<?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 Saverio Ferrara on Medium]]></title>
        <description><![CDATA[Stories by Saverio Ferrara on Medium]]></description>
        <link>https://medium.com/@fsferrara?source=rss-bf68b929425f------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/2*eMWKMPBeloeyT8Xm_rNFLg.png</url>
            <title>Stories by Saverio Ferrara on Medium</title>
            <link>https://medium.com/@fsferrara?source=rss-bf68b929425f------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Sun, 22 Nov 2026 05:11:53 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@fsferrara/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[Technical Documentation Mini-Guide]]></title>
            <link>https://medium.com/@fsferrara/technical-documentation-mini-guide-47cd2ad8ece7?source=rss-bf68b929425f------2</link>
            <guid isPermaLink="false">https://medium.com/p/47cd2ad8ece7</guid>
            <dc:creator><![CDATA[Saverio Ferrara]]></dc:creator>
            <pubDate>Mon, 27 Apr 2026 21:13:14 GMT</pubDate>
            <atom:updated>2026-04-27T21:13:14.640Z</atom:updated>
            <content:encoded><![CDATA[<p>Technical documentation exists to give people accurate information fast.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*qLxLTVumpvPgGxf9" /></figure><p>That is the whole job. If a document is hard to find, hard to scan, out of date, duplicated, vague, or padded with unnecessary words, it fails the job no matter how smart the content is.</p><p>Good documentation does more than answer questions:</p><ul><li>It reduces repeated interruptions and chat messages.</li><li>It increases team efficiency.</li><li>It clarifies thinking by forcing ideas into clear language.</li><li>It gives teams training and onboarding material for free.</li><li>It stabilizes and standardizes knowledge across time, people, and systems.</li></ul><p>Think of a document as a script for a human. Code succeeds when instructions are clear, correctly placed, and consistently named. Documentation works the same way.</p><p>Writing also exposes weak understanding. If you cannot explain a system simply and accurately, you probably do not understand it well enough yet.</p><h3>Write for the actual reader</h3><p>Write with the reader’s behavior in mind, not the behavior you wish they had.</p><ul><li>Readers do not read linearly. They scan and satisfice.</li><li>They are trying to complete a task, not enjoy a reading experience.</li><li>The document is a means to an end, so every extra second feels expensive.</li><li>If they have to think too hard, they leave.</li><li>If the document is wrong once, they may stop trusting it entirely.</li><li>Nobody wants to read documentation longer than necessary.</li><li>Nobody wants to come back later to update it either, so write with maintenance in mind.</li></ul><p>This is why speed matters so much. A document that is technically correct but slow to use still fails.</p><h3>Make documentation part of the work</h3><p>Treat documentation the way you treat tests: part of the change, not an optional follow-up.</p><p>Whenever behavior changes, ask:</p><ol><li>What document needs to change?</li><li>Does that document already exist?</li><li>If not, should it exist?</li></ol><p>Add or update documentation when:</p><ul><li>People will need the information again.</li><li>The information is easy to forget.</li><li>Teammates ask the same question repeatedly.</li><li>The knowledge matters beyond one person or one moment.</li></ul><p>Do not create a new page just because you can. Add content only when it will help future readers. Budget time for documentation when estimating work. That is how documentation stays current.</p><p>Old documentation is worse than no documentation. Missing information creates friction. Wrong information destroys trust.</p><h3>Put documentation in the right place</h3><p>Before you write, decide where the content belongs. Documentation quality depends on information architecture as much as writing quality.</p><h3>Choose the right home</h3><p>Different kinds of information belong in different places.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Yeg8AsTcSU_8kIfO6KM4Cw.png" /></figure><p>Use conversations for conversation. Use durable documentation for durable knowledge.</p><p>Do not leave important answers buried in:</p><ul><li>Chat threads</li><li>Page comments</li><li>Code review discussions</li><li>Tickets</li><li>Email chains</li></ul><p>If a conversation produces knowledge someone will need again, move that knowledge into the document itself.</p><h3>Build one hierarchy</h3><p>It does not matter how good the content is if no one can find it.</p><ul><li>Build a clear hierarchy for your documentation.</li><li>Maintain it once it exists.</li><li>Honor it instead of creating exceptions.</li><li>Do not rely on search as the primary navigation model.</li><li>Do not make readers guess which tool, space, repo, or page to trust.</li></ul><p>One hierarchy works. Multiple competing hierarchies create hesitation, duplication, and drift.</p><h3>Group related content</h3><p>Keep related topics together so the reader can predict where something belongs.</p><ul><li>Put setup near setup.</li><li>Put troubleshooting near troubleshooting.</li><li>Put architecture near architecture.</li><li>Put shared policies in one shared location instead of scattering them across teams.</li></ul><p>If a topic could reasonably live in more than one place, choose one canonical home and link to it from elsewhere.</p><h3>Choose an ordering system</h3><p>Navigation communicates meaning even before a reader opens a page. Choose an order on purpose.</p><p>Useful ordering patterns include:</p><ul><li>Most important first</li><li>Most frequently used first</li><li>Chronological order</li><li>Dependency order, where one thing must exist before the next</li></ul><p>Do not leave ordering accidental. Accidental order makes navigation feel arbitrary.</p><h3>Limit depth or limit breadth</h3><p>Hierarchies usually work best in one of two shapes:</p><ul><li>Flat: fewer levels, more options per level</li><li>Deep: more levels, fewer options per level</li></ul><p>Pick one bias and keep it usable.</p><ul><li>Flat hierarchies work well when the domain is simple and the top-level choices are obvious.</li><li>Deep hierarchies work well when the domain is complex but each click remains a clear, unambiguous choice.</li></ul><p>Do not create a hierarchy that is both too broad and too deep. That gives readers too many choices at every step and slows them down.</p><h3>Make the structure scalable</h3><p>A useful hierarchy works today and still works after the content grows.</p><p>When you design navigation, ask:</p><ul><li>Where will future topics go?</li><li>Will this category still make sense when it has ten times as much content?</li><li>Are there top-level buckets that can absorb related pages over time?</li></ul><p>Scalability matters because documentation is a living system, not a one-time artifact.</p><h3>Remove anything that is not helping</h3><p>Anything that is not contributing is in the way.</p><p>Remove:</p><ul><li>Outdated content</li><li>Duplicate content</li><li>Empty sections</li><li>Stub pages</li><li>Dead links</li><li>Misleading page titles</li><li>Navigation paths that lead nowhere</li><li>Long introductions that delay the answer</li><li>Sections that exist only because a template created them</li></ul><p>This is not cleanup for its own sake. It is performance work.</p><p>When documentation becomes noisy, readers learn that searching it is a waste of time. Once that happens, they ask people directly instead of using the docs.</p><h3>Remove outdated content first</h3><p>Outdated content is the highest-cost failure mode because it teaches readers not to trust anything.</p><p>When you find stale content:</p><ul><li>Fix it immediately if you can.</li><li>Remove it if it cannot be validated quickly.