Your Agents Are Stuck In Your Org Chart

Someone at a large company told the newsletter's author this week that their department has to grow output by 30% this year while losing a fifth of its headcount, and the plan is to lean on AI agents to close the gap. Watch the token spend too, they were told. This kind of mandate is landing on desks everywhere right now, and almost none of the underlying organizational work has been done to make it achievable.

The piece leans on a June pulse survey of 212 people in data roles, and the numbers are blunt: three out of four data teams have nobody who owns their data products, and teams operating in that kind of anarchy spend 45% of their week firefighting instead of building. Melvin Conway figured out why back in 1968. His observation, now called Conway's Law, says that whatever you build ends up mirroring how your organization actually communicates. You ship your org chart, whether you mean to or not. Infrastructure has a clear owner 85% of the time because it usually sits inside one team. Data products, the dashboards and models that cut across marketing, sales, finance, and support, live in the gaps between teams, and gaps don't get owners.

Drop agents into that gap and something strange happens. People instinctively route around organizational messes: they Slack the one analyst who actually understands the payments data instead of filing a ticket, and they carry a mental list of which tables are stale and which numbers have been wrong since some guy got laid off in 2023. Agents don't have that instinct. They stop dead at the hard boundaries, like missing permissions or join keys that never existed because two teams never talked. But they sail straight through the soft boundaries, the undocumented judgment calls, and hand back a confident answer built on rotten data. A swarm of agents pointed at a siloed company won't dissolve the silos. It'll build more of them, just faster and with better test coverage.d

The fix the article points to comes from Matthew Skelton and Manuel Pais's Team Topologies: the reverse Conway maneuver, where you deliberately restructure teams so the systems you want actually come out the other end. Most data orgs have nailed the platform-team half of that equation, which explains the 85% infrastructure-ownership number. Almost none have built stream-aligned teams that own an entire domain end to end, the way a payments team might own its own pipeline outright.

The tempting shortcut, letting an agent just be the owner, doesn't survive contact with what ownership actually means. Ownership means getting paged at 9pm, knowing why a model was built a certain way, and being the person who answers for the wrong number in the board deck. An agent can't absorb that kind of accountability, at least not yet. Ownership stays scarce partly because it's a hot potato nobody wants and partly because nobody has time to sit down and assign it, which is exactly the kind of slow, political, unglamorous work that AI mandates are now forcing companies to finally confront.