The Quiet Architecture Shift Behind Phi-4
A field report on long-context workflows and what it changes for product managers.
For most of the last year, the conversation around computer-use agents has been louder than the evidence. That is starting to change.
Teams that win with computer-use agents tend to share a habit: they write the evals before they write the prompts. Everything else follows from that.
Ramp has been quietly running onboarding through ChatGPT Canvas 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. Spotify's team treats theirs the way an SRE team treats a runbook.
The cost curve matters here. Mistral Large 3 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.
Teams that win with tool-first agents tend to share a habit: they write the evals before they write the prompts. Everything else follows from that.
What Cohere actually shipped with Llama 4 is less a single capability and more a cluster of small, compounding improvements — the kind that only show up when you put a real workflow on top.
None of this guarantees a clean story. Anthropic 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.