</li><li>Redirect to the correct source if another page has replaced it.</li></ul><p>Never leave known-bad instructions in place just because removing them feels inconvenient.</p><h3>Remove duplication</h3><p>Keep documentation DRY.</p><p>Say something once in one canonical place, then link to it.</p><p>Duplicate content causes at least one of these failures:</p><ul><li>Readers waste time deciding which version is correct.</li><li>Writers update one copy and forget the others.</li><li>Search results show multiple similar answers with different levels of freshness.</li></ul><p>Single-sourcing is a documentation quality feature, not just a maintenance preference.</p><h3>Link the answer instead of retyping it</h3><p>When someone asks a question that is already documented, send the link to the document.</p><p>Do not re-answer the question in chat unless the document is wrong. If the document is wrong or unclear, fix it first, then send the link.</p><p>This reinforces the habit you want:</p><ul><li>Read the docs first.</li><li>Ask people second.</li></ul><h3>Structure pages for scan speed</h3><p>A page should reveal its answer quickly, even to someone who only reads the headings and list items.</p><p>Use this structure whenever it fits:</p><ol><li>Start with the answer or outcome.</li><li>Put prerequisites near the top.</li><li>Break the content into clear sections.</li><li>Use lists for parallel information.</li><li>Put exact commands, paths, names, and versions where readers need them.</li><li>End with validation steps, troubleshooting, or next actions if they help.</li></ol><p>For procedural pages, a strong default shape is:</p><ol><li>Purpose</li><li>Prerequisites</li><li>Steps</li><li>Verification</li><li>Troubleshooting</li><li>Related links</li></ol><p>Do not force every page into the same template, but do give readers a predictable pattern.</p><h3>Write sentences that move quickly</h3><p>The best documentation style reduces mental processing time.</p><h3>Say it once</h3><p>Do not repeat the same explanation across multiple pages or multiple sections of the same page.</p><ul><li>Write the primary explanation once.</li><li>Link back to it when needed.</li><li>Avoid near-duplicate summaries that will drift over time.</li></ul><h3>Cut extra words</h3><p>Be concise. Extra words slow the reader down.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*3H7TqE-TIElW5Hv8cxY8uw.png" /></figure><p>Common filler to remove:</p><ul><li>please</li><li>simply</li><li>just</li><li>in order to</li><li>you can</li><li>this document will explain</li></ul><h3>Be specific</h3><p>Specific writing is faster than vague writing.</p><p>Prefer:</p><ul><li>Exact file paths over “the config file”</li><li>Exact versions over “recently”</li><li>Exact team names over “the team”</li><li>Exact commands over paraphrases</li><li>Exact page names over “the other doc”</li></ul><p>Every vague phrase creates work for the reader.</p><h3>Use one term for one thing</h3><p>Be consistent.</p><p>If a component is called event processor, do not also call it worker, task runner, and background job in the same guide unless those names mean different things.</p><p>The same rule applies to links and headings:</p><ul><li>Link text should match the destination heading.</li><li>Section names should stay stable unless the meaning changes.</li></ul><p>Consistency reduces hesitation.</p><h3>Prefer lists over paragraphs</h3><p>Lists are easier to scan than dense prose.</p><p>Use a list when:</p><ul><li>Items are parallel</li><li>Steps must be completed in order</li><li>You are naming options, requirements, causes, or outcomes</li></ul><p>Use paragraphs when the reader actually needs connected explanation. Do not turn everything into prose by default.</p><h3>Chunk content under useful headings</h3><p>Every distinct idea deserves its own heading or subheading.</p><p>Good headings help the reader:</p><ul><li>Skip irrelevant content</li><li>Build a mental map of the page</li><li>Return later and find the same section again</li></ul><p>Long unbroken text blocks are hard to scan, hard to maintain, and easy to abandon.</p><h3>Start headings and list items with verbs</h3><p>Verb-led phrasing is easier to process because it tells the reader what action or concept matters.</p><p>Prefer:</p><ul><li>Use and Contribute over Using and Contributing</li><li>Format a change request over Change request format</li><li>Set up the component over Component setup</li><li>Include the component on your page over Including the component on your page</li></ul><p>This also helps you avoid nominalizations: turning verbs into nouns when direct action would be clearer.</p><h3>Use active voice</h3><p>Active voice is usually shorter and clearer.</p><p>Prefer:</p><ul><li>Use the package manager to install packages. over The package manager is used to install packages.</li><li>We configured the build to... over The build has been configured to...</li></ul><p>Passive voice hides who does what and usually adds words.</p><h3>Use simple language</h3><p>Write the way engineers actually talk when they are being clear.</p><p>Prefer:</p><ul><li>must over one must</li><li>use over utilize</li><li>before over prior to</li><li>if over in the event that</li></ul><p>Simple language is not less technical. It is faster.</p><h3>Drop politeness padding</h3><p>Instructional writing does not need conversational padding.</p><p>Prefer:</p><ul><li>Install the version manager and use it to install the runtime. over Please install the version manager and then you can use it to install the runtime.</li><li>Use the sidebar to navigate the documentation. over For navigating the documentation better please use the sidebar.</li><li>Choose one of these options. over You can choose one of these options.</li></ul><p>Clarity matters more than courtesy formulas in a procedural document.</p><h3>Avoid noun strings</h3><p>Long stacks of nouns force the reader to unpack the sentence.</p><p>Prefer:</p><ul><li>the process for restoring a failed cache after deployment</li></ul><p>instead of:</p><ul><li>post-deployment cache restoration failure response process</li></ul><p>If a phrase sounds compressed or bureaucratic, rewrite it.</p><h3>Avoid unnecessary formality</h3><p>Formal language often adds distance without adding precision.</p><p>Write as if you were explaining the work clearly to another engineer at a desk. Keep the tone direct, plain, and factual.</p><h3>Follow a repeatable writing workflow</h3><p>Use a consistent process so documentation quality does not depend on inspiration.</p><ol><li>Define the audience.</li><li>State the question the document answers.</li><li>Choose the canonical home for the content.</li><li>Check whether a page already exists and extend it if it should.</li><li>Sketch the headings before writing paragraphs.</li><li>Draft the content quickly.</li><li>Revise for accuracy, hierarchy, scan speed, and concision.</li><li>Remove duplication and move durable answers out of conversations.</li><li>Validate commands, links, versions, and names.</li><li>Publish it where readers already expect to find it.</li></ol><p>The earlier you make the structural decisions, the less rewriting you need later.</p><h3>Check the document before you publish it</h3><p>Use this checklist before you ship a new page or update an existing one.</p><h3>Confirm purpose and audience</h3><ul><li>Can you state the one question this document answers?</li><li>Is the intended reader obvious?</li><li>Does the page lead with the answer, outcome, or purpose?</li></ul><h3>Confirm placement and hierarchy</h3><ul><li>Is this the correct home for the content?</li><li>Does an existing page already cover this topic?</li><li>Is there one canonical source instead of several competing pages?</li><li>Can a reader find it through navigation without relying on search?</li><li>Is the navigation order clear and intentional?</li><li>Will the hierarchy still work as more content is added?