A Working Map for the Next Phase

Most senior leaders I talk to are past the should we? phase with AI. Their teams have experimented. They've watched a few people in their org do impressive things with chatbots and a few others do nothing. Now they're into something harder: what do we actually do now?

Different companies are in different places — some far along on data policy and barely started on workflow patterns, some with prompt-fluent operators in marketing while their tech teams catch up, some with culture work to do, some with skills work, many with both. There's no scorecard or timetable. AI is many things, but it's also gunpowder. The rules aren't written yet, the stakes are real, and the work has to happen in the open — between people who are willing to compare notes honestly.

Here's a working map of the friction points companies are actually wrestling with. Four of them. None of them solved. All of them places where business needs and human needs are both real, and where the work has to be deliberate.

We've been here before

When email rolled out across businesses in the 90s, nobody knew when to email versus call versus walk down the hall. Nobody knew what "reply all" was for, or whether silence meant agreement, or how to write a subject line that meant anything. There were no email best practices because there had never been email before. The norms got worked out over twenty years, mostly by people improvising and watching what worked.

We're in that stretch with AI now. The pattern library is being written in real time. The people doing the writing are spread across companies, mostly not coordinating. There's room out here, and we owe each other better than guessing alone.

The patterns don't exist yet

Most companies are still using AI as a basic chatbot. You ask a question, you get an answer, the session ends, and the next one starts from scratch. Nothing accrues. Every new use case begins at zero.

Compounding wins — wins that build on each other instead of starting from zero each time — need something underneath them. A way to keep context across sessions. A way to share patterns across teams. A way to know what's worked. That kind of plumbing is still mostly built for individuals, not for big companies. The few operators figuring it out are doing it by hand, often without organizational support.

This isn't a budget problem or a willpower problem. The tools are new. The patterns don't exist anywhere yet. There is no playbook to dust off — which is why the people building the patterns now are doing real work for everyone else who will follow.

The boring infrastructure

Everything that has to get worked out before any new technology lands in an enterprise. None of it is exclusive to AI:

Data policy — what data can be sent to which models, what gets logged, what gets retained
Budget and license management — who pays for which subscriptions, how seats are allocated, how to evaluate which tools are worth the money
Security, audit, and legal — log retention, audit trails for AI-assisted decisions, vendor security posture, IP exposure
Risk and liability — who carries the bag when AI makes a mistake that hurts a customer or creates legal exposure
HR and labor — what AI use is permitted in performance evaluation, hiring, employee communications; disclosure to employees, customers, regulators
Procurement — how new AI tools get approved, what standard contract language looks like
And so many more — records retention, accessibility, training and certification, vendor consolidation, internal communications, change management, partner contracts, customer disclosure, brand standards, A, B, C, 1, 2, 3, boom, bam, bim. ALMOST EVERYTHING.

Each is its own thing. None of them is "AI strategy." They're the infrastructure any new technology needs to land in a big company — and even the boring infrastructure transformation is blowing my mind.

The instinct in some organizations is to centralize all of this under one big AI initiative. That instinct is mostly wrong. AI is broadly useful technology — like email, like spreadsheets, like the web. The work is for each group to figure out what role it plays in their work, and for centralization to handle only what genuinely crosses group boundaries. The 1995 question was what role does email play in your group? The 2026 question is the same — with one word swapped.

The cultural pressure

The cultural friction is the place where this gets human, and it's the place we cannot afford to handle clumsily.

Five distinct pressures, each with its own dynamics:

Identity and security. Three nested worries: will my role exist? If it changes, into what? Can I become the new version of myself in time? Most companies talk about the first one and ignore the other two — which is where the actual day-to-day worry sits.
Trust between management and labor. The old tension between productivity, cost, and job security has always been there. AI sharpens it because the gains become measurable. If managers hadn't built trust before AI showed up, AI makes that obvious fast. And even where trust was strong, fear can wear it down — people are watching to see what AI means for them, and trust takes work to hold onto when the ground is moving.
Role boundaries. AI changes what work belongs to whom. With boundaries unsettled, ambitious people grab and defensive people retrench — both reactions harden positions before the org has had time to think clearly.
Tech teams and the context fallacy. Technology teams sit closest to the tools, so they look like the natural automators. But automation requires deep workflow context they don't have. Tech serves the domain experts here, not the other way around — not tech automating workflows over the heads of the people who do the work.
The temporal trap. Right now, business pressure is pushing toward "automate now, cut headcount, rely on third-party solutions" — at exactly the moment when nobody knows yet what knowledge has to stay in humans and what can be safely automated. Companies cutting too aggressively right now are placing a bet at the worst possible time. The knowledge that walks out the door doesn't come back.

Most organizations don't have a forum where any of this gets surfaced honestly. The work to build those forums is part of the friction.

The skills haven't been built yet

The most under-recognized of the four. The people who can do this work well need a skill set that barely existed as a recognized category three years ago.

Some pieces of it pre-existed AI and got more important. Others are genuinely new:

Workflow thinking — looking at a process end-to-end and seeing where AI changes what's possible
Spec writing — describing what you want clearly enough that an AI can do it without you in the room
Evaluation discipline — knowing whether AI output is actually good, reproducibly, not just "looks fine to me"
Pattern recognition — recognizing that a workflow problem is a context-loading problem, or a verification problem, or a handoff problem, and reaching for the AI move that fits
Editorial judgment — saying specifically what's wrong with AI output so it can be improved
Translation between domain and tooling — turning domain expertise into something an AI can execute against
Knowing when not to use AI — the judgment to recognize tasks where AI is slower or worse than the manual approach

Most organizations don't recognize these as skills. They can't hire for them. They don't train for them. The bottleneck on AI adoption isn't AI capability — it's the human capability to design with it, and we have to build that capability deliberately or watch it accumulate only in the few people who got to it on their own.

The work we owe each other

Four friction points. Each one different. Each one demands different work, by different people, on different timelines.

Most of the public conversation picks one and calls it the answer. It's a tools problem. It's a culture problem. It's a skills problem. All of those are partly true and individually insufficient. The companies that handle this well are going to be the ones holding all four at once and working on each of them in parallel — and the only way through is collaborative. Comparing notes. Sharing what's worked and what hasn't. Naming the hard parts honestly instead of waiting for someone to package the answer.

We must not leave another generation of factory workers behind. We don't have to, if we're deliberate.

These friction points are exactly where we have to lean in — for business reasons and for ethical reasons, each sacred in their own right.