<?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 Nadger on Medium]]></title>
        <description><![CDATA[Stories by Nadger on Medium]]></description>
        <link>https://medium.com/@nadger?source=rss-8b43adae74a8------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*YNCKnyBgC1RL0DKpCkYuAA.png</url>
            <title>Stories by Nadger on Medium</title>
            <link>https://medium.com/@nadger?source=rss-8b43adae74a8------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Fri, 20 Nov 2026 23:46:34 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@nadger/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[Is Software About to Have Its Bank Teller Moment?]]></title>
            <link>https://medium.com/@nadger/is-software-about-to-have-its-bank-teller-moment-c41dcf856a73?source=rss-8b43adae74a8------2</link>
            <guid isPermaLink="false">https://medium.com/p/c41dcf856a73</guid>
            <category><![CDATA[software-engineering]]></category>
            <category><![CDATA[productivity]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[programming]]></category>
            <category><![CDATA[software-development]]></category>
            <dc:creator><![CDATA[Nadger]]></dc:creator>
            <pubDate>Wed, 21 Oct 2026 11:45:54 GMT</pubDate>
            <atom:updated>2026-10-21T11:45:54.657Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*LKhAahIg57sNsrr4d_QdHQ.png" /></figure><p><em>Every mass-employment industry eventually learns the same lesson: automate the repeatable part of the work, and you end up with far fewer people doing it, far better. Car plants did. Shipyards did. Steel mills did.</em></p><p><em>Software never has. For seventy years, the industry has only known one direction: more code needed, more hands to write it. Recession-proof, headcount-hungry, always hiring.</em></p><p>Here’s my postulation: that’s about to change. Software is entering the same contraction these other industries went through, a much smaller, far more capable workforce replacing a much larger one that did more repeatable work. If that’s right, getting into software stops looking like “<em>learn to code and get a job</em>” and starts looking like getting into medicine, law or accountancy: good grades, demonstrated aptitude, years of supervised training before anyone trusts you with real responsibility.</p><p>I don’t think the right comparison for how this happens is car plants, though.</p><h4>Not machines replacing muscle. AI replacing judgment-adjacent execution.</h4><p>Car-making and shipbuilding are stories about machines replacing muscle. Software is a knowledge trade. The sharper precedents are other knowledge trades that lost a single <strong>cognitive</strong> task embedded inside a bigger profession, not a whole physical production line.</p><p>Telephone operators. Between 1920 and 1940, AT&amp;T engineered the switchboard operator’s job out of existence across most of the US network. A linked-census study found the operators who lost that job didn’t move up with everyone else; a decade later they were disproportionately in lower-paying work, or out of the labour force entirely.</p><p>Hot-metal compositors. Setting type for print was a decades-to-master craft until desktop publishing collapsed it within a single generation. US printing employment is down roughly 60% from its peak. What survived, in far fewer hands, was editorial and design judgment, exercised directly through the new tools.</p><p>NYSE floor traders. The trading floor shrank from about 5,500 people at its peak to a few hundred today. What vanished was manual order-matching. What grew, in both stature and pay, was quantitative strategy and risk judgment, with entry now resembling medicine or law: specialist degrees, credentials, brutal selectivity.</p><p>Travel agents. US travel agency employment tripled to 340,000 by 2000, then fell 60–70% once Expedia and Priceline arrived. What survived serves complex, high-value bookings. The commodity task, finding a fare, went self-service.</p><p>Four cases, one shape: a discrete, mechanical task nested inside a broader profession got automated out, and the people who remained moved up into judgment. That’s a much closer match for “AI now writes the code” than anything from a car plant.</p><h4>The counter-example that should worry you if you believe this too quickly</h4><p>Bank tellers were supposed to go the same way when ATMs arrived. They didn’t. US teller employment stayed roughly flat from 1980 to 2010, and briefly <strong>doubled</strong>. Why? ATMs made branches cheaper to run, so banks opened more branches, and total demand for tellers rose even as tellers per branch fell. Automation lowering the cost of an activity can expand its own market for decades before a second, unrelated shift (online banking) finally finishes the job.