July 2026

9 entries from July 2026. The latest entries are on the AI Feed itself.

• 31 Jul 2026
  The unit price is not the bill

  Unit prices falling while bills rise, a scheduler that doubles throughput without touching the model, and the best model on a benchmark behaving worst on it.

• 30 Jul 2026
  The controls have not moved into the harness

  A brake with no trigger on it, a protocol that deletes the layer where state used to sit, and the same model scoring 13.3 and 38.3 depending on the harness.

• 29 Jul 2026
  Verification is the constraint, not capability

  A rewrite of half a million lines in eleven days, a licence that changes with your revenue, and a result that took a week to find and a month to trust.

• 28 Jul 2026
  The benchmark is measuring the harness, not the model

  Three headline scores that belong to an assembled system rather than a model, two of them disclosed in the vendor's own footnotes.

• 27 Jul 2026
  Constraint is moving from the prompt into the plumbing

  An effort budget that moves the outcome more than the model choice, a system prompt cut by four fifths, and a corpus that settles provenance where it is built.

• 26 Jul 2026
  The largest MCP revision changes how you run it, not what it does

  The biggest overhaul of the protocol wiring agents into enterprise systems makes almost nothing more capable. It makes MCP something a firm can run on ordinary infrastructure an...

• 26 Jul 2026
  Part 2: agents need infrastructure, not just models

  A serving split that halves response time without touching the model, and a firewall that keeps real credentials out of the sandbox entirely.

• 25 Jul 2026
  The model stopped being the variable

  A multi-agent system delivered as one model, inference split across two kinds of hardware, and a voice channel wired straight to an execution path.

• 24 Jul 2026
  Autonomy now needs a filed decision policy

  Agent decision rights written into law, and the plumbing that makes answering for them possible.