</li></ul><h3>Confirm content quality</h3><ul><li>Is every instruction still accurate today?</li><li>Did you remove outdated, duplicated, or empty content?</li><li>Did you move important knowledge out of comments, chat, or tickets?</li><li>Are page titles and headings truthful and specific?</li><li>Are links current and useful?</li><li>Did you avoid examples tied to a specific vendor, product, or organization unless the page requires them?</li></ul><h3>Confirm writing quality</h3><ul><li>Did you remove filler such as please, simply, just, in order to, and you can?</li><li>Did you replace vague wording with exact names, paths, versions, and commands?</li><li>Did you use one term consistently for each concept?</li><li>Did you convert long paragraphs into lists where appropriate?</li><li>Do headings and list items start with strong verbs where possible?</li><li>Did you prefer active voice, simple language, and direct phrasing?</li><li>Did you avoid noun strings and unnecessary formality?</li></ul><h3>Use this standard for every document</h3><p>The same principles apply whether you are writing:</p><ul><li>A README</li><li>A setup guide</li><li>An architecture overview</li><li>A runbook</li><li>A troubleshooting page</li><li>An onboarding guide</li><li>A policy page</li><li>A design document</li></ul><p>In every case, the standard is the same:</p><ul><li>Put the content in the right place.</li><li>Keep one trusted source of truth.</li><li>Make it easy to scan.</li><li>Keep it accurate.</li><li>Remove anything that gets in the way.</li></ul><p>If people reliably use the documentation before asking someone for help, the documentation is doing its job.</p><p><em>Originally published at </em><a href="https://fsferrara.github.io/"><em>https://fsferrara.github.io</em></a><em> on April 27, 2026.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=47cd2ad8ece7" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Multi-phase Process Prompt]]></title>
            <link>https://medium.com/@fsferrara/multi-phase-process-prompt-072658e5d41c?source=rss-bf68b929425f------2</link>
            <guid isPermaLink="false">https://medium.com/p/072658e5d41c</guid>
            <category><![CDATA[prompt-engineering]]></category>
            <dc:creator><![CDATA[Saverio Ferrara]]></dc:creator>
            <pubDate>Wed, 31 Dec 2025 16:36:00 GMT</pubDate>
            <atom:updated>2025-12-31T16:36:00.017Z</atom:updated>
            <content:encoded><![CDATA[<p>Complex tasks often overwhelm AI models. When you ask for a multi-step workflow in a single prompt, the model often rushes to the end, skipping crucial intermediate steps or hallucinating details.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*ol1Yozg06wQ1H6HUIoDq7g.jpeg" /></figure><h3>The Problem</h3><p>Without a structured process, the model tries to do everything at once:</p><pre>Analyze this data, write a summary, create a marketing plan based on it, and then write three email drafts.</pre><p>The model might write a weak summary, a generic plan, and forget the emails entirely because its context window or attention span gets diluted.</p><h3>The Solution</h3><p>The <strong>Multi-phase Process</strong> technique forces the model to work sequentially. You explicitly define phases and steps, instructing the model to complete one phase before moving to the next.</p><p>This mimics how humans handle complex projects: we don’t write the code, the tests, and the documentation simultaneously. We do them in phases.</p><h3>How to Apply This</h3><ol><li><strong>Define the Goal</strong>: State the ultimate objective clearly.</li><li><strong>Break it Down</strong>: Split the workflow into high-level <strong>PHASES</strong>.</li><li><strong>Detail the Steps</strong>: Within each phase, list specific <strong>STEPS</strong>.</li><li><strong>Enforce Sequence</strong>: Add a “Gatekeeper” instruction (e.g., “Only proceed to Phase 2 once Phase 1 is executed”).</li></ol><h3>Example in Practice</h3><p>Here is how you can structure a prompt to guide an AI through creating a technical tutorial.</p><p><strong>Multi-phase Prompt Template:</strong></p><pre># Goal<br>Write a high-quality technical tutorial based on a user-provided topic.<br><br>Carefully read and execute the instructions in the following files IN SEQUENCE, following their numbered PHASE.  Only proceed to the next PHASE once the previous numbered PHASE has been executed.<br><br>PHASES to follow in sequence:<br><br># PHASE 1: Outline &amp; Audience<br>Step 1. Ask the user for the topic and target audience level (Beginner, Intermediate, Advanced).<br>Step 2. Generate a detailed outline with main headings.<br>Step 3. WAIT for user approval or feedback on the outline.<br><br># PHASE 2: Drafting Content<br>Once the outline is approved:<br>Step 1. Write the Introduction (Hook + Prerequisites).<br>Step 2. Write the Body paragraphs step-by-step.<br>Step 3. Write the Conclusion and &quot;Next Steps&quot;.<br><br># PHASE 3: Review &amp; Refine<br>Step 1. Review the draft for clarity and tone.<br>Step 2. Check that all code snippets (if any) are properly formatted.<br>Step 3. Output the final tutorial in Markdown format.</pre><p>By using this structure, the model stops after Phase 1, allowing you to correct the course before it wastes tokens generating the wrong content.</p><h3>Advanced Patterns</h3><p>To make your process even more robust, borrow concepts from programming:</p><h3>1. Conditional Logic</h3><p>Use <strong>IF/THEN</strong> statements to handle variables.</p><blockquote>“IF the user provides a topic, proceed to Phase 2. IF NOT, ask for a topic.”</blockquote><h3>2. Loops</h3><p>Instruct the model to <strong>LOOP</strong> through items.</p><blockquote>“For EACH section in the outline, generate a summary.”</blockquote><h3>3. Templates</h3><p>Provide a <strong>Template</strong> for the model to fill out.</p><blockquote>“Populate the response using this JSON structure…”</blockquote><h3>Conclusion</h3><p>The Multi-phase Process technique transforms the AI from a simple text generator into a workflow agent. It ensures accuracy, allows for human-in-the-loop feedback, and handles complexity that would otherwise break a single-shot prompt. Use this for any task that requires planning, execution, and review.</p><p><em>Originally published at </em><a href="https://fsferrara.github.io/"><em>https://fsferrara.github.io</em></a><em> on December 30, 2025.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=072658e5d41c" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Use placeholders in your prompts]]></title>
            <link>https://medium.com/@fsferrara/use-placeholders-in-your-prompts-c05cfa726555?source=rss-bf68b929425f------2</link>
            <guid isPermaLink="false">https://medium.com/p/c05cfa726555</guid>
            <category><![CDATA[prompt-engineering]]></category>
            <dc:creator><![CDATA[Saverio Ferrara]]></dc:creator>
            <pubDate>Wed, 31 Dec 2025 16:24:45 GMT</pubDate>