</p><p>That’s Jevons’ Paradox, and it has an unbroken track record in software specifically. Compilers, fourth-generation languages, CASE tools, low-code platforms: every one of them was supposed to shrink the need for programmers. Every one of them was followed by <strong>more</strong> programmer demand, not less, because cheaper production unlocked software that wasn’t economical to build before. BLS still projects software developer employment up 10% through 2035.</p><p>So which is it: compositor, or bank teller?</p><h4>The evidence that it’s already happening</h4><p>One thing is not really in dispute: the narrow “programmer” role, writing code, has been shrinking for twenty years, long before generative AI existed, driven originally by offshoring. Programmer employment is at its lowest level since 1980. Meanwhile “software developer,” the broader design-and-systems role, keeps growing, and so does systems architecture. The craft is moving up the stack. That part looks solid. The part in dispute is whether total software headcount shrinks, or the mix just changes while the industry keeps growing.</p><p>Here’s what tips me toward shrink. The biggest, best-informed employers in the industry are currently answering this question in public, with money. US tech firms cut roughly 140,000–150,000 jobs in the first half of 2026 alone, more than the equivalent period in 2025, <strong>while</strong> posting record revenue and committing some $725B in AI infrastructure spending between just four companies. This is not distressed downsizing. It’s profitable companies explicitly betting they can grow output and shrink the team that builds it, simultaneously. Salesforce’s CEO says the company hired no new engineers at all in fiscal 2026. Block’s Jack Dorsey, cutting close to half his workforce, put the belief as plainly as it gets: “a significantly smaller team… can do more and do it better.”</p><p>That belief isn’t unanimous. Palo Alto Networks’ CEO has said the opposite on the record: “the fallacy is that organizations are going to get 30, 40, 50, 60% more productive… so we need less people. No, I need more.” A recent survey of tech leaders found most are using AI’s productivity gains to build <strong>more</strong>, not to cut heads. So the largest, most visible firms may be leading a real shift, or they may just be the loudest early movers in what’s still a minority position.</p><p>And at the small end of the market, there’s a genuinely interesting complication: new business formation is at record highs, not falling, and a handful of AI-native software companies are posting startling revenue-per-employee numbers (Cursor-maker Anysphere at roughly $300M ARR with a few dozen people; Lovable at $400M ARR with 146 staff). That’s either early evidence of contraction arriving at small companies first, or evidence that lean firms simply multiply and eventually hire at scale, just later and leaner than the last generation. Too early to call.</p><h4>Why this isn’t medicine, law or accountancy. Yet.</h4><p>If software really is contracting into a smaller, more capable, more revered profession, the obvious comparison is to medicine, law and accountancy: good grades, demonstrated aptitude, years of supervised training before anyone trusts you with real responsibility.</p><p>Software already tried this, and the market rejected it. A national Software Engineering PE exam ran from 2013 to 2019 and drew a grand total of 81 candidates. It died from lack of demand.</p><p>That failure makes sense once you see why medicine, law and accountancy got their protections in the first place. None of them professionalised by choice. Each was forced into it by a visible, costly failure of trust: the 1910 Flexner Report exposing medical schools as diploma mills; the 1929 crash and the Securities Acts that made audited financial statements a legal requirement. In each case, an information gap between practitioner and client, combined with a public disaster, made informal market filtering too slow and unreliable.</p><p>Software never had that moment. Failures were diffuse, blamed on a vendor rather than “the profession.” Employers could filter candidates faster than any national board. That may be exactly what’s now changing: code is increasingly generated at a volume no single human fully traces, software sits inside life-critical infrastructure far more than it did in 2013, and a mature body of engineering knowledge (IEEE’s SWEBOK, now in its fourth edition) exists to credential against for the first time. If a real AI-generated-software trust failure happens publicly, which is a when more than an if, the gatekeeping that follows probably won’t look like a PE stamp again. More likely: mandatory senior sign-off for safety-critical software, liability regimes that price credentialed teams differently, or a voluntary certification that spreads because it’s cheaper than vetting from scratch, the way the CPA designation actually did.</p><h4>The part nobody’s industry has solved</h4><p>Here’s what worries me more than any of the employment numbers. Medicine, law and accountancy sustain a small, elite workforce because they built deliberate, multi-decade systems, residencies, articling, audit years, specifically to keep producing seniors. That machinery is the whole point. Without it, “revered and curated” collapses in one generation.</p><p>Software is cutting the bottom of the pyramid right now, without having built anything like that. I know of at least one employer with over 2,000 staff, roughly half of them technical, and turnover above $1.3B, that runs no graduate scheme, no internship programme, no junior-entry route of any kind. The stated logic: hire seniors, get revenue traction faster. By report, this isn’t an outlier position among comparable companies; it’s a shared industry stance. I’m using this anonymously on purpose, because it isn’t a story about one company. It’s a live instance of exactly the pattern this whole piece is about.</p><p>One company’s hiring policy doesn’t prove an industry-wide pipeline collapse; juniors might still be getting made elsewhere, by smaller firms, bootcamps, career-changers. But if “hire seniors, skip juniors” is now the common position among employers of this size, the people this document imagines as revered and curated in 2040 have no mechanism to be grown. The professions we’re comparing software to solved this problem deliberately, decades ago. Software hasn’t started.</p><h4>Where I land</h4><p>I think the postulation holds up better than I expected going in, but not uncomplicated. The knowledge-trade precedents (compositors, floor traders, telephone operators) are a much better fit for what’s happening to coding than manufacturing ever was. The biggest employers in the industry are already betting money on contraction, not just efficiency. And the junior pipeline problem is real, present-tense, and currently unaddressed by anyone.</p><p>But Jevons’ Paradox has beaten this exact prediction every single time it’s been made about software before, the bank-teller case shows the same cost mechanism can expand a market for decades before it shrinks it, and the lean-startup data is genuinely ambiguous about which way this goes.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=c41dcf856a73" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Lightning Development in the AI Age]]></title>
            <link>https://medium.com/@nadger/lightning-development-in-the-ai-age-4da67d256098?source=rss-8b43adae74a8------2</link>
            <guid isPermaLink="false">https://medium.com/p/4da67d256098</guid>
            <category><![CDATA[lightning]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[productivity]]></category>
            <category><![CDATA[software-development]]></category>
            <category><![CDATA[programming]]></category>
            <dc:creator><![CDATA[Nadger]]></dc:creator>
            <pubDate>Sun, 11 Oct 2026 11:58:55 GMT</pubDate>
            <atom:updated>2026-10-21T07:50:25.028Z</atom:updated>
            <content:encoded><![CDATA[<h3>When AI can generate the specification, the architecture and the code, the two-week sprint is the slowest thing left in the building. Lightning Development is a way of working that takes a product idea to a working Alpha in a couple of days, with three disciplines in one room and one pen between them.</h3><h3>One triangle, one pen</h3><p>Picture a triangle pivoted around its centre. Product sits at the top, Architecture at the bottom left, Engineering at the bottom right. Whoever is at the top holds the pen.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/800/1*gfW6bSiSnrbcU6FhKONf0g.gif" /></figure><p>All three disciplines are in the room for the whole journey. Only one of them is writing at any moment. The other two are not waiting for a handover: they are challenging, correcting and supplying what the pen-holder does not know.</p><p>The pen, in every case, is an AI. The human holding it is not typing the specification, the decision records or the code. They are directing the generation, judging the output and deciding when it is good enough to turn the triangle.</p><p>Three turns take an idea to a running Alpha. Each turn ends with an artefact, and no turn is ever fully closed.</p><h3>Turn one: Product holds the pen</h3><p>Product starts. They discover, gather requirements and, with Architecture and Engineering beside them, generate a specification.</p><p>This is not a feature list. A list of features tells an AI what to build and nothing about what the result has to be like. The specification has to carry the characteristics too:</p><ul><li><strong>Features:</strong> what the product, service or feature does, and for whom.</li><li><strong>Running cost target:</strong> what it may cost to operate, per month or per transaction.</li><li><strong>Throughput:</strong> the transactions per second it has to sustain.