<?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 Tom Siebeneicher on Medium]]></title>
        <description><![CDATA[Stories by Tom Siebeneicher on Medium]]></description>
        <link>https://medium.com/@tom.siebeneicher?source=rss-27ae99b12291------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*dmbNkD5D-u45r44go_cf0g.png</url>
            <title>Stories by Tom Siebeneicher on Medium</title>
            <link>https://medium.com/@tom.siebeneicher?source=rss-27ae99b12291------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Tue, 17 Nov 2026 18:04:45 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@tom.siebeneicher/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[Nobody Wrote a Strategy Paper for AI]]></title>
            <link>https://medium.com/@tom.siebeneicher/nobody-wrote-a-strategy-paper-for-ai-63d324ee5cde?source=rss-27ae99b12291------2</link>
            <guid isPermaLink="false">https://medium.com/p/63d324ee5cde</guid>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[agile-transformation]]></category>
            <category><![CDATA[technology-trends]]></category>
            <dc:creator><![CDATA[Tom Siebeneicher]]></dc:creator>
            <pubDate>Thu, 23 Jul 2026 14:52:47 GMT</pubDate>
            <atom:updated>2026-07-23T14:52:47.424Z</atom:updated>
            <content:encoded><![CDATA[<p><strong>Every transformation of the last twenty years needed the organisation to distribute it. This one didn’t. That should change how we read the last twenty years.</strong></p><p>The first physical evidence that a large European corporate was going to become agile was a document.</p><p>Not a team, not a tool, not a working prototype. A strategy paper, written by a colleague in the back office, someone who would never have to deliver under the new way of working. It carried a bold, quantified promise: two to three times faster delivery, at the same capacity, with dynamic replanning on a biweekly cadence.</p><p>I was a consultant then, and I tried to find where that number came from. There was no benchmark behind it. No pilot result. As far as anyone could reconstruct, it had been lifted from a slide deck, which had lifted it from an article. An unsourced figure, laundered through two layers of citation, arriving in a strategy paper as a commitment the organisation was now accountable for.</p><p>That is not a story about one badly run programme. It is how enterprise transformation has worked for twenty years: a sponsor becomes convinced, a document appears, a consultancy produces a blueprint, coaches are hired, teams are assessed, and progress is reported upward in the vocabulary of the blueprint rather than the vocabulary of the work. Martin Fowler named the ecosystem that supplies this sequence the Agile Industrial Complex and called its imposition of methods on people a travesty. He co-wrote the Agile Manifesto. When the failure rate of agile transformations is estimated at 47% by Jeff Sutherland, who co-created Scrum, the criticism is no longer coming from outside.</p><p>But the interesting question is not why the machinery failed so often. It is why the machinery existed at all.</p><p><strong>The dependency nobody named</strong></p><p>None of these transformations could be adopted by one person alone at their desk.</p><p>You cannot do Scrum by yourself. You cannot work in a two-week cadence unless the people upstream replan on the same cadence and the people downstream accept the output. You cannot practise continuous delivery if the release board still meets monthly. DevOps without the Ops half is a hobby. Every wave, be it Agile, DevOps or DevSecOps, requires simultaneous behavioural change across a group of people who had not individually chosen it.</p><p>That requirement is what the entire transformation apparatus was built on. If change needs coordinated consent, someone has to manufacture the coordination, and that someone becomes the programme office, the coach cadre, the maturity model, the steering committee. The machinery was not a parasite on the transformation. It was the delivery mechanism, because the transformation had no other way to travel.</p><p>Once you see this, the two classic failure modes stop being mysterious.</p><p>The first: <strong>the instrument measures the input and calls it the change.</strong> On one programme, the number of people trained rose every week. It was reported upward, and it was true. Meanwhile the teams still worked to deadlines rather than cadence, and still waited for management approval rather than deciding themselves, because nothing had changed about who was accountable for a missed deadline or who carried the risk of a wrong call. The old behaviour remained the rational behaviour, and people behaved rationally. Training changed what they knew. It could not change what it made sense for them to do. The dashboard climbed anyway.</p><p>The second: <strong>the organisation performs the adoption.</strong> By the mid-2010s, Agile was the industry standard, so admitting you didn’t know how to do it was a career risk. The ceremonies ran, the boards filled, the maturity scores rose, the name without the practices. A colleague described the underlying posture better than any framework I’ve read:</p><p><em>“Please wash my car, but keep driving.”</em></p><p><strong>What I got wrong</strong></p><p>For a while I believed a good proof of concept would spread on its own merit.</p><p>It was a reasonable belief, and we acted on it. Faced with resistance on that programme, our answer was a small protected pilot, deliberately built to generate enough visible success that others would want in, would manufacture desire, because none existed natively. The pilot worked. Pilots almost always work; a small, loosely coupled bucket can adopt nearly any way of working.</p><p>It went nowhere. Not because the people around it were unconvinced, because many were convinced. But enterprises run on hierarchy, and even highly motivated people follow the system they are actually accountable to. Without management leading by example and endorsing the new way, the merit of the pilot was irrelevant to everyone’s incentives. Good ideas do not travel upward through an organisation that is not asking for them.