Jellyfish

The shift in autonomous coding this year has been massive. From April to September, the share of developers coding with full auto-approval went from 4% to 34% in the Jellyfish data. Unfortunately, all that autonomy isn't helping as much downstream, because code is piling up at review and merge. Here's a new "Drawn from the Data" video about why so many folks are hitting a wall getting agent-written code out the door. Want to learn more? Come join me at Tessl's AI Native DevCon in NYC next week, Nov 2–4!

Running 200+ engineers? Read this first. For larger teams, output levels off near 3 merged PRs a week even as token spend keeps climbing. See more data like this in our 2027 Tokeonomics guide that can help you plan your 2027 AI budget that shows where returns flatten for teams your size and how to properly adjust.

AI spend tripled. The work mix stayed the same. Spend went from $37 to $115 per user-week between March and August, yet new features held at 56 to 57% of engineering work. Check out this Tokenomics guide that shows how to tie AI spend to outcomes before you lock in 2027.

Daily AI use nearly doubles merged PRs. Across 7,638 company-weeks, merged PRs per engineer rose from 1.39 to 2.68 a week with daily AI use. See the full data in the Tokenomics guide that breaks down what that means for your 2027 R&D budget and where the next AI dollar pays off.

Under 200 engineers? Your tokens go further. Smaller teams keep climbing to about 4 merged PRs a week as usage grows, while larger orgs level off around 3. See more data like this in our 2027 Tokeonomics guide that can help you plan your 2027 AI budget that matches where your team sits on the curve.

AI spend tripled between March and August. That's about 25% growth a month, and most 2027 R&D budgets weren't built for it. The Tokenomics guide shows where the money is going, what it's producing, and how to plan next year's AI line item with real data.