</li><li><strong>Cloud constraints:</strong> providers, regions, managed services that are in or out of bounds.</li><li><strong>Reference material:</strong> the house rules the output must follow, such as RESTful API guidelines, security standards and naming conventions.</li></ul><p>Architecture and Engineering earn their seats here. They are the ones who know that a cost target and a TPS target are in tension, or that a constraint quietly rules out the obvious design. Catching that in the first hour is cheap. Catching it in the code is not.</p><p>When everyone is happy, the triangle turns.</p><h3>Turn two: Architecture holds the pen</h3><p>Architecture is now at the top, and the output is a set of Architecture Decision Records. Each ADR captures one decision: the context, the options considered, the choice made and what it costs.</p><p>Product and Engineering stay closely involved. Product answers the questions the specification left open. Engineering says what will be painful to build, test and run.</p><p>This turn may well change the specification. That is the process working, not failing. A requirement that looked harmless on paper turns out to double the running cost, and Product, who is sitting right there, decides in minutes whether it stays.</p><h3>Turn three: Engineering holds the pen</h3><p>The final turn puts Engineering at the top with the code-generation pen. They create the service or application, and that means all of it: the code, the tests, the QA, the CI/CD pipeline, all the way to a deployed Alpha.</p><p>The other two are still in the room. Generated code is the hardest test a specification or an ADR will ever face, so both may be modified again. A decision that read well as prose can look very different once it is running.</p><p>Code generation is the one stage where the room can sometimes thin out while the AI works. That is the exception, not the rule.</p><h3>Nobody leaves the room</h3><p>The whole process runs in real time. Everyone is present at each stage, and nobody leaves the “room” until the artefact for that stage is complete, even if it is not final.</p><p>Complete and final are different things. A complete specification is one you can build from today. It will still change tomorrow, when the ADRs or the code push back on it.</p><p>The room is small: ideally no more than three or four people, totally focussed on this one task for the duration. It may well be that all three roles are one person. The triangle still turns, because the three ways of thinking are still distinct, and it matters which one you are doing.</p><p>What the room removes is the queue. In a conventional lifecycle a question from Engineering to Product waits for a meeting, a ticket or a refinement session. Here it is answered before the AI has finished generating the next file.</p><h3>Bolts, not sprints</h3><p>The duration is the other main facet. This is not a two-week sprint. It is a couple of days at most, broken into a series of <strong>bolts</strong>: very short sessions, measured in hours, each of which spawns an artefact at the end.</p><p>Planning shrinks to match. There is no backlog of epics and stories to groom, because nothing lives long enough to need grooming. You state the outcome you want and go and get it.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/465/1*oT7lTZLnsP87jGQEg0XVRw.png" /></figure><p>The last row matters most. Story points measure effort, and effort is the thing AI has just made cheap. Outcomes are what is left to count.</p><h3>Individuals are empowered. Teams are de-powered.</h3><p>This is the uncomfortable truth about AI-accelerated product development. The individual has been massively empowered, and the team has lost much of its advantage.</p><p>An individual who owns the code base, holds the vision and understands the business case is far more effective than a team sharing the code generation. Everything the AI needs to be steered well is in one head, and nothing is lost in translation between heads.</p><p>Share the generation across a team and the old problems return at AI speed. Several people generating simultaneously against the same code base produce pull requests faster than anyone can review them, and merge conflicts between changes that no human wrote and no human fully remembers.</p><p>Small teams work better with AI for a plain reason: they cut out the decision-making ceremonies and drive directly to code generation. Every person added to the room adds a conversation that has to happen before the pen can move.</p><h3>Capability beats headcount</h3><p>None of this works without skilled and talented humans. The multiplier is the human, not the AI.</p><p>Give the same model to two people and you will get two very different products. What makes the difference is not the AI alone. It is three things the human brings:</p><ul><li><strong>Knowing what good looks like.