Somewhere right now, a VP is drawing a new org chart. Boxes, arrows, a shiny “AI Enablement” box bolted onto the side. They believe that when the reshuffle is done, the transformation will be done. It won’t. You cannot reorganize your way into a culture that doesn’t exist yet — you can only draw a prettier picture of the one you already have.

That’s the trap The Incongruent Org diagnosed: a company where half the team moved up the stack and half didn’t, and the seams show in the review queue, the estimates, the standup silences. That post was the X-ray. This one is the treatment plan. And the treatment is not a reorg.

Here’s the whole thing in one sentence, and it’s an ordering as much as a list: empathy first, foundation second, language third. Skip the empathy and you lose the people you needed most. Skip the foundation and you scale chaos at machine speed. Skip the shared language and no one in the building can reason together about what you’re all supposedly building. The order isn’t decorative. Get it out of sequence and each later step lands on ground that can’t hold it.

Lead with empathy, because the fear is real

Before the tooling, before the harness, before any of the smart stuff — start with the fact that a lot of your best people are quietly scared.

Not scared in the way that shows up in a survey. Scared in the way that shows up as a senior engineer who suddenly has strong architectural objections to every AI-assisted PR. As the tenured dev who “just prefers to write it by hand” and frames twenty years of hard-won mastery as a stylistic preference. The New Caste System named the thing underneath that: a new ladder is forming, it has nothing to do with pedigree, and the people highest on the old ladder often have the furthest to fall. Of course they’re bracing. They’re not being difficult. They’re being human about the possibility that the thing they were the best at just stopped being the thing that’s scarce.

You will be tempted to shame the holdouts. Don’t. Every eye-roll at the “AI skeptic,” every all-hands where adoption gets framed as enthusiasm-versus-cowardice, tells your most experienced engineers that the safe move is to hide. And a senior engineer in hiding is the single most expensive failure mode in this whole transition, because their judgment is exactly the input the machine can’t supply — and now they’re spending it defending their pride instead of catching the model being confidently wrong.

Leading with empathy is not softness. It’s the coldly correct play. You bring the holdouts across — you don’t leave them on the far bank as a cautionary example. That means naming the fear out loud, protecting people while they’re bad at the new thing, and being visibly bad at it yourself in front of them. The manager who won’t fumble through their own first agent session in public has no standing to ask anyone else to. Fear doesn’t respond to a mandate. It responds to cover.

Build the foundation before you roll out the tools

Empathy gets people willing to move. Now, do not hand them the tools yet.

I know how that sounds in a series that has spent six posts telling you to move up the stack. But the most reliable way to make a mess of this is to buy every engineer a license, declare victory, and roll out capability into an org that has no floor under it. Easy Bake Oven Developers was about the individual version of this — the person who tosses ingredients at a toy oven and blames the appliance. The org version is worse, because it’s that mistake multiplied by headcount and compounding through shared context.

The foundation is the stuff that has to exist before the tools do real work: the shared context every agent reads from, the harness — your common layer of rules, memory, and conventions — the review culture that catches bad output before it ships, the conventions that make one engineer’s work legible to the next. Orchestration was the single-engineer discipline of keeping intent in force across a long run. The foundation is that same discipline made institutional — the thing that keeps a team’s intent in force across everybody’s runs at once.

Ship tools before that floor exists and you don’t get leverage. You get Easy Bake Oven Developers at scale, all pointing unverified output at the same shared context, each one’s confidently-wrong assumption fanning out into every future prompt everyone else writes. The burndown chart will look incredible. That’s the tell. You’re not accelerating — you’re shipping tech debt at a rate no human team could have produced on its own, and the fastest-looking teams are digging the deepest holes.

Foundation before rollout. Slower on day one. The only order that doesn’t compound the mistake.

Teach the org a shared language

Empathy gets people moving. The foundation gives them somewhere solid to stand. But a team standing on solid ground and speaking six private dialects still can’t build together — because they can’t even agree on what “done” means anymore.

This is the part almost everyone skips, and it’s the one I’d fight hardest for. When one engineer says “just have the agent do it” and means hand it a structured spec and review the diff, and another hears type a vibe and merge whatever comes back, they are not disagreeing. They are failing to communicate, in a meeting where everyone nods. The Team Dimension made the point that a glossary the agent doesn’t read is a diary — and the same is true of a team. A word that means five things means nothing.

So teach the vocabulary on purpose, the way you’d teach any craft. Judgment — the scarce human call the model can’t make. Intent — the structured expression of what to build, which is the real work product now; the code is just the exhaust. Leverage — output per unit of your attention, the number that actually matters, not tickets closed. Eval — the measurable check that replaces “it worked when I tried it.” And the one this whole series has been walking toward:

A shared language isn’t corporate hygiene. It’s the precondition for a group of people to reason together — to disagree productively, to hand work off without it degrading, to mean the same thing when they say the moat is what we’re protecting. Without it, every conversation quietly re-litigates its own definitions and nobody notices the meeting accomplished nothing.

The moat was never the models

Which brings me to why the order matters at all — why you can’t just buy your way here.

The instinct of every leader under pressure is to treat this as a procurement problem. Get the best models, the best licenses, the best tooling, and win. But the models are the one thing your competitor can have too, at the same price, by dinner. If your entire advantage is which foundation model you’re renting, you have no advantage — you have a subscription, and so does everyone across the street.

The moat is the compounding. It’s the context your org has accumulated and made legible; the taste your engineers have sharpened about what’s worth building; the judgment, distributed across a team that speaks the same language, about when to trust the machine and when to keep hands on the wheel. That doesn’t come in a license. It’s grown, deliberately, over time — and it’s exactly what the ordering produces. Empathy keeps the people whose judgment is the moat. The foundation is the substrate the context compounds in. The shared language is what lets it compound across a team instead of trapping the good stuff in one person’s head. Skip any of the three and the moat never forms — you just have expensive tools and a prettier org chart.

And here’s the honest, uncomfortable part, the reason nobody sells you this in a keynote: it doesn’t measure well early. The reliable industry number people quote — that around 70% of transformations “fail” — turns out to be soft on inspection, traceable to an unsourced consulting line rather than a clean measured rate. But McKinsey’s own 2018 survey of 1,793 leaders found only 16% said their digital transformation had both improved performance and equipped them to sustain it. Sixteen. And the through-line in the ones that stuck was never the technology — it was the people side: capability, culture, communication. The exact three things a reorg can’t install and a purchase order can’t buy.

That’s the whole series, really. Back in Acquiring Talent in the Age of AI I argued the interview should stop grading code production and start grading judgment. Everything since has been the same claim pointed at a different surface — the individual, the tools, the metrics, the team, the org. It always lands in the same place: the scarce thing is human, the durable thing is human, and the leader’s job is to protect and compound the human thing while the cheap, clonable thing gets cheaper.

You can’t org-chart your way to that. You can only lead people to it — in order. Empathy so they’ll move. A foundation so they don’t move into chaos. A language so they can move together. And a moat, at the end, that no competitor can buy because you didn’t buy it either.

So here’s the question I’ll leave the whole series on. Look at whatever you’re actually doing this quarter to “adopt AI” — the licenses, the mandates, the new box on the chart. Strip out everything a competitor could purchase by Friday. What’s left? Because whatever’s left after you subtract the buyable — that’s your moat. And if the honest answer is nothing yet, you don’t have an adoption problem. You have the most important, least urgent-looking work in front of you, and no purchase order will do it for you.