Move all the info from 100 emails into 1, once a day.
A daily AI digest written for where you actually are, plus one verified skill you can use tomorrow.
A reading service for people who already read too much.
We run a small newsroom that reads everything written about AI today and writes you one email about it tomorrow morning. The page is long. The newsletter is short.
- i.
One email a day.
Delivered at the local time you pick. Read it on the train. If today was quiet, the email is short. If today was loud, the email is still short.
- ii.
Framing that fits where you are.
Same news, different version for a parent than for a senior ML engineer. We sort readers across 45 cells and write each story 45 ways.
Same news. Forty-five different framings.
Three technical depth levels, three awareness levels, five curriculum tracks. We sort each new reader into one cell on signup, then the digest is rewritten for that cell. A senior ML engineer and a worried parent get the same headlines, in entirely different words.
The cell is not a label. It changes as you read. Click any cell at right to preview the framing in section 03.
A specimen digest.
Each digest has three parts: an editorial lead written for you, today's curriculum lesson, and one verified skill.
- Lead, written for your cell~3 min
- Today's curriculum lesson~1 min
- Today's skill~1 min
Switch the persona at section 02 to see the same news re-framed.
Claude 5: 1M context, sub-agents are first-class, function-calling spec rewritten.
Decide if this changes your migration plan in tomorrow's standup. Don't waste two days re-reading the changelog.
Pricing is roughly flat to Sonnet 3.7 under 200k tokens. The function-calling SDK has a breaking change on `tools[].input_schema` (was `parameters`). Sub-agents replace the recursive loop pattern and cut p99 latency by 40% on the eval we ran. Migration diff is in the link.
Open the diff link, scan the 12-line breaking change list, decide before standup whether you want a one-day spike or a two-week migration.
Function-calling versus tool-use, drawn out
Function-calling is a JSON-schema contract the model fills in. Tool-use is the broader pattern where the model decides which tool, in what order, with what inputs. Today's lesson: a side-by-side trace of the same task in both modes, with notes on where each one wins.
If you are not sure a citation is real, omit it. Do not invent.One email. Once a day. At the time you pick.
Free at launch. The first digest lands tomorrow at the local time you choose during signup.