</strong> You cannot steer towards a standard you have never seen.</li><li><strong>Writing specifications precise enough to build from.</strong> Vague in, plausible-looking rubbish out.</li><li><strong>Catching what the AI got wrong.</strong> It will be wrong, fluently and confidently, and someone has to notice.</li></ul><p>That changes how an organisation should think about people.</p><blockquote><em>Every resourcing decision we make, hiring, allocation, team shape, should reason about capability, not head count.</em></blockquote><p>Adding heads to a Lightning team slows it down. Adding capability speeds it up. And capability cannot be bought in a hurry:<strong> the people who multiply output take years to grow. Protect them.</strong></p><h3>Spec, ADRs and code: one picture, one repo</h3><p>After a few iterations the balance between the artefacts shifts. The code becomes the most precise statement of what the product does, and the specification settles into a different role: it becomes the story.</p><p>That is a change of job, not a retirement. The specification no longer dictates every detail, so it diminishes in importance, but it still says what the product is for, who it serves and the targets it has to hit. No one can recover that from reading source files.</p><p>The ADRs fill the other gap. Why this database, why this boundary between services, why the cheaper option was rejected. None of that is in the code base either.</p><p>So the full picture is all three: the specification is the story, the ADRs are the why, and the code is the what. Lose any one and the other two stop making sense.</p><p>That means they have to be kept in sync, and not only on the way to Alpha. It applies throughout, including post-deployment maintenance. A bug fix that changes behaviour changes the specification. A scaling fix that changes a design choice needs a new ADR.</p><p>Humans are bad at this kind of housekeeping, which is why documentation has always drifted. The answer is to put all three in the same location, the same repo, so that the AI manages the synchronisation. A change to any one of them is a change the AI can see, and it can carry it through to the other two in the same commit.</p><h3>The craft has moved up the stack</h3><p>The craft of software development has not disappeared. It has shifted up the stack, to system design, architecture and operations.</p><p>The code only matters in the abstract. Does it have the right shape, the right boundaries, the right behaviour under load and under failure? Those are the questions a skilled human now asks. The choice of loop or variable name is no longer where the skill shows.</p><p>This is also why the triangle gives Architecture a turn of its own instead of folding it into Engineering. The decisions made in that turn are the ones that outlive every regeneration of the code.</p><h3>“But surely…”</h3><p>Three objections come up every time this approach is described. Each has an answer, and each answer is already built into the process.</p><h3>“Too fast to be safe”</h3><p>It is also traceable. Every feature leads back to a line in the specification, and every structural choice leads back to an ADR that records what was decided and why. Many slower projects cannot say as much.</p><h3>“Bus factor of one”</h3><p>A single owner sounds like a single point of failure. It would be, if the knowledge lived only in that person’s head.</p><p>It does not. The specification, the ADRs and the code are the handover. They were produced as the work happened, not written up afterwards from memory, and together they say what the product is for, why it is built this way and exactly what it does. A capable successor with an AI can read all three and be productive in hours.</p><h3>“What about juniors?”</h3><p>If small teams of experienced people do the work, where do the next experienced people come from?</p><p>They are grown through review. The skills that matter now are knowing what good looks like and catching what the AI got wrong, and both are learned by reviewing: reading specifications, challenging ADRs, checking generated code against intent, with a senior explaining what they would have caught. A junior in the room for a bolt sees more decisions made, and the reasons for them, than in months of working tickets.</p><h3>Get to Alpha, and fail there if you must</h3><p>The overall goal is to reach Alpha fast. We want to fail the business case or use case as soon as possible, with minimal effort.</p><p>A working Alpha in front of real users answers questions that no amount of analysis will. If the idea is wrong, you have lost a couple of days and a handful of bolts. If it is right, you already have the code, the decisions and the specification to build on.</p><p>So proofs of concept are kept to a minimum. A PoC is a side-hustle: a short effort to tease out a couple of ADRs. It is not a wholesale technology deep dive or a massive, in-depth market analysis. The Alpha is the proof of concept that counts.</p><h4><strong>Lightning Development is, in the end, a bet on people. Put a few highly capable ones in a room, give them an AI and a clear outcome, turn the triangle three times, and do not let them leave until something real exists.</strong></h4><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=4da67d256098" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[“Read all about it!” — Newspaper style software delivery]]></title>
