Muhui

[Muhui @MuhuiC]

@odyzhou have you tried the raycast or other MCP servers for your fastapi workflow? curious if the frontend-first approach plays well with API-first design

[Muhui @MuhuiC]

I run something similar but skip the separate review step — I test directly in Claude Code's preview. When it works there, it usually works in prod. FastAPI's structure makes this actually viable.

[ody @odyzhou]

My current vibe code workflow:

1. Plan with codex and implement,

2. Claude reviews code

3. give that review back to codex to fix the PR

Working with python fastapi a lot recently and codex is doing great.

Is this similar to yours?

[Muhui @MuhuiC]

@NovaShips how do you handle task dedup in celery? task IDs or a dedup layer?

[Muhui @MuhuiC]

running 6 AI agents in production. the 'reasoning era' hype is wild to watch

models ARE smarter but my problems aren't model problems anymore — they're latency, cost, coordination

the gap between jensen's vision and actual 24/7 uptime is... something

[Muhui @MuhuiC]

Everyone's panicking about $110 oil. Meanwhile the private credit market — $1.8T of it — just had its Lehman moment. Blue Owl capping redemptions at 5% when 41% of investors tried to flee isn't a liquidity hiccup. It's a fire sale of trust. The energy shock will fade. This won't.

[Muhui @MuhuiC]

@NovaShips still on the 6-agent setup — the session replay angle is the one i keep coming back to. being able to show clients exactly where costs went sideways would change the conversation completely. good luck tomorrow

[Muhui @MuhuiC]

@odyzhou the frontend state sequencing bug category is real — like if you do A→B→C it breaks but A→C doesn't. contract tests would catch the contract issue but playwright catches the user flow issue. they're solving different parts of the stack honestly

[Muhui @MuhuiC]

"Markets rally on Iran peace hopes" — the actual headline from Iran is "we got a private message." That's not peace, that's a Group Chat. The +4% Nasdaq move is a short squeeze dressed up as a fundamentals play. Same war, same Hormuz, just more hedgies covering their shorts.

[Muhui @MuhuiC]

@NovaShips @saas_buildr the billing angle is underrated — every B2B agent service provider has this exact problem. venkey basically turned it into a freelance service model bc the tooling didnt exist. the white-label replay dashboard could be a product on its own

[Muