Oracle 10-K filing, June 2026

An SVP I will keep nameless had a slide that read Agile by Q3. He had the budget, the certified trainers, the team rosters redrawn into squads, and a launch date with his name on it. What he had never thought to check was whether anyone in his organization trusted anyone else to make a call. A year and a half later the squads were running stand-ups that were status meetings in disguise, the backlog was a wish list the executives reshuffled on a whim, and velocity had become a number people massaged before every review. He reached the conclusion most executives in his chair reach. Agile does not work.

Somewhere online a practitioner read that same story and reached the opposite conclusion with equal certainty. He did not do real Agile. The certifications were a weekend, the coaches were order-takers, the squads were the old org chart wearing new labels. The practitioner is not wrong about a single detail. He is also, in his own way, avoiding the thing that actually broke, and that half of the story I am holding for next week. This week I want to stay with the executive, because his mistake is the one repeating itself right now inside nearly every AI rollout I look at.

Start with what the executive actually wanted, which was a quick win. He wanted a lever he could pull from the top of his department and have the whole thing run faster and cleaner by the next quarter. Agile was never going to be that lever, because the entire purpose of it is to take the lever out of his hand. He could mandate the ceremonies. He could have a stand-up running by Tuesday and a wall of sticky notes by Friday. What he could not do by Tuesday, or by mandate at all, was create the one condition the ceremonies exist to express, which is that the people closest to the work are trusted to decide how the work gets done.

That word, trust, gets waved around these conversations until it stops meaning anything, so let me make it concrete. Trust is what you show when you hire a person and then let them make the decisions that sit inside the scope of the job you hired them for, without sending each one back up the chain. It means pushing the authority to decide down as far as it will travel, to the person actually standing where the decision has to be made. An organization that already works this way is most of the way to Agile before it runs a single ceremony. An organization that does not cannot purchase its way there with a framework. When the trust is missing, that is the first thing to build, and it is slow, unglamorous work that shows up on no quarterly slide.

So how do you build it, if you cannot order it into being? The same way you build any behavior in an organization, by rewarding the things that let it grow and declining to reward the things that choke it. The hinge is what happens when someone carries bad news into the room. Where trust can take root, the person who surfaces a problem early or owns a miss instead of burying it is met with gratitude, and over time with more responsibility, because they just saved the organization something real. Where it cannot, that same person collects a mark against their name, and everyone watching takes the lesson to protect themselves and never be the one left holding the problem when it finally shows. You cannot extend trust downward while punishing the honesty it runs on.

The incentive that does the deepest damage is the one that rewards control for its own sake. Praise a leader for tightening constraints and layering on process and you have handed that leader a flawless alibi for the day the plan comes apart. The story tells itself. Those people did not do what I told them, and had they listened, we would have succeeded. The failure never attaches to that leader, because the leader did the precise thing the organization rewards, which is to control. It looks entirely different when the result lands under a leader who owns the outcome, who has no alibi to reach for because they took the outcome as theirs from the start. That kind of leader does not show up by luck or good intentions. An organization gets one by building an incentive structure on purpose to produce and protect them, one that rewards the person who owns a bad result over the person who can prove the rules were followed.

This is why the transformations rot into theater. A leader installs self-organizing ceremonies on top of a structure that still routes every decision of consequence upward for sign-off, and the daily stand-up becomes a meeting where capable adults narrate their progress to the person who still makes all the real calls. The shape is Agile. The substance is the command hierarchy it always was, only slower now for the added overhead. The executive watches the poor result and blames the method, when the method was never the thing actually running.

Now watch the same person meet AI. The hunger is identical. A quick win, a lever pulled from the top, the department remade by a technology this time rather than a methodology. So he buys the licenses, declares the organization AI-first, mandates the adoption, and waits for the productivity to arrive. What arrives is the same costume he got the last time. The pilots that return nothing anyone can trace to the business. The workforce that reads the rollout, understands it as a threat to their own standing, and declines it. The record quarter announced in the same breath as a layoff that names AI as its cause in a legal filing. None of these are new failures. They are the Agile failure recompiled on far more expensive hardware.

The cause has not changed. AI does not repair a low-trust organization. It runs on top of the one you have and accelerates whatever was already there. Put a capable tool in the hands of people who are trusted to exercise judgment and it multiplies that judgment. Put the same tool in the hands of people who have learned that every decision gets reversed from above, and you have automated the reversing. The technology magnifies the decision culture it lands in. Drop it into a place where nobody is trusted to decide, and it will help that place produce the wrong things faster than it ever managed before.

Here is the part neither the Agile executive nor the AI executive wants to say in the room. It is far more comfortable to announce that a methodology failed, or that a technology was oversold, than to admit you were never willing to loosen your grip on a single decision, or that the organization you built does not trust its own people to make one. Blaming the tool keeps the verdict on the tool. The honest admission lands on the person who chose not to let go, which is exactly why it stays unspoken. The tool makes a convenient defendant. It never argues back.

James Wright and I gave the opening chapter of our book to this pattern and named it the successful failure, the organization that posts healthy metrics and delivers nothing real. Underneath it sits a framework we shorten to REAL, results through empowerment, accountability, and leadership, where empowerment means decisions genuinely moving down to the people doing the work. We describe a progression that runs from compliance to commitment to ownership, and it explains why a mandated transformation stalls at the first step every time. You decreed it, so people comply. Nobody chose it, so nobody owns it. No tool carries an organization across that gap, and no framework does either. Only the patient extension of trust gets you across.

The honest path is the one that offers no quick win at all. Build the trust first. Push the authority down and leave it there even on the day a decision comes back wrong, because leaving it there is the only way it eventually comes back right. Accept that Agile and AI both succeed as byproducts of an organization that already trusts its people, and never as the program you install to manufacture that trust from the top. Next week I will turn the same honesty on the practitioners, the ones who guard the methodology and mistake purity for results, because they are avoiding something of their own. For now the question belongs to the corner office. Before you write off your last transformation, or bet the next one on a model, what is the last decision you handed away that you did not strictly have to?