            <link>https://medium.com/@nadger/read-all-about-it-newspaper-style-software-delivery-912579861634?source=rss-8b43adae74a8------2</link>
            <guid isPermaLink="false">https://medium.com/p/912579861634</guid>
            <category><![CDATA[continuous-deployment]]></category>
            <category><![CDATA[continuous-delivery]]></category>
            <category><![CDATA[software-development]]></category>
            <category><![CDATA[continuous-integration]]></category>
            <dc:creator><![CDATA[Nadger]]></dc:creator>
            <pubDate>Thu, 29 Feb 2024 17:05:09 GMT</pubDate>
            <atom:updated>2024-02-29T17:05:09.630Z</atom:updated>
            <content:encoded><![CDATA[<h3>“Read all about it!” — Newspaper style software delivery</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/900/0*olqSjAfv93Czfvja.jpg" /></figure><p><em>Before I start, let’s get something straight; I’m not suggesting this is the </em><strong><em>only</em></strong><em> way or even the </em><strong><em>best</em></strong><em> way, it’s just </em><strong><em>a </em></strong><em>way that suited us and it worked in this case.</em></p><h3>The Story</h3><p>Picture the scene; a 20 year old, browser based product that looks dated, behaves poorly and is really damaging our brand value. It has been tweaked, features shoehorned in and generally messed about with over the course of its lifetime so now it’s a hotchpotch of PHP, Ruby, Java, C, Comet, Websockets, JQuery, Prototype, Isomorphic, Smarty, React. Angular, blah blah.</p><p>We need to rewrite the whole thing as a modern Single Page Application (SPA), a federated experience using React, a Backend for Frontend (BFF) service and our own company design library. On top of that, our new CEO decided to publicly announce that we were undertaking this, that our customers would be wowed by the new experience and it would gain a great independent Gartner review in time for the next publication in 9 months time!</p><p>A wonderful collection of stress monkey contributors; <strong><em>CEO visibility, public expectation and an immovable deadline</em></strong>, great.</p><h3>The Challenge</h3><p>In truth, we weren’t really starting from scratch, we had been doing some small scale work on componentising the UI into React, we also had a burgeoning BFF and design library. We had also pioneered a novel microfrontend delivery approach. Effectively we had proven the techniques we wanted to use, the methodology and we had a good idea of the challenges.</p><p>We also had great cohesion between Product, UX, Developers, QE and DevOps, all marshalled by excellent Programme Management.</p><p>The major challenge was that with such a large project, we didn’t want to “go dark” for nine months and turn up with a shiny new product — projects like that don’t have a great history of success and go against the whole ethos of “fail fast”. We wanted to deliver demonstrable and tangible incremental progress towards the final objective, keep everyone onside, focused, believing and at the same time maintaining morale and team spirit when there was inordinate amounts of pressure.</p><h3>The Approach</h3><p>I worked in the newspaper industry many years ago and back then (<em>probably still similar today for print journalism</em>) a day in the life of a daily newspaper publisher followed a strict and prosaic pattern.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/620/1*yMIE9v5pRCB1icD3ArvzVA.png" /></figure><p>This flow has some particular characteristics;</p><ol><li>Fixed schedule</li><li>Every day starts at zero</li><li>Every day has different challenges</li><li>A single objective — get the paper out!</li><li>Deadlines, deadlines, deadlines</li></ol><p>The point I’m making is that there is an immovable daily deadline. Subjects need to be chosen, stories written, edited, proofread, laid out on pages, typeset, presses run and newspapers delivered to the distributors. All in time to hit the streets for people to get when they wake up each morning.