The Harness Is the Product
As agents move from prototype to production, the engineering discipline around the model — state, durability, oversight — matters more than the model itself.
Insights
Reviews, opinions and analysis on what is actually happening in AI.
As agents move from prototype to production, the engineering discipline around the model — state, durability, oversight — matters more than the model itself.
A catastrophic third-party database exposure and a local model beating a frontier rival on a single consumer GPU arrived in the same week — and they argue the same point.
Three separate incidents this week involved agents that did exactly what they were built to do — into a sandbox, onto the internet, and against a target nobody chose.
An agentic CRM launch and a public vendor-audit checklist arrived in the same week — and together they raise the same question: do you know what is already running in your business?
A critical local-filesystem escape in a desktop AI assistant showed that the riskiest agent is the one with access to your own machine — not the one in the cloud.
Enterprises are moving from renting frontier models to fine-tuning their own on private data — and OpenAI's first hardware product suggests the vendors expect that shift to accelerate.
Anthropic's $30B valuation and Azure availability arrived alongside a public AI-security competition result that says more about defensive maturity than any funding round.
The EU's streamlined AI Act approval landed the same week self-hosted agent frameworks disclosed security flaws — a reminder that compliance paperwork and engineering reality are different things.
A small non-transformer model matching rivals three times its size signals that capable private AI is arriving on ordinary hardware — while shadow AI becomes the leading incident cause.