            <atom:updated>2025-12-31T16:29:39.546Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*8XDtCc0njkUjd0UW0WHEow.jpeg" /></figure><p>Placeholders make your prompts reusable and adaptable. Instead of rewriting prompts for every variation, use variables to swap in different content, saving time and maintaining consistency.</p><h3>The Problem</h3><p>Without placeholders, you rewrite the same prompt repeatedly:</p><pre>Write feedback for Sarah about her work as a Senior Developer.<br><br>Write feedback for Marcus about his work as a Product Manager.<br><br>Write feedback for Alex about their work as a Design Lead.</pre><p>Each version requires full rewriting, introducing inconsistency and wasting effort.</p><h3>The Solution</h3><p>Replace variable content with placeholders using [BRACKETS]:</p><pre>Write feedback for [NAME] about [HIS/HER] work as a [ROLE].</pre><p>Now reuse it by filling in the brackets:</p><ul><li>[NAME] = &quot;Sarah&quot; | &quot;Marcus&quot; | &quot;Alex&quot;</li><li>[HIS/HER] = &quot;her&quot; | &quot;his&quot; | &quot;their&quot;</li><li>[ROLE] = &quot;Senior Developer&quot; | &quot;Product Manager&quot; | &quot;Design Lead&quot;</li></ul><h3>How to Apply This</h3><ol><li><strong>Identify variables</strong>: What changes between similar tasks? (people, dates, topics, quantities, tones)</li><li><strong>Replace with placeholders</strong>: Use [DESCRIPTIVE_NAME] for each variable</li><li><strong>Document the template</strong>: Save it for reuse</li><li><strong>Fill and regenerate</strong>: Substitute values and run the prompt as many times as needed</li></ol><h3>Example in Practice</h3><p>Once you master basic placeholders, combine them with structured prompting (Role, Task, Instructions, Context) for even more powerful templates.</p><p><strong>Structured Template with Placeholders:</strong></p><pre>Role: You are a professional feedback writer preparing performance evaluations.<br><br>Task: Write concise feedback for [NAME] about [HIS/HER] performance as a [ROLE].<br><br>Instructions:<br>- Write in third person (intended for the manager to read)<br>- Keep the feedback under 150 words<br>- Balance achievements with growth opportunities<br>- Use specific examples, not generalizations<br><br>Context:<br>Key contributions:<br>- [CONTRIBUTION_1]<br>- [CONTRIBUTION_2]<br><br>Areas for development:<br>- [SUGGESTION_1]<br>- [SUGGESTION_2]</pre><p>This template uses placeholders that maintain quality while adapting to different people and situations.</p><p><strong>Instance Example:</strong></p><pre>Role: You are a professional feedback writer preparing performance evaluations.<br><br>Task: Write concise feedback for Sarah about her performance as a Senior Developer.<br><br>Instructions:<br>- Write in third person (intended for the manager to read)<br>- Keep the feedback under 150 words<br>- Balance achievements with growth opportunities<br>- Use specific examples, not generalizations<br><br>Context:<br>Key contributions:<br>- Led the backend refactoring that improved API response time by 40%<br>- Mentored two junior developers through their first major feature<br>- Consistently delivered code reviews with actionable feedback<br><br>Areas for development:<br>- Document architectural decisions more thoroughly for team reference<br>- Take on cross-team collaboration on the data pipeline project</pre><h3>Variants</h3><p>Some systems are formalizing placeholder syntax further:</p><ul><li><strong>Double curly braces</strong>: {PLACEHOLDER}</li><li><strong>Percent signs</strong>: %PLACEHOLDER%</li><li><strong>Angle brackets</strong>: &lt;PLACEHOLDER&gt;</li><li>etc…</li></ul><p>An alternative definition for a prompt with placeholders is <strong>dynamic prompt</strong>.</p><h3>Conclusion</h3><p>Placeholders transform ad-hoc prompting into a systematic, scalable workflow. Start simple with a few variables, then evolve your templates as you identify patterns in your work. The time invested in building reusable templates pays dividends every time you skip rewriting the same prompt from scratch.</p><p><em>Originally published at </em><a href="https://fsferrara.github.io/"><em>https://fsferrara.github.io</em></a><em> on December 27, 2025.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=c05cfa726555" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Prompt Engineering Crash Course]]></title>
            <link>https://medium.com/@fsferrara/prompt-engineering-crash-course-f3bed870e7b8?source=rss-bf68b929425f------2</link>
            <guid isPermaLink="false">https://medium.com/p/f3bed870e7b8</guid>
            <category><![CDATA[prompt-engineering]]></category>
            <category><![CDATA[ai]]></category>
            <dc:creator><![CDATA[Saverio Ferrara]]></dc:creator>
            <pubDate>Tue, 09 Dec 2025 19:26:45 GMT</pubDate>
            <atom:updated>2025-12-09T19:29:25.252Z</atom:updated>
            <content:encoded><![CDATA[<p>The quality of AI-generated output directly mirrors the quality of your input. When you submit an imprecise or unclear prompt, expect generic or irrelevant responses in return. Conversely, a thoughtfully constructed prompt can generate remarkably accurate, actionable results that surface valuable insights or spark creativity.</p><p>This crash course introduces you to the essential concepts and techniques for effective prompt engineering with generative AI.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*nAgb5eRqC0GdtDoXqfIkHw.jpeg" /></figure><h3>Prompt Structure</h3><p>Effective prompts follow a deliberate structure. Understanding the difference between naive and structured approaches will transform your results.</p><p>A beginner usually writes naive prompts, which is the most straightforward approach.</p><ul><li><strong>Naive Prompt</strong>: You provide the model with a task description and nothing more.<br>Example: “Write a children’s fable.”</li></ul><p>Naive prompts may work but they have limitations: they may produce imprecise, potentially biased, or even incorrect outputs.<br>To enhance prompt effectiveness, consider adopting a more structured approach.</p><p><strong>Structured Prompt</strong>: Break down your prompt into clear components:</p><ul><li><strong>Role</strong>: The persona or expertise you’d like the AI to embody<br>Example: children’s author</li><li><strong>Task</strong>: The specific action or deliverable you need<br>Example: write a children’s fable with a moral lesson</li><li><strong>Instructions</strong>: The specific approach or constraints for execution<br>Example: Use simple vocabulary, include animal characters, convey a clear lesson, keep under 300 words</li><li><strong>Context</strong>: Relevant background or situational information<br>Example: target audience age group, desired moral theme, preferred narrative style</li></ul><p>The resulting structured prompt would be:</p><blockquote>“As a children’s author, write a fable that includes animal characters and conveys a clear moral lesson. Use simple vocabulary suitable for ages 5–8, and keep the story under 300 words.”</blockquote><p>Try it yourself and observe the difference in output quality!</p><h3>Prompt Principles</h3><p>Once you’ve mastered structure, these five principles will help you fine-tune your prompts for even better results.</p><p>Beyond structure, mastering these five principles will elevate your prompt engineering. We’ll demonstrate each using the “Write a children’s fable” task.</p><h3>1. Give Direction</h3><p>Start by describing the desired style or tone in detail. Adding direction helps the model understand not just <em>what</em> you want, but <em>how</em> you want it.</p><p><strong>Naive prompt:</strong></p><pre>Write a children&#39;s fable.</pre><p><strong>Directed prompt:</strong></p><pre>Write a children&#39;s fable in the style of Aesop. Focus on animals as characters and include a clear moral lesson at the end. The tone should be warm and slightly humorous.</pre><p>By specifying the style reference, character types, and tone, you steer the model toward more targeted output.</p><h3>2. Specify Format</h3><p>Define the exact structure you expect in the response. Format specifications eliminate ambiguity and make outputs more predictable.</p><p><strong>Directed prompt:</strong></p><pre>Write a children&#39;s fable in the style of Aesop. Focus on animals as characters and include a clear moral lesson at the end. The tone should be warm and slightly humorous.</pre><p><strong>Format-specified prompt:</strong></p><pre>Write a children&#39;s fable in the style of Aesop with this structure:<br>- Title: [One-line fable title]<br>- Characters: [List of animal characters with brief descriptions]<br>- Story: [150-200 words]<br>- Moral: [One clear takeaway lesson in a single sentence]</pre><p>Specifying format produces consistent, well-organized results that are easier to use and iterate on.</p><h3>3. Provide Examples</h3><p>Include one or two examples of excellent fables to establish the quality and style baseline you’re targeting.