Engineering notes from the field.
Essays and field notes on AI, software engineering, design, and the craft of building product teams that ship. Written by the engineers doing the work.
Anthropic's $30B Run Rate and the 3.5GW Footnote
Anthropic passed OpenAI on revenue this month and locked in 3.5 gigawatts of Google/Broadcom TPU capacity for 2027. The footnote on the Broadcom filing — compute commit is conditional on Anthropic's continued commercial success — is the part worth reading twice.
Chrome Auto Browse: The Browser Is the Agent Surface Now
Google shipped Auto Browse and Chrome Skills to enterprise at $6/month this week. The interesting question isn't whether it works — it's what it does to the handful of startups selling browser agents as a standalone product.
Kimi K2.6 and the Open-Weight Frontier: Reading Moonshot's GA Release
Moonshot dropped the 'Preview' label on Kimi K2.6 and shipped a 1T-parameter open-weight model that trades blows with the closed frontier and leads it on Humanity's Last Exam with tools. The real story isn't the numbers — it's what an open weight at this tier does to the buy-vs-host math.
Nearshore, Offshore, Onshore: The Tradeoffs Buyers Actually Feel
Cost per hour is the easy comparison. Timezone overlap, communication overhead, and what each model breaks under stress are the ones that actually decide whether a project ships. A practical read for teams choosing between geographies.
Staff Augmentation vs Project Delivery: The Engagement Model Is the Decision
The wrong engagement model turns good engineers into bad outcomes. A framework for choosing between embedded headcount, fixed-scope delivery, and a full hire — before the contract gets signed.
AI Development vs AI Training: Two Different Programs, Two Different Buyers
They sound interchangeable. They are not. AI development builds AI into products; AI training improves the model itself. Confusing the two is how teams end up with a contract that solves the wrong problem.