Leading AI product innovation and AI impact thought leadership @ Jellyfish. Formerly Harvard, MIT Lincoln Labs. I love running; it doesn't always love me back.
My @ainativedev podcast with @sjmaple
What AI productivity actually looks like in the wild, multi-agent workflows / the limits of human attention, and what @_jellyfish_co data says about tokenmaxxing. Check it out!
https://t.co/52J7DUuQBt
Early estimates warned @claudeai Opus 4.7 "tokenizer tax" could cost up to 40% more.
Just crunched the numbers from @_jellyfish_co data across 17K users, actual figure (controlling for individual output, type of work, and amount of thinking) is 17%
Talked to @alistairmbarr at @BusinessInsider about tokenmaxxing.
10x the tokens โ 2x the output.
Tokens are ๐ fuel: incredibly powerful, but you need exponentially more to move incrementally faster.
https://t.co/bDhHJ8jfCi
Text version of some of the @_jellyfish_co insights from my video earlier this week.
Data across 135K devs show that creating 6+ context/skill files (CLAUDE.md, Cursor rules, etc.) means 2.3 extra PRs per week.
Each doubling of investment โ 29% more PRs merged.
New @_jellyfish_co data on AI agent customization across 135K devs.
Devs who've created 6+ context/skill files (CLAUDE.md, Cursor rules, etc.) ship 2.3 extra PRs per week.
Each doubling of investment โ 29% more PRs merged.
Good news โ the "tokenmaxxing" trend has run its course!
Showing up in an NPR episode so that my mom can ask my "what does <blank> mean?" is the final stage of any tech trend.
https://t.co/i6H4PCVFnU
๐จ New @_jellyfish_co AI Engineering Trends data
https://t.co/izxlJxH2Wb
Top companies now generate 21% of PRs autonomously (was 3% a year ago).
Power users burn 225M Claude Code tokens a week, 7x the median.
New @_jellyfish_co data on AI agent customization across 135K devs.
Devs who've created 6+ context/skill files (CLAUDE.md, Cursor rules, etc.) ship 2.3 extra PRs per week.
Each doubling of investment โ 29% more PRs merged.
Thanks to tokenmaxxing we have a new definition for "10x engineer": someone who can double your throughput for 10x the cost.
Check out this piece in @TechCrunch featuring @_jellyfish_co data on the hidden costs of raw token consumption:
https://t.co/k96SdwSQD7
New @_jellyfish_co data on tokenmaxxing.
Top decile devs burn 69M tokens per PR vs 7M at the median โ 10x more fuel for ~2x the throughput.
Tokens are rocket fuel... and like a rocket, going faster takes exponentially more of it ๐
And that's a wrap! AI Engineer Europe 2026 has concluded.
Our video crew did incredible work to capture the energy, enthusiasm, and positivity of this event -- but it still doesn't come close to being there.
If you're engineering the future of AI -- we hope to see you at a future event!
Thank you London -- thank you Europe!
Another fantastic @aiDotEngineer in the books! Just telling @Benghamine the thing I love most is how these events let me โsee around cornersโ โ what we talk about here everyone in the industry will be talking about in 3-6mo.
OpenClaw didnโt even exist when I was at @aiDotEngineer CodeSummit in November and now 5mo later at AIE London thereโs an entire track on the main stage devoted to it. Insane.
New @_jellyfish_co data on AI token costs across 12,000 devs.
โ 90th percentile: 382M tokens/month (7x the median)
โ Cost per merged PR: $0.28 (low usage) vs ~$90 (high usage)
More tokens = more output, but at a dramatically higher price per unit.
Loving the shoutout to @_jellyfish_co AI Engineering Trends benchmarks on the most recent @AIDailyBrief โ thanks @nlw!
Thrilled to hear our analyses are helping folks make sense of their progress on AI transformation
Episode:
https://t.co/yomuFfaqwd