</p><p>That was my error, and it is worth stating plainly because everything that follows depends on it: <strong>merit alone never moved anything. It always needed permission.</strong></p><p>Until it didn’t.</p><p><strong>The break</strong></p><p>I first heard about ChatGPT in a Slack thread between developers saying it was skyrocketing. That was the entire distribution mechanism. No strategy paper, no sponsor, no blueprint, no coach. One person tried it. Then a lot of people did.</p><p>It spread that way for a structural reason, not an emotional one: <strong>AI is the first transformation in this lineage that delivers its full value to a single individual on the first day, without requiring anyone else to change anything.</strong> A developer using an assistant does not need the release board’s permission. An analyst drafting with a model does not need her colleagues on the same cadence. There is no coordination to manufacture, so there is nothing for the coordination machinery to do, and no hierarchy standing between the proof of concept and the next person who wants it.</p><p>The numbers confirm what anyone inside an enterprise can see. Microsoft’s Work Trend Index found (already in 2024) that three-quarters of knowledge workers were using generative AI at work, most bringing their own tools, while a majority of leaders admitted having no plan. MIT’s <em>GenAI Divide</em> research found employees in over 90% of surveyed organisations using personal AI tools while less than half of those organisations had bought anything at all. And the use is not junior staff evading policy: recent research found senior decision-makers twice as likely as their own reports to use unapproved tools.</p><p>The pattern those numbers describe is the inversion of everything in the first half of this article. <strong>The formal programme fails while the informal adoption succeeds, in the same buildings, among the same people.</strong> For twenty years, adoption without a programme was impossible. Now the programme is the thing arriving late.</p><p>And this time the stakes of not seeing it are different in kind. Shadow IT was ungoverned tooling, by losing track of files, unknown vendors, solvable by inventory. Shadow AI is ungoverned judgement: analysis, drafting, prioritisation, recommendation. The artefact you would need to govern is a thought process, and it leaves no trace.</p><p><strong>The freedom was taken first</strong></p><p>Most large organisations exist to produce stability, control and structure. That is not a flaw; it is their function. But underneath it sits an assumption: that autonomy is something the organisation grants in advance, inside a frame it designs.</p><p>That assumption is now void. People have already taken the freedom. The frame can no longer be granted. It can only ratify what is happening or criminalise it.</p><p>The evidence on criminalising is in, and it is unambiguous: control produces concealment, not compliance. Two-thirds of professionals in one large 2026 survey had used AI at work while believing it was against policy. Punishment doesn’t reduce use; it reduces visibility. The organisations aiming hardest at control know the least about what is actually happening inside them, and a maturity assessment in that environment measures how much of existing practice you have managed to see, and it will report a comfortable number. Exactly like a training count that rises every week while nobody’s accountabilities change.</p><p>So the position I will defend is this: <strong>freedom in a frame still works, but only if you endorse the humans already inside the frame.</strong> Endorsement is not a soft word. It means making the sanctioned path faster and better than the unsanctioned one, so that use surfaces voluntarily. It means removing the career risk from saying “I used a model for this.” And it means management leading by example, which, it turns out, is the same condition that decided every transformation before this one. Hierarchy didn’t stop mattering. It stopped being able to prevent adoption. It can still prevent honesty about it.</p><p>This is not only a coaches’ problem. If you run delivery, your capability map is now wrong: the differences between individuals doing nominally identical work are large, growing, and captured by no role description or performance system you own. If you run technology, your governance surface is not a tool list; it is thousands of daily acts of delegated judgement you cannot inventory. If you write the policy — the research says you are probably violating it.</p><p><strong>What worries me</strong></p><p>The gap between what organisations are capable of and what the individuals inside them have already become is too wide to close gracefully. And the reflex I see everywhere is to close it fast with tooling, for example a platform, an enterprise licence, a copilot rollout, rather than with a change in how people are expected to think and decide.</p><p>Tooling is the fastest thing an enterprise knows how to procure. It arrives with a vendor, a budget line, a rollout plan and a dashboard that will climb every week.</p><p>It is, in other words, the same reflex that once produced a strategy paper with an unsourced number in it: buying the artefact of a change instead of making one.</p><p>Nobody wrote a strategy paper for AI. It spread anyway. Sit with what that says about the last twenty years, before we spend the next five proving we learned nothing from them.</p><p><strong>Sources</strong></p><ul><li>Fowler, M., <em>The State of Agile Software in 2018</em> (Agile Australia keynote); <em>The Agile Imposition</em> (2006)</li><li>Sutherland, J., <em>Why 47% of Agile Transformations Fail</em> (2020)</li><li>Microsoft &amp; LinkedIn, <em>Work Trend Index</em> (2024)</li><li>MIT NANDA, <em>The GenAI Divide: State of AI in Business</em> (2025)</li><li>Wakefield Research / PagerDuty, <em>Shadow AI Survey</em> (2026)</li><li>TrustedTech with Censuswide, shadow AI research (2026)</li></ul><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=63d324ee5cde" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>