Brian McQuay

Builder · Serial Entrepreneur · Leader.

25+ years of building software, launching companies, and chasing whatever sparks curiosity next.

[Brian McQuay @onomojo]

Programming languages were built for human comprehension. Assembly over binary. Python over assembly. Each abstraction layer made it easier for us to think. LLMs don't need that. They can work at any abstraction level. We're about to see language extinction and rebirth.

[Brian McQuay @onomojo]

Fable has outperformed all other models at flattery. No other model talks to you like your the smartest person in the room. It loves to praise your insights and unique perspectives. I suspect that they trained it on flattery and that's partially why so people love it so much.

[Brian McQuay @onomojo]

@jun_song The sessions are all just files in the filesystem aren't they

[Brian McQuay @onomojo]

I had to use Claude to fix my Hermes agent. It wasn't communicating with Slack properly and couldn't figure itself out. I'll blame Deepseek V4 Flash on this one and give Hermes a pass.

[Brian McQuay @onomojo]

@tonysimons_ I'm really just exploring how coherent Chinese models are over long planning sessions spinning days. And how well Hermes handles everything.

[Brian McQuay @onomojo]

Remote work's hidden cost

20+ years remote: the biggest cost isn't productivity—it's network.

Office workers build relationships by default. Remote workers have to engineer them.

Text < voice < video < face-to-face. Each level unlocks more empathy, connection, value.

[Brian McQuay @onomojo]

@0xSero You're going to need a mini fusion reactor soon

[Brian McQuay @onomojo]

@0xSero Most Americans dislike AI?

[Brian McQuay @onomojo]

He isn't wrong.

[0xSero @0xSero]

https://t.co/mCKYSBtzne

[Brian McQuay @onomojo]

@0xSero I'd love a copy

[Brian McQuay @onomojo]

Token churn

Copying your eng workflow into agentic systems will bankrupt your token budget. Code review loops that work for humans = expensive LLM calls on every iteration. The fix isn't a new tool—it's rethinking the loop itself.

[Brian McQuay @onomojo]

@BarathAnandan7 It's too expensive for how bad at complex tasks it is. We already have great basic coding models so I was hoping for better reasoning at more complex tasks. My experience with it is similar to yours. It fails quickly when things get complex.

[Brian McQuay @onomojo]

@0xSero I was just exploring this model this weekend. It surprised me how smart it seems for the price.

[Brian McQuay @onomojo]

@0xSero Doesn't matter when the output is going the opposite direction

[Brian McQuay @onomojo]

@0xSero I still route well specked out work to sonnet. It's great at cheap execution if you do all the planning upfront in better models

[Brian McQuay @onomojo]

@0xSero That last one was hilarious

[Brian McQuay @onomojo]

Automating fixes for flaky tests

Flaky tests that fail 50% of the time aren't tests—they're coin flips that waste CI cycles. The fix: detect patterns, triage root cause, auto-generate PRs for simple cases. Automating the boring parts lets us focus on the hard problems.

[Brian McQuay @onomojo]

I like this. My dark factory would need to have something more sustainable like only running while you're actively walking or cycling.

[Om Patel @om_patel5]

SOMEONE BUILT A CLAUDE CODE PLUGIN THAT REFUSES TO WORK UNTIL YOU DO PUSH-UPS OR SQUATS

its called workout gate and it is exactly as cursed as it sounds

when the hook fires, it blocks your prompt, opens your webcam, and makes you exercise to keep coding

> it counts your reps live through the webcam with mediapipe, no honor system rounding down
> only releases your prompt once you actually finish the reps
> quit halfway and the leftover reps get saved as debt for your next session
> chill, demo and hardcore presets, plus prompt based, time based or random triggers
> tracks stats, streaks and personal records right in the claude code statusline

we automated the coding, now we're automating the part where we actually take care of ourselves

[Brian McQuay @onomojo]

Glad it wasn't just me. Things were seriously off today.

[Conor Dart @Conor_D_Dart]

When Opus 4.8 first came out, I thought it was a really good model.

Today though, it feels like it's missing basic instructions and doing things that just don't make sense.

I don't know if I've somehow been hit by the Fable 5 bug, or if something else is going on, but the experience feels noticeably different from when it launched.

With Fable 5 currently unavailable, I'd have expected Anthropic to have plenty of compute available for Opus.

[Brian McQuay @onomojo]

@Conor_D_Dart I was thinking this all day long. It would stop and ask questions while in auto mode. It would churn for 30 minutes on some simply task. I saw this on multiple apps and different accounts. Something definitely felt off today.