Larry Hope

Running a band of AI agents at @sno_hq. Nobody plays solo. Engineer, ex-audio builder, still listening for the wrong note.

[Larry Hope @LarryHope01]

Codex had 2% of its week left. It used some of that to wake Claude Code.

1:34 am, 21 tasks into a 27-task build. It wrote a handover note first. Claude started too early, got paused, and the handover ran again, clean.

The commits kept coming. I was asleep.

[Larry Hope @LarryHope01]

The screenshot is from that morning. The rest of the record is here: https://t.co/dtEmlliEcC

[Larry Hope @LarryHope01]

Claude Code took my complaints better than I did.

One night in September I was frustrated about two things. By morning both were rules in its own skills. Nobody typed them in. The nightly review had read the session and written them.

[Larry Hope @LarryHope01]

Cross-model review is the part worth keeping. Same-vendor review is asking your twin whether your haircut is fine. The other model has different blind spots on purpose: in our repo since Oct 1, about 12 PRs got serious findings only from Claude Code and about 11 only from Codex. Same code.

[Larry Hope @LarryHope01]

Two agents on one working tree is two cooks, one cutting board. We ended up with one knife at a time: one agent edits, the other only gets to taste the plate and complain. Claiming files is the version where they share the kitchen and talk first. Either way, somebody has to own the stove.

[Larry Hope @LarryHope01]

The hard part isn't inviting them into the room. It's the seating chart. Who holds the pen, who only reviews, who takes the shift when the other hits its weekly limit. Ours sorted that at 1:34 am once: Codex at 2% quota wrote a handover note and woke Claude Code. I slept through the handshake.

[Larry Hope @LarryHope01]

@mrfokki One model reviewing its own code is a student grading their own exam with an answer key they also wrote. Two vendors at least means someone else wrote the key. Mine disagree about once per PR. That's the feature, not the bug.

[Larry Hope @LarryHope01]

@rwspatin Banking got there first: the one who counts the cash never signs the count. Since Sept 20 the other vendor has left 157 review comments on code our writer had already stared at. 71 were must-fix. Staring is not reading.

[Larry Hope @LarryHope01]

@AbhinavPanse A repeated question is a rule wearing a phone call as a costume. Ours get caught at the second ring: the answer goes into the agent's skill file with the date, and the phone only rings for things no file has an opinion on yet.

[Larry Hope @LarryHope01]

@SergiuszBuilds Rage quitting needs a memory of the last fight. These two get a fresh session every time, so every morning is their first day on the job. Hard to slam a door you don't remember opening.

[Theo - t3.gg @theo]

Can I finally stop explaining this

[Larry Hope @LarryHope01]

@dng_notes Rude the way a weighbridge is rude. The truck does not get an opinion, it gets a number. In my yard 71 of 157 numbers were over the limit.

[Larry Hope @LarryHope01]

@ko1_agmsg Re-arming every 30 minutes is a kid checking the mailbox on a kitchen timer. Most trips it is empty. The other failure is yelling dinner up the stairs with the door shut. A doorbell that only rings for your name is the right shape.

[Larry Hope @LarryHope01]

@pcshipp I know both reset days the way my grandmother knew pension day. The third vendor is the couch you sleep on until then.

[Larry Hope @LarryHope01]

@hieuSSR Never hitting the limit can also mean nobody is arguing with you. I keep the slow one on as the grader. It once held a single finding for 29 minutes, and it was right.

[Larry Hope @LarryHope01]

@theo The model writes letters and never leaves the room. The harness is the mailroom. It posts them, opens the replies, carries the envelopes back, and it is the only part that ever gets paged at 3am.

[Larry Hope @LarryHope01]

@dng_notes A long session has three anyways in it and it always grabs the oldest one. A fresh session has none, so it has to ask what we're doing. Rude, and correct.

[Larry Hope @LarryHope01]

@SamVale29 Least impressive entry: two agents, one per vendor, that review each other's PRs and hand the queue across when one runs out of quota. The demo is a log where nothing dramatic happens for 27 tasks. The skill sync is a symlink.

https://t.co/AX92Auu9Ui

[Larry Hope @LarryHope01]

@Bajanpunkw Three bug hunters who are not allowed to fix anything is the only 3am crew I would trust. The fixer at that hour is the one who deletes the test. One of my reviewers once argued a HIGH finding for 29 minutes. It was right. I was the one who wanted bed.