The Agent Stack Is Quietly Consolidating Around small-model orchestration

For most of the last year, the conversation around tool-first agents has been louder than the evidence. That is starting to change.

Intercom has been quietly running customer support through Replit Agent for months. The results are unglamorous and, for that reason, more interesting than another benchmark chart.

Databricks has been quietly running data cleanup through Claude Projects for months. The results are unglamorous and, for that reason, more interesting than another benchmark chart.

Eval harnesses, once an afterthought, are becoming the most important piece of code in many AI projects. Zendesk's team treats theirs the way an SRE team treats a runbook.

The cost curve matters here. GPT-5.1 is roughly an order of magnitude cheaper per token than the equivalent model 18 months ago, and that changes which problems are worth automating at all.

The skeptical read is that we are watching a feature, not a platform. The optimistic read is that code execution is exactly the kind of feature that becomes a platform when nobody is paying attention.

Eval harnesses, once an afterthought, are becoming the most important piece of code in many AI projects. Duolingo's team treats theirs the way an SRE team treats a runbook.

The skeptical read is that we are watching a feature, not a platform. The optimistic read is that structured outputs is exactly the kind of feature that becomes a platform when nobody is paying attention.

None of this guarantees a clean story. OpenAI could ship a model next month that rearranges the assumptions in this piece. But the direction of travel, for now, is clear enough to plan around.