</p><p><strong>Format-specified prompt + examples:</strong></p><pre>Write a children&#39;s fable in the style of Aesop with this structure:<br>- Title: [One-line fable title]<br>- Characters: [List of animal characters with brief descriptions]<br>- Story: [150-200 words]<br>- Moral: [One clear takeaway lesson in a single sentence]<br><br>Example:<br>- Title: The Tortoise and the Hare<br>- Characters: Tortoise (slow but determined), Hare (fast but overconfident)<br>- Story: [Example story about a race and its lesson]<br>- Moral: Steady effort beats natural talent without discipline.</pre><p>Examples anchor the model’s output to your quality expectations far better than words alone.</p><h3>4. Evaluate Quality</h3><p>Once you receive output, assess it against specific criteria to identify what works and what doesn’t.</p><p><strong>Evaluation criteria for fables:</strong></p><ul><li>Does the moral feel authentic and age-appropriate?</li><li>Are the animal characters distinct and memorable?</li><li>Is the story concise without sacrificing clarity?</li><li>Would a child find this engaging?</li></ul><p>After generating a fable, rate it on these criteria. This process reveals which prompt elements drive better quality and helps you refine future prompts.</p><h3>5. Divide Labor</h3><p>For complex tasks, break them into sequential steps. Generate components separately, then combine them.</p><p><strong>Multi-step approach:</strong></p><pre>Step 1: Generate a list of 3 animal character pairs with contrasting traits (e.g., fast vs. slow, honest vs. cunning).<br><br>Step 2: For each pair, outline a short conflict or challenge that could teach a valuable lesson.<br><br>Step 3: Choose the most compelling scenario and write the full fable using the format from Principle 2.<br><br>Step 4: Have the model rate the moral for clarity and age-appropriateness (1-5 scale).</pre><p>Breaking work into smaller, focused tasks reduces errors and gives you control over each component.</p><h3>Conclusion</h3><p>This is only the beginning of your prompt engineering journey. There are other advanced techniques to explore, and all of them must be re-evaluated as AI models evolve. To tell the truth, I think prompt engineering is not an exact science but rather an art form that improves with practice and creativity.</p><p><em>Originally published at </em><a href="https://fsferrara.github.io/"><em>https://fsferrara.github.io</em></a><em> on December 5, 2025.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=f3bed870e7b8" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[How Do I Get Things Done]]></title>
            <link>https://medium.com/@fsferrara/how-do-i-get-things-done-b992a546cc67?source=rss-bf68b929425f------2</link>
            <guid isPermaLink="false">https://medium.com/p/b992a546cc67</guid>
            <category><![CDATA[gtd-getting-things-done]]></category>
            <category><![CDATA[time-management]]></category>
            <category><![CDATA[gtd]]></category>
            <category><![CDATA[task-management]]></category>
            <category><![CDATA[productivity]]></category>
            <dc:creator><![CDATA[Saverio Ferrara]]></dc:creator>
            <pubDate>Sun, 06 Aug 2023 22:00:13 GMT</pubDate>
            <atom:updated>2024-08-22T10:59:08.137Z</atom:updated>
            <content:encoded><![CDATA[<p>Almost everyone knows GTD. It is the acronym of “Getting Things Done” which is a productivity framework introduced by David Allen in his book “Getting Things Done (2002) — The Art of Stress-Free Productivity”.<br>The key reason to have a such framework is that our mind is great at creating stuff, but it’s terrible at tracking it. And yet there’s a good chance you’re tracking tons of stuff in your head right now. Stuff that drains your energy and clogs your creativity.</p><p>I became a GTD user during a period in which I was so full of things to do that I couldn’t work well. Just keeping the task list in mind completely occupied my brain and I could no longer concentrate on the task in progress.<br>I GTD love it since it keeps track of my tasks and allows me to focus on each one of them individually. Also, it allows me to work on several workstreams (or projects) during the same period (e.g. a quarter).</p><p>I’d like to recap in this post how I applied the framework. My main goals were to keep the entire procedure lightweight, portable, and almost paperless.</p><h3>The tools</h3><p>Let’s get straight to what everyone wants to know: the tool/software I adopted.</p><p>The first tool is a digital calendar. I don’t suggest a specific calendar over another since they are all similar and I bet you already have one. In my calendar, I keep track of ticklers, recurring tasks, and, of course, meetings.</p><p>Another tool is… a notebook! Even if I can easily replace it with a digital solution, I can’t stop using paper and pen.</p><p>The last tool is a Trello Board.</p><h3>GTD Trello Board</h3><p>The reasons I chose Trello are several, I can list some from my top of mind:</p><ul><li>a board allows you to have a quick look at all your lists,</li><li>it allows you to quickly move cards (i.e. stuff) from one list to another,</li><li>Trello cards may have attachments like images, checklists, documents, etc.,</li></ul><p>I shared <a href="https://trello.com/b/JD7ituj7/my-gtd-template">my Trello Board Template for GTD</a>. It is ready to use and contains useful instructions in the right place to reduce friction during its usage.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*Nvvj9Kp5xhlCveDo.png" /></figure><p>In the next sections, each list of this board is described.</p><h4>Horizons of Focus</h4><p>I like to start top-down and define what is my purpose first. Keeping it in mind allows me to define priorities for each task and discard what is not useful to pursue the purpose.</p><p>Allen suggests working on the big picture thinking with 5 increasingly top-down horizons:</p><ul><li>Horizon 5: Purpose and principles — The purpose, mission, and values at the heart of your existence;</li><li>Horizon 4: Vision — A broad overview of what success looks like for you on a 3–5 years timeline;</li><li>Horizon 3: Goals — The larger outcomes you’d like to make manifest in the next 1–2 years;</li><li>Horizon 2: Areas of Focus and Accountabilities — The roles and areas of life you are committed to focussing on in the next 12 months;</li><li>Horizon 1: Current projects — The list of any outcome you are committed to realizing in &lt; 12 months. Personally, for this horizon, I prefer defining activities that can be done in a quarter (i.e. &lt;= 3 months).</li></ul><p>In my GTD implementation, I defined a Trello list “Horizons of Focus” containing a card for each horizon from H5 (i.e. Horizon 5) to H2 (i.e. Horizon 2). In those cards, there is just a textual description of the related horizon.</p><p>The Horizon 1 is an exception since it is simplistic to describe it in a single card, especially if you are keeping track of multiple workstreams. For H1 (i.e. Horizon 1) I defined a dedicated list named “Workstreams”.</p><p>The “Ground” for those horizons are the tasks/actions at the core of getting things done. They are all the cards contained in the lists: “Later”, “This Week”, “Today”, and “Completed”.</p><h4>Workstreams</h4><p>The Workstreams (that are Projects and Outcomes) are listed here. I have one card for each of them containing a description and split into several milestones. If needed, in the workstream card I can define the list of tasks using a “Checklist”. Trello allows me to convert a checklist item into a new card so that I can pull tasks from the workstream when I need them.</p><p>I suggest reviewing the “Workstream” list monthly. It is good to keep in mind what are the current activities and what is their status.