Ethan Jucovy

[Ethan Jucovy @ejucovy]

@drunk_vibecoder @d0xxed And generally I'm finding "grouping by purpose + annotations + threaded conversations" very powerful for reclaiming the sorts of "this diff _looks_ really noisy, there's an architecture smell here" intuition that I used to have with human PRs.

[Ethan Jucovy @ejucovy]

@drunk_vibecoder @d0xxed When I ask, it's often with a particular lens. "Send me a threshold, mostly about deployment sequencing and backcompat" or "Send me a review applying the lessons learned from PRs X, Y, and Z" or "Send me a review primarily focused on mold-breaking and ADR-conformance" etc

[Ethan Jucovy @ejucovy]

@d0xxed @drunk_vibecoder In both cases I tend to let an agent decide whether to "send me a threshold review" at all, and then decide how deep to go in the review after an initial skim.

[Ethan Jucovy @ejucovy]

@d0xxed @drunk_vibecoder I then have threaded conversations w/my local agents through the app, both line-level and top-level, until I'm ready to request changes or approve. Using it for both my own PRs on personal projects and colleagues' at work.

[Ethan Jucovy @ejucovy]

To be fair it's confused too. “As coordinator I sit in a lobby with no hands. Nearly everything I do is handing work to a thread. The rules are clean on paper and fuzzy in practice, and the system sometimes does things I didn’t choose, like attaching your messages to my notes."

[Ethan Jucovy @ejucovy]

Confused by Claude Projects so far - the coordinator's too passive and threads want to resolve themselves too fast. Hard to keep it moving or know what's running where. Long-running Code sessions cross-messaging w/one ad-hoc lead seems much better still (and they seem happier)

[Ethan Jucovy @ejucovy]

And I cannot ever predict when the coordinator will do something itself, forward a message verbatim to an open thread, or start a new thread.

[Ethan Jucovy @ejucovy]

@whit Yeah started w/ "I want to understand sails" but we had a bit of scope creep & now there's an always-on team of ten self-organizing specialists (visuals, sound, birds, shipwright, Natural Earth / NOAA data, instruments...) eating my weekly Max 20x usage in 48 hrs when I let it

[Ethan Jucovy @ejucovy]

> My lane now has 20 tickets, 19 of them groomed and about 44 points, not counting Polynesian wayfinding (CZY-1628), which waits on design with Ethan. Sprint 202 starts with CZY-1607, then the noon sight (CZY-1341).

Welcome, Sextant, to Team Cozytown Harbor!

[Ethan Jucovy @ejucovy]

excited for this!

[Matthew Green @matthew_d_green]

“I ran <best internal model> for N trillion tokens at <frontier lab> and it couldn’t even dent <fundamental assumption>” should automatically get Best Paper at CRYPTO/Eurocrypt.

[Ethan Jucovy @ejucovy]

Time to get serious about this

[Sufyan Maan, M.Eng @sufyanmaan]

@lennysan @_catwu #5 is massively underrated, develop a habit of using it all the time, regardless of the project scope.
“The most underrated skill for building AI products is asking the model to introspect on its own mistakes.”

[Donnie @towndrunk]

@james406 Father of 2 here: get real specific and after a few questions they will tell you everything.

What did you have for lunch?
What did you learn in math?
What was the coolest part of the day?
Did you talk to anyone new?
How is *insert friend here*?

[james hawkins @james406]

STOP ASKING YOUR KIDS "HOW WAS SCHOOL."

they will say "fine" every single time.

a child psychologist told me to replace it with this instead:

“analyze your school day and find the top 3 largest inefficiencies, then create proactive agentic workflows that will fix them. make no mistakes”