Your AI mandate is the wrong end of the lever

The AI mandate is going to collide with the org chart, and the org chart is going to win.

By now you’ve seen one of these memos. The CEO writes to the whole company. AI is no longer optional. Every team will use it. Performance reviews will reflect it. Managers will be measured on adoption.

Shopify did it. Box did it. Duolingo, Klarna, IBM. The script is consistent. The expectation is consistent. The result, so far, is also consistent: nobody really knows what’s happening on the ground.

Employees are doing one of three things. Some are genuinely adopting the tools and changing how they work. Some are openly resisting, sometimes loudly, often privately. The largest group is the quiet middle: people using AI on the side, not telling their manager, performing the rituals of adoption in meetings while doing essentially the job they did six months ago. Shadow AI, the same way there used to be shadow IT, except this time the thing being hidden is productivity.

Meanwhile, the same companies issuing the mandates are quietly telling their investors that AI is going to reshape the workforce. Each employee will produce more. Headcount will flatten. Layers will compress. The org chart of 2027 will look very different from the org chart of 2024.

I think this is mostly true, eventually. I also think the path between here and there is going to be far messier than any press release suggests. And the reason is Conway’s Law.

A 1968 observation that finally matters

Melvin Conway wrote, in 1968, that organizations design systems that mirror their own communication structures. If your team has three people who don’t talk to each other, you will ship three products that don’t talk to each other. If your customer service org is siloed from product, you will ship customer experiences where help is bolted on rather than designed in. The org chart is destiny. You ship your org whether you mean to or not.

This started as a software observation. It applies more broadly. Anything an organization produces, the org structure shapes. The product, the marketing, the customer experience, the way bugs get fixed, the way decisions get made. All of it is a faithful rendering of how the people talk to each other.

Here is the part most companies haven’t worked out yet. They are mandating a new input, AI, into a communication structure that was designed for humans operating at human throughput. The reporting lines haven’t changed. The meeting cadence hasn’t changed. The handoffs between teams haven’t changed. The way work gets reviewed, approved, shipped, and measured hasn’t changed.

Conway’s Law tells you what comes next. The output will reflect the structure. You will ship the org you have, not the org you’re imagining. The org you have has not yet figured out how to integrate AI work into its actual operating rhythm.

The seams are already showing

This is already visible in shipped product. Look at the AI features added to enterprise software in the last twelve months. Most of them feel bolted on, because they were. Most have an interaction model that doesn’t compose with the rest of the product, because the team that built the AI feature was a new pod that didn’t share rituals with the rest of the org. The seams are visible. The customer feels them.

It is also visible in customer service. Companies that announced AI agents in the contact center are quietly walking it back, not because the agents can’t answer questions, but because the agent’s answers can’t flow back into the rest of the operation. The agent resolves something, but the resolution doesn’t propagate. The customer gets bounced. The org didn’t redesign the handoffs, so the handoffs broke.

You are watching, in real time, organizations ship their actual communication structure under a thin coat of AI paint. The paint is not the problem. The structure underneath is.

The throughput mismatch

The flatter-org thesis is going to compound this. Companies want each employee to produce more, move faster, operate with more range. But the hierarchy hasn’t compressed. VPs still have the same span of control. Directors still have the same number of reports. The expectation at the individual level is real, but the structures around the individual have not been redesigned to absorb it.

What you get is a throughput mismatch. Individual contributors do more, faster, with AI help. The layers above them, designed for slower input, become the bottleneck. The work piles up at the review step, at the approval step, at the let-me-run-this-by-my-boss step. The org chart quietly throttles the AI it just mandated.

The honest version of the productivity story is not “everyone is now ten times faster.” It is “some people are now ten times faster, in some tasks, and the organization is mostly the same shape, so a lot of that speed is being absorbed by the friction it always had.”

What gets redesigned

The companies that get this right are going to do something much harder than buying licenses. They are going to redesign their communication structures. Smaller teams with more end-to-end authority. Fewer handoffs. Different review cadences. Different definitions of what a manager does. The hierarchy will compress, but only after the communication patterns compress first.

Conway’s Law works in both directions. If you want to ship a different product, you have to ship a different org. The org has to change first. The product follows.

This is the part nobody can mandate. You can mandate tool adoption. You cannot mandate that two teams who have spent five years not talking to each other suddenly coordinate well. You cannot mandate that a VP who built her career on managing twelve directs decide she should manage four. The cultural and political work of changing a communication structure is slow, painful, and largely invisible from the outside. It does not photograph well in a quarterly earnings call.

What’s coming

My prediction is this. The next two years will produce a lot of AI-shaped products that do not feel coherent, from companies that announced AI-first strategies that did not survive first contact with their own org charts. The customer will feel the seams. The employees will quietly know why. The CEOs will be confused why the productivity gains did not show up the way they expected.

The small number of companies that ship something genuinely new, AI products that feel designed instead of assembled, will be the ones that did the unglamorous work of redesigning how their people talk to each other before they redesigned what they ship.

That is the work I find most interesting right now. Not the tools. The communication structures. The org charts. The cadences. The seams.

The tools are the easy part. The org chart is the product.