</p><p>A note about acquiring new skills. Learning is something everyone should continue doing with constancy. In some cases, it is essential to continue being employable. I use a dedicated workstream to keep track of this activity and be sure to dedicate to it a portion of my time.</p><h4>Inbox</h4><p>The “Inbox” list is where I stockpile everything. I’m careful not to leave stuff on this list for too long. I usually review this list once or twice a day following the Step 1–2–3 procedure described below.</p><p><strong>Step 1: Stockpile</strong></p><p>Stockpiling is “Step 1” to enter in my GTD system and means capturing all your outstanding stuff. It’s a vital step because anything that doesn’t get in your GTD system will stay on your mind. And anything that stays on your mind eats energy and kills creativity.</p><p>Trello allows me to define an inbox email address I can use to forward stuff (usually emails). I can also stockpile from my smartphone since Trello is accessible from it.<br>I may have other stuff in my physical notebook. At the end of the day, I store all the pending stuff in the “Inbox” so I don’t leave it off the system.</p><p>Once stockpiling is done, the “Inbox” list is not executable yet. Each item in the list needs to be clarified in “Step 2”.</p><p><strong>Step 2: Clarify</strong></p><p>Clarifying an item means answering questions about it. These questions are:</p><ul><li>What is it?</li><li>What’s the desired outcome?</li><li>What’s the priority?</li><li>Is it important for my workstreams?</li><li>Can I delegate it?</li><li>How much time is required to do it?</li><li>Is it a time-sensitive activity? Does it have an ETA?</li></ul><p>Usually, I clarify my card by adding details to the card. In particular, I use labels to specify the priority.<br>Once the clarification of a card is done, a card should leave the “Inbox” list and go to another one. This is the “organization” phase (i.e. Step 3).</p><p><strong>Step 3: Organize</strong></p><p>Organizing is the process of addressing the cards on the “Inbox” list. I use an algorithm like this one for each card:</p><ul><li>If I can do the activity described in the card in less than 2 minutes, then I proceed and do it. Once completed, the card goes into the “Completed” list.</li><li>If the card can be safely delegated, then I’ll delegate it and wait. The card goes into the “Waiting For (delegated)” list. I check this latter list daily to ask if there are updates.</li><li>If the card scope is too big, then it is non-actionable ‘stuff’. It is like a “project”, so something that needs to be refined. In this case, I can either plan a time slot to refine the card or (if it is not urgent) just store the card content in a proper place for future reference (e.g. amongst my computer documents).</li><li>If it is something not urgent, then I’ll defer the activity and the card goes into the “Later” list.</li><li>If it is something urgent, then I’ll move the card to the “Today” or “This Week” list. It depends on how much urgent is it.</li><li>Trashing what’s left. The entire stockpile-clarify-organize process allows us to detect activities that are not important.</li></ul><h4>Later</h4><p>In the “Later” list there are the cards I have to execute. Each week, I review this list and define the week’s scope by moving the cards I want to execute during that week to the “This Week” list.<br>If a card is in the “Later” list for a long time, you may want to move it to the “Someday/Maybe” list.</p><h4>This Week</h4><p>In the “This Week” list there are the cards part of the current week’s scope. Each morning, when I plan the activity for the day, I move some cards from the “This Week” list to the “Today” list.</p><h4>Today</h4><p>The “Today” list contains the cards that are in the scope of the day.<br>Here is where my notebook comes into play. I don’t particularly appreciate stating at this digital list all day long. For this reason, I copy the card titles on paper (like a task list) and then execute them.<br>At the end of the day, the done cards will be moved into the “Completed” list.</p><p>If you want to go completely paperless, then this list can be used instead of the notebook.</p><h4>Someday/Maybe</h4><p>The “Someday/Maybe” list contains outcomes or actions you may like to take one day, under different circumstances.<br>Each quarter, or when I start a new workstream, I review the entire GTD board. On this occasion, I may re-organize the cards by moving them around. During this review, you can trash the cards that are in the “Someday/Maybe” list that are no more relevant.</p><h3>Are you using GTD?</h3><p>In conclusion, implementing the GTD framework has transformed my productivity and focus. With a digital calendar, trusty notebook, and Trello board, I’ve achieved a lightweight, organized, and almost paperless workflow.</p><p>The Horizons of Focus provided clarity on my purpose, goals, and priorities, while Trello streamlined task management from “Inbox” to “Completed.”</p><p>If you’re willing to use GTD, as you embrace it, I suggest you to customize it to fit your needs. Embrace stress-free productivity and enjoy a fulfilling life with purpose and efficiency.</p><p>Happy organizing and best wishes!</p><p><em>Originally published at </em><a href="https://fsferrara.github.io/"><em>https://fsferrara.github.io</em></a><em> on August 6, 2023.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=b992a546cc67" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Error budget burn rate alerts for major outages]]></title>
            <link>https://medium.com/@fsferrara/error-budget-burn-rate-alerts-for-major-outages-3e999f8a200d?source=rss-bf68b929425f------2</link>
            <guid isPermaLink="false">https://medium.com/p/3e999f8a200d</guid>
            <category><![CDATA[reliability-engineering]]></category>
            <category><![CDATA[alerting]]></category>
            <category><![CDATA[slo]]></category>
            <category><![CDATA[error-budget]]></category>
            <category><![CDATA[sli]]></category>
            <dc:creator><![CDATA[Saverio Ferrara]]></dc:creator>
            <pubDate>Thu, 02 Feb 2023 23:00:48 GMT</pubDate>
            <atom:updated>2024-08-22T10:59:37.396Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*r823nIWT6JuBODvzffge0Q.jpeg" /><figcaption>Photo by <a href="https://unsplash.com/ko/@introspectivedsgn?utm_source=unsplash&amp;utm_medium=referral&amp;utm_content=creditCopyText">Erik Mclean</a> on <a href="https://unsplash.com/photos/mmPMCNt6cPM?utm_source=unsplash&amp;utm_medium=referral&amp;utm_content=creditCopyText">Unsplash</a></figcaption></figure><p>As a software engineer, if you’re reading this, for sure you already know that the error budget is defined when we select a <em>Service Level Indicator</em> (SLI) and for it set a target <em>Service Level Objective</em> (SLO) over a specific time period. If you don’t, head to the “references” section below which suggests introductory resources.</p><p>How fast can we consume the error budget? This question was not easy to answer for me. That’s why I’d like to share my notes.</p><p>Let’s focus on a specific example so that we’ll be able to quantify the error budget. We’ll consider an availability SLI that, as you surely know, is the proportion of the good events over the total number of events. In addition, we’ll consider:</p><ul><li><strong>SLO target</strong>: 95%</li><li><strong>Budget window</strong>: 720h (i.e., 30 days)</li></ul><p>In case you’re not familiar with those concepts, at the end of this article there is the “terminology” section that can help.</p><p>The error budget is the amount of not “good events” derived from an SLO’s target percentage (100% — target percentage). In our case, the error budget is 100% — 95%. Hence:</p><ul><li><strong>Error budget</strong>: 5%</li></ul><p>Google introduced the concept of <strong>burn rate</strong> which is a number which represents how fast, relative to the SLO, the service consumes the error budget. It is defined as:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/296/0*S4wsqsi2pYNs6QiE.png" /></figure><p>By definition, a burn rate of 1 means that the service is consuming the error budget at a rate that leaves you with exactly 0 budget at the end of the budget window.