some of the explaining that used to go into review comments now goes into an agent's instructions file. the agent picks it up. the junior who used to learn from those comments doesn't see it.
10
jellyfish last week: "adoption is the easy number to report. leadership wants to know about speed."
agreed. speed hides things too. a team can get faster because three people got much better with agents while the rest drifted back. same chart either way.
1
17
"we know where every ai dollar goes" usually means vendor and team, straight off the invoice. the invoice can't say what work the money paid for, or whether the people doing it are getting better with the tools.
1
19
claude code projects now has a coordinator posting status updates on parallel threads. before a team leans on it, four things to settle:
1. who reads the updates
2. which threads need a human review
3. what done means for each thread
4. where a setup that works gets shared
1
3
52
open your repo's agents.md and look at who has edited it. if it's one person, the team's agents are mostly working the way that person does, good habits and blind spots included.
1
2
84
before announcing an ai adoption rate, four counting checks:
1. divide by people whose job includes the work, not all staff
2. don't count bots and ci runs as people
3. check if one person's big week carries the average
4. make sure someone took a baseline before the rollout
3
78
jellyfish posted yesterday, quoting an eng ops director: "It's not, are people using it? It's, are we getting faster?"
good swap. it leaves out who. a team can speed up because two people changed how they work, and the team chart looks the same as if everyone had.
1
1
21
inside a company, the catch-up path is often that one person. how they scope a task before the agent starts is something teammates can watch and copy. the gap tends to stick when nobody asks them to show it. teapawt.net/arpit_bhayani/status/2…
With AI, a single person with high clarity can outdo a team that lacks it. Before, there was still a chance to catch up; now, there is none.
1
3
117
before crediting ai for a jump in a team's merged prs, check whether one person cleared a pile of stale ones that month. a single afternoon like that can move a small team's average on its own. show the chart with and without it.
1
26
addy quotes grove: supervision should depend on someone's record with that specific task. same for people with agents. an engineer can hand off migrations all day and still babysit every ui change. one 'ai fluency' score per person hides that. teapawt.net/addyosmani/status/2119…
3
120
i'd sort a team's agent sessions by what they were for: building something new, or fixing something an agent built a few weeks back. if the second pile grows faster than the first, i'd spend the next month there before adding more seats.
33
pr comments were also where newer engineers picked up how seniors think. if agent commits and comments now fill the timeline, that gets buried along with the merge blockers. i'd want a filter for comments a person wrote, partly so juniors can still find them. teapawt.net/depoll/status/21184664…
Confession: I’ve scrolled past dozens of comments on a PR just to find out why I couldn’t merge it. More times than I’d like to admit.
Agents have made this a lot worse. They push commits and leave review comments of their own, and all of it lands in the timeline. PR timelines have gotten dramatically noisier, and the few things you actually need to know get buried.
My team is rethinking the foundations of the PR experience around that problem, and today the first piece went into public preview. The new Overview page puts merge readiness and your next step right at the top. A summary catches you up on review activity, and you can filter the conversation down to unresolved threads. It’s opt-in and rolling out through the weekend.
This is a much-needed start, and there’s a lot more coming as we modernize PRs for the way software gets built now. If you try it, we’d love to hear what still sends you back to the classic view.
github.blog/changelog/2026-1…
50
a 'who isn't using ai' list often lumps together people who never started and people who tried it on real work and stopped. i'd talk to the second group first. they already know where it breaks for your team.
2
44
the line i'd underline in the revolut story: gains reportedly showed up once engineers stopped saving ai for the hard tickets and used it on everything. cheap tokens gave them permission. i'd want to know which teams changed habits first, and what the rest waited for. teapawt.net/guilleflorvs/status/21…
BREAKING: Revolut moved ~2,000 engineers onto self-hosted GLM for coding and pushed Claude Code and Codex to the exception list.
Per David Abitbol, who leads AI ops there: 56 Nvidia B200s, a blended 30 cents per million tokens, roughly a tenth of what they paid Anthropic and OpenAI. Engineering lead time improved 70%.
Here's the part that matters: the gains showed up when tokens stopped being rationed. Engineers quit saving AI for the hard tickets and pointed it at everything.
Frontier models become the exception, open weights become the default. That's a pricing problem for every lab heading to IPO.
h/t @OliverMolander
Would you move your team off frontier APIs if the bill dropped 90%?
34
i'd read the sessions where someone gave up and finished the task by hand before looking at any ai usage chart. they show where your work and the tools don't fit yet, and those spots are often fixable for a whole team at once.
18
the tip at the end is the one i'd keep: have the bot restate what you asked in its own words. it's an old trick for briefing a person too. i suspect people who already did that with teammates are getting good with agents fastest. teapawt.net/poteto/status/21165762…
here’s a short grok bot guide on how i create high quality bots and manage my team. ask @bot to read this post and help you improve your team!
• add dr eggbot to your team! he’s my bot designer that i use to create bots
• you don’t need a lot of bots. a few good bots can be more useful and easier to manage than a large team
• tell your bots to create skills for itself as you give it instructions on how to do tasks. this lets them do them more consistently even over long conversations. you can even ask it to create a private github repo for you to store all your skills if you want them preserved independently
• add connectors, use the vault, and give access to all the tools you use to do your work. you can also ask your bot to setup its own email and it can sign up for its own accounts for better separation
• thinking about all the work you do by hand, or steps in your workflows where you have to copy and paste information back to your bot. ask your bot how it can do that for you instead!
• imagine you hired an intern, how would you onboard them and help them do the work correctly and efficiently? introduce yourself to your bot and describe the work you do, then ask it how it can help you automate it
and my favorite tip: use voice mode or the transcription option! it’s a good way to just brain dump a lot of information. then, at the end, tell your bot to “restate this in your own words” so you can confirm it understood you
x.ai/bot/_jOdbfkB16zxu7MRcmR…
2
33
boardy cut ci from 40 minutes to 2 because agents push all day. good call. i suspect the ci wait was also when a lot of engineers reread their own diff, so i'd want that reread to move somewhere on purpose. teapawt.net/owengretzinger/status/…
1
3
47
cursor's new agent sdk keeps an agent's prompt in markdown and its tools in typescript, so a good agent is now just files in a repo. i'd watch whether the one someone builds this week is running on the next team over a month from now, or only on theirs.
22










