby and I am happy to share my experiences. If you’re interested, have a look at <a href="https://marcsuma.medium.com/">my other notes</a>.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=4c4099443357" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>