The diary,
out loud.

A daily recap of what the labs announced in the last 24 hours, then one deep dive. Every story answers one question: what changes in our work?

Podcast feed (RSS)

Receipts

Every claim in episode 1 comes from a company's own announcement. Vendor numbers are the vendor's.

• OpenAI: GPT-6 for everyone
• Google: Gemini at work
• Google DeepMind: SynthID Detector
• Anthropic: Claude Haiku 5.5
• Anthropic: Cyber Mission

One test you can run Monday

1. Pick your three most expensive agent tasks.
2. Re-run each on the cheaper model, same prompts, same checks.
3. Record pass rate and cost per completed task, not per token.
4. If the cheaper model fails a check, keep the expensive one for that task and write down why.

My take, not a measured result.

Same standard. Different format.

Source links sit beside the player. Unsupported numbers get cut, not disguised by confident narration.