</p><p>This struck me as being a great process for us to adopt in producing our new UI at pace. Start the day with an agreement of what is going out today, work on the items and publish to production at the end of the day. You could think of it as very lightweight “daily sprints”. This is what it looked like;</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/620/1*qvTk_qcbbZ1zzcOyXexG-A.png" /></figure><p>This is a “<em>little but often</em>” practise overlaid on our Agile (Scrum of Scrums) ceremonies/processes and it requires a number of things to be in place for it to function;</p><p><strong>Buy In</strong> — everyone needs to be onboard with the methodology, understand the urgency each day and where in the process they are. Introducing something radical like this attracts a lot of nervousness and it’s imperative that concerns are heard and accounted for.</p><p><strong>QE/CI/CD </strong>— solid automation that incorporates testing, verification, zero-downtime delivery to production and extremely efficient methods of rolling forwards and backwards. DORA metrics, Defect Reports and measures of quality (e.g. QE cycles for tickets) help drive improvement.</p><p><strong>Customer Validation/Feedback</strong> — allow the customers to see and participate in validating the new version of the product by providing a means by which they can switch between Classic and New. This feeds into the programme metrics to show the effectiveness of the process and drive remediation.</p><p><strong>“Fix First Fast” Mentality</strong> — rather than the first thought always being <strong>rollback!</strong> whenever there is turbulence, always be thinking roll-forward. The cadence of releases reduces risk (small changes), holding back releases or rolling back actually raises the risk level due to the increased delta.</p><p><strong>Keep Everyone Informed </strong>— automated but curated daily report of what was released and by whom. Builds confidence that progress is on-track and if not, a clue to the required course correction.</p><blockquote>That last bit cannot be overemphasised — we actually turned this into an electronic Daily Newspaper that we took turns editing. It contained interesting narrative about the teams activities, funnies, personal observations, cooking tips, shout-outs etc. but it also contained a list of the released components and links to the associated Jira tickets, repos and developers that made the commits.</blockquote><blockquote>The links to the developers included their image from GitHub. This aspect was one of the best morale boosters we could have come up with. To see your name and face in print (all be it electronic), distributed to the whole company, created a real buzz around the project, built pride and provided much needed acknowledgement for the people in the engine room</blockquote><h3>Outcome</h3><p>Well in short, it worked!</p><p>We met the deadline, produced a best in class, distinctive UI that embodied modernity in look &amp; feel, quality and delivery and won some awards for the final product and the processes involved in making it.</p><p>Perhaps even more importantly, we put smiles on the faces of our customers and end-users by improving their efficiency and removing some of the grit in the vaseline of their daily grind.</p><h3>Conclusions</h3><p>Over 50 engineering (SET, FE, BE, DevOps), Product and UX staff were continuously involved in this endeavour and just like any good process, we evolved it as our confidence grew.</p><p>We eventually dropped the morning triage meeting —<strong> if tickets are complete, the CI/CD pipeline and Smoke tests pass, then we ship them.</strong></p><p>Perhaps the biggest cultural change was that we changed our view of the final release sign-off. <strong>We stopped doing it.</strong></p><p>Instead of there being a big “<strong>Green</strong>” button to go, we created a big “<strong>Red</strong>” button to halt. Automation is end-to-end, farm to fork and runs completely unattended. Only if something is flagged from our pipelines, in-place observability, synthetics, customers etc. do we stop the rollout.</p><p>The benefits of this approach are manyfold, but here are a few significant ones;</p><p><strong>Enthused and Incentivised Staff</strong>— the thrill of getting a feature/fix from keyboard to production within a day is infectious and fulfills that deep desire of all creatives to be able to feel valued, acknowledged and appreciated.</p><p><strong>Customer Satisfaction </strong>— turning around bugs in days not weeks/months impresses customers, improves their view of our support organisation and shows that we can be extremely responsive to their issues/requests.</p><p><strong>Invigorated Product/Sales/Marketing </strong>— getting features out quickly and safely adds to their armoury and bolsters the company&#39;s innovation credentials, a great selling feature.</p><blockquote>I hope you get something from this read — the takeaway I would leave with you is this; when faced with this challenge, it was important to show that we’re going to do something different, not the usual grind, liven things up a bit. That raised enthusiasm, commitment and pride in the endeavour.</blockquote><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=912579861634" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>