<br>A burn rate of 2 means that the service is consuming the error budget twice as fast as a burn rate of 1.</p><p>The burn rate can’t be an infinite number. Considering we’re experiencing a complete service outage, then our error rate is 100%. So, the <em>maximum burn rate</em> is:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/252/0*iRfA-O7FqKCyW6hA.png" /></figure><p>In our example the maximum burn rate is 100% divided by 5%, hence:</p><ul><li><strong>Max Burn Rate</strong>: 20</li></ul><p>So, how fast can we burn the error budget? The maximum burn rate tells us that we can consume the error budget up to 20 times faster than normal. This doesn’t answer our question!</p><h3>How big is my error budget?</h3><p>Let’s reason in hours and see if we can make it clearer. The key concept here is that we should consider a “complete outage” of the service.</p><p>An error budget of 5% over a budget window of 720h tells us that we are allowed to have a complete outage of 36h without violating the SLO target. This is the <strong>Window Budget</strong>:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/510/0*rY_6Anrtc-CZ6Yqq.png" /></figure><p>Returning to the initial question… how fast can we burn the error budget? The answer is 36 hours.</p><h3>How much error budget can I consume in one hour?</h3><p>This question is more interesting and allows us to set alerts. Instead of considering the entire budget window of 720 hours, let’s focus on a restricted alert window of 1 hour. In case of a complete outage, the maximum amount of <strong>budget consumed</strong> is:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/462/0*eamVMxJ2Kz_kU6yc.png" /></figure><p>In our case:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/430/0*qaRJaR4RKiKsOc12.png" /></figure><p>Careful here! The <em>maximum budget consumed</em> is 2.77% of the entire error budget. So, in one hour we can consume 2.77% of 5%. Always considering a complete outage.</p><h3>How to set a burn rate alert for a complete outage?</h3><p>Having an alert on the burn rate, means we want to be alerted if our budget is consuming fast over a restricted time period (i.e., the alert window). Monitoring tools allow us to set such kinds of alerts and we should only define how long should be the alert window and what is the burn rate threshold. Like before we’ll suppose 1-hour alert window and the burn is calculated as:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/569/0*JmiMbm8Si7pLnlSc.png" /></figure><p>Let’s do the math:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/361/0*y4-RyUt5FigJNPtV.png" /></figure><p>This is exactly the value we got before for the maximum burn rate. In fact, we’re still considering a complete outage.</p><h3>How to set a burn rate alert for a major outage?</h3><p>Alerting on a complete outage is very extreme and, in some cases, may lead to poor alert recall. A more reasonable approach is to be alerted when 80% of the calls are failing. Let’s do this.</p><p>If in case of a complete outage, we can burn 2.77% of the budget, then we should decrease it by 20% if 80% of the calls are failing. The resulting budget consumed in 1 hour is then ~2.22%. Let’s apply the burn rate formula again, and we’ll have:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/359/0*D0a9-gS6a7JmFQv7.png" /></figure><p>Answer: having an alert with a burn rate of 16 will warn us in case we’ll experience a drop of 80% of the good events during the 1-hour period.</p><h3>Terminology</h3><p><strong>SLI</strong> (%): Service Level Indicator. A quantifiable measure of service reliability that tells you if things are working.</p><p><strong>SLO</strong> (%): Service Level Objective. A reliability target for an SLI that tells you if the users are happy or sad.</p><p><strong>SLA</strong> (%): Service Level Agreement. A contract that the service provider promises users on service reliability. Agreed reliability targets should be more relaxed than the internal SLOs to give enough room for unexpected outages.</p><p><strong>Target</strong> (%): SLO Target. It is the availability target (%) defined by the SLO.</p><p><strong>Budget</strong> (%): Error Budget. Either the number of requests out of the total requests or the amount of time for a given period, a system can afford to be unreliable before users become unhappy.</p><p><strong>Burn Rate</strong> (number): It is an indicator that defines how fast, relative to the SLO, the service consumes the error budget.</p><p><strong>Budget Window</strong> (h): The length of the error budget window. We recommend 30 days (720 hours). It usually is 7, 30, or 90 days.</p><p><strong>Alert Window</strong> (h): It is a time window (in hours) in which we’d calculate the burn rate and notify the team if the budget is burning too fast. For example 1 hour.</p><p><strong>Budget Remaining</strong> (%): Error Budget Remaining. The remaining error budget within the Budget Window</p><p><strong>Budget Consumed</strong> (%): It is the percentage of Error Budget consumed within a time period. For alerts, it refers to the Alert Window.</p><p><strong>Current Status</strong> (%): It is the SLI value over the budget window.</p><p><strong>Budget Consumption Time</strong> (h): Time (hours) until the error budget is entirely consumed within the time window.</p><p><strong>Window Budget</strong> (h): This is the total amount of time we can spend with a complete outage that will consume the entire Budget.</p><h3>References</h3><ul><li><a href="https://sre.google/workbook/implementing-slos/">Google SRE: implementing SLOs</a></li><li><a href="https://sre.google/workbook/alerting-on-slos/">Google SRE: alerting on SLOs</a></li><li><a href="https://medium.com/expedia-group-tech/error-budget-policy-adoption-at-expedia-group-7d80d41c4a8b">Error budget policy at Expedia Group</a></li><li><a href="https://medium.com/expedia-group-tech/error-budget-policies-in-practice-4c98f56a28c1">Error budget policies in practices</a></li></ul><h3>Appendix: Error Budget Formulas</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/653/0*uIo8REVcg0AKrWY7.png" /></figure><p><em>Cheat-sheet is available at </em><a href="https://fsferrara.github.io/"><em>https://fsferrara.github.io</em></a><em>.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=3e999f8a200d" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Working Together in a Global Company]]></title>
            <link>https://medium.com/expedia-group-tech/working-together-in-a-global-company-d242cee6f440?source=rss-bf68b929425f------2</link>
            <guid isPermaLink="false">https://medium.com/p/d242cee6f440</guid>
            <category><![CDATA[distributed-teams]]></category>
            <category><![CDATA[collaboration]]></category>
            <category><![CDATA[remote-working]]></category>
            <category><![CDATA[global-workforce]]></category>
            <category><![CDATA[software-engineering]]></category>
            <dc:creator><![CDATA[Saverio Ferrara]]></dc:creator>
            <pubDate>Tue, 10 Mar 2020 14:01:01 GMT</pubDate>
            <atom:updated>2024-08-19T17:03:59.784Z</atom:updated>
