When the senior engineer in the room is a model
Architecture conversations now happen with a participant who has read every paper, every blog post, and every Stack Overflow answer. Here is how the meetings change.
The change is not "automation"
I have spent the last eighteen months watching enterprise architecture meetings reorganise themselves around a new participant. The participant is a model — sometimes Claude, sometimes GPT, occasionally an in-house tuned variant. The participant is not "doing the work." It is participating in the conversation.
This is a more interesting change than the productivity narrative captures.
What changed in the meetings
In 2023, the model in the room was a search engine with better manners. In 2026, it is a peer. The architect asks "what do we think about Kafka here?" and the model offers a position, with citations, with caveats, with the same hedging that a careful senior engineer would offer.
The result is that the meeting moves faster. The architect no longer has to fetch context. The context is in the room.
What did not change
The decision is still the architect's. The accountability is still the architect's. The political work — getting three teams to agree on a contract — is still the architect's.
Three things I tell new architects
1. The model is a peer, not a tool. Treat it like a peer: argue with it, ignore it sometimes, defer to it sometimes.
2. Write the questions down. The questions you ask the model are the same questions you should be asking the room.
3. The model does not know your politics. It will suggest the technically optimal solution. Your job is the politically possible solution.
Filed in AI-Native Engineering Platforms
Tags: AI agents · Architecture · Leadership