Insights

Notes from the work.

What we learn building automation into businesses that already run on something, and running the data office underneath it.

The Clartelia Brief, free every week

Sonnet 5.5 and GPT-6.1 Sol: which of your jobs still need a top AI model?
Anthropic and OpenAI both released a stronger middle model in the last week of September, Claude Sonnet 5.5 and GPT-6.1 Sol, and both cost $2 and $10 per million tokens. Keep the top models for open-ended work that needs judgment, and move the scoped, repeatable jobs to the middle tier once they pass a test on your own examples.
October 5, 2026en
Low, medium, high or max: how much thinking should you pay an AI model for?
Claude and GPT-6 both let you choose how hard the model thinks, from low to max, and the thinking is billed as output. Start at medium, test low on real examples, and raise the level only for the tasks that fail.
September 28, 2026en
Claude Opus 5.5 vs GPT-6: which model is the real rival, and which one should you use?
Claude Opus 5.5's real rival is GPT-6 Astra, not GPT-6 Sol. At the top, Claude Fable 5.1 and GPT-6 Astra cost the same $10 and $50 per million tokens, while Opus 5.5 costs $4 and $20 and Anthropic says it matches Fable 5.1 on most work. Sol is priced exactly like Claude Sonnet 5, and Luna is ten times cheaper than Claude Haiku.
September 26, 2026en
Why does the close get slower every time you add an entity?
Because what scales isn't the data — it's the disagreement. Each entity arrives with its own definition of revenue, margin and cut-off, and the close absorbs the cost of making them agree, every month. New tooling makes the mechanical work faster without deciding what the numbers mean, which is why the close date rarely moves.
August 22, 2026en

Notes from the work.

Sonnet 5.5 and GPT-6.1 Sol: which of your jobs still need a top AI model?

Low, medium, high or max: how much thinking should you pay an AI model for?

Claude Opus 5.5 vs GPT-6: which model is the real rival, and which one should you use?

Why does the close get slower every time you add an entity?