            <content:encoded><![CDATA[<h4><a href="https://medium.com/expedia-group-tech/software/home">EXPEDIA GROUP TECHNOLOGY — SOFTWARE</a></h4><h4>Seven tips for collaborating across geographies</h4><p>I still remember when I joined the <a href="http://hotels.com/">Hotels.com</a>™ (part of Expedia Group™) technology team and how excited I was to start such a new challenge. Since the very early days, I noticed that working in a global company is quite different from working in the same building. That is obvious, I know, but it wasn’t easy to get used to the new way of working.</p><p>I missed getting in touch with new teammates, and sometimes felt isolated from others and had the feeling of being a <em>remote worker</em>. Later on, only when I started working from home for 2 days per week, I realized that working in a <em>global team</em> has a lot in common with the <em>remote work</em>.</p><p>In this article, I will list <strong>my personal seven rules</strong> that allowed me to collaborate effectively with colleagues scattered across the world.</p><figure><img alt="Photo of a person holding a globe facing a sunset" src="https://cdn-images-1.medium.com/max/1024/1*ioltQQBG3UKp8bDQIjmfmg.jpeg" /><figcaption>Photo by <a href="https://unsplash.com/@benwhitephotography?utm_source=unsplash&amp;utm_medium=referral&amp;utm_content=creditCopyText">Ben White</a> on <a href="https://unsplash.com/s/photos/global?utm_source=unsplash&amp;utm_medium=referral&amp;utm_content=creditCopyText">Unsplash</a></figcaption></figure><h3>1. Know every collaboration tool</h3><p>Get familiar with the standard company collaboration tools for <strong>chat</strong>, <strong>wiki</strong>, <strong>ticketing</strong>, and <strong>video-conferencing</strong>. They are now my best friends in day-to-day work. This <em>virtual technology</em> is key to working together from different locations.</p><p>But there’s more! I like to create <em>virtual whiteboards</em> using Google Drawings, <em>share files</em> with Dropbox, and collaborate on <em>documents</em> with any modern documents suite. Based on the team’s needs, it is important to pick the right tools to work together.</p><p>As a good practice, I also have backup tools for chat and video-conferencing — in case one of them doesn’t work for some reason, I don’t waste time, I just move to the next one.</p><figure><img alt="Decorative separator" src="https://cdn-images-1.medium.com/max/1024/1*tFEnh5nNCzSTWKXdMaaImA.png" /></figure><h3>2. Communicate asynchronously</h3><p>It is not recommended to chase people and ping them in person. Having a quick 1:1 chat in the office would exclude those who are working from home or from another location. In a global environment, the most effective way to communicate is asynchronous.</p><p>But please, stop writing tons of email! Start using issue mentions and chat tools, since Chat and Ticketing tools are way better! Also, tickets are public and can be read from the entire team and whoever is interested. For this reason, it is extremely important to properly use those tools even if the message you are communicating is meant for someone that is sitting next to you.</p><p>Communicating asynchronously is not always the best choice. For example, if I need a long discussion, I schedule a call. And then I use a headset because it highly improves the audio quality for all the participants.</p><figure><img alt="Decorative separator" src="https://cdn-images-1.medium.com/max/1024/1*tFEnh5nNCzSTWKXdMaaImA.png" /></figure><h3>3. Communicate your status</h3><p>Keep in mind that people can’t see you.</p><ul><li>If you are working from home, folks who are in the office can’t see you.</li><li>If you are working in the office, remote workers can’t see you.</li><li>In both cases, people that are in a different office location can’t see you.</li></ul><p>That’s why I start my day by writing this in my team’s chat channel:</p><figure><img alt="A chat message reading “Good morning” followed by an icon of the sun" src="https://cdn-images-1.medium.com/max/1000/1*Q-HsszJXp41M7t1jYzCdww.png" /></figure><p>I write that message independently if I am in the office or working from home. In this way, the entire team knows the time when I have started to work.</p><p>Another similar example is the following:</p><figure><img alt="A chat message asking a technical question, with a response that reads “Let me have a look”" src="https://cdn-images-1.medium.com/max/1000/1*bCxoE0M52J9fDiuqWf-NEA.png" /></figure><p>Giving an acknowledgement that you read the message and you are working on it has two advantages:</p><ol><li>The requester will know the message was received,</li><li>And the team will know that someone is handling the request.</li></ol><p>In general, make it clear to everyone what you are working on. Do it using the tool that is most appropriate for your team.</p><p>Be 100% available in chat during working hours. This is usually the main communication channel and if I am not available in chat then people just assume I am probably out of office. Some chat tools offer the possibility to set the current status so that everyone can know what you are doing. Examples of status are: “Having lunch”, “In a meeting”, “Out of office”, and “Vacationing”.</p><figure><img alt="Decorative separator" src="https://cdn-images-1.medium.com/max/1024/1*tFEnh5nNCzSTWKXdMaaImA.png" /></figure><h3>4. Plan the future</h3><p>When collaborating across locations, it is important to plan the next actions.</p><p>The team’s daily stand-up is an example. During such a meeting, do not focus too much on the past (i.e., what you have done the previous day), but use the meetings to plan the future.</p><p>Sharing a plan will increase the chances of working together even if someone is not working in the same place.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*tFEnh5nNCzSTWKXdMaaImA.png" /></figure><h3>5. Do be seen</h3><p>Do you normally cover your face in the office?</p><figure><img alt="Tattooed woman covering her face with her hands" src="https://cdn-images-1.medium.com/max/1024/1*j7kt4mdljbHGIXil_H5oBA.jpeg" /><figcaption>Photo by <a href="https://unsplash.com/@anniespratt?utm_source=unsplash&amp;utm_medium=referral&amp;utm_content=creditCopyText">Annie Spratt</a> on <a href="https://unsplash.com/s/photos/hide-face?utm_source=unsplash&amp;utm_medium=referral&amp;utm_content=creditCopyText">Unsplash</a></figcaption></figure><p>Then why hide your face during an online meeting? Nonverbal communication plays a big role, so don’t be shy. Proudly show your face during online meetings.</p><p>Besides the meetings, many tools allow you to set a profile picture. Of course, I recommend updating your profile picture with a recent photo of you. This facilitates the association between faces and names.</p><p>I haven’t tried any always-on video portals like <a href="https://perch.co/">Perch</a>, but I would like to have one of those portals in the relaxation area to enjoy a virtual coffee break with remote colleagues.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*tFEnh5nNCzSTWKXdMaaImA.png" /></figure><h3>6. Meet in person</h3><p>A global environment has its drawbacks. Working remotely or from a small office can give you the sensation of loneliness or isolation. Although virtual meetings help to mitigate this aspect, the real cure is to meet in person regularly.</p><p>My suggestion is to take advantage of any team-building opportunities. For example, during business trips, try to spend time with others and have fun together.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*tFEnh5nNCzSTWKXdMaaImA.png" /></figure><h3>7. Say thank you</h3><p>If someone helps me in some way, then I make sure to highlight it and give him/her credit. It always feels good to receive thanks, and this will make future collaborations easier.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*AZpg_TdELqc_lgxbZedekg.png" /></figure><p><strong>If you enjoyed this post, you might want to check out the </strong><a href="https://fsferrara.github.io/"><strong>slide deck</strong></a><strong> for your team.</strong></p><blockquote><a href="https://lifeatexpediagroup.com/brands">https://lifeatexpediagroup.com/brands</a></blockquote><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=d242cee6f440" width="1" height="1" alt=""><hr><p><a href="https://medium.com/expedia-group-tech/working-together-in-a-global-company-d242cee6f440">Working Together in a Global Company</a> was originally published in <a href="https://medium.com/expedia-group-tech">Expedia Group Tech Blog</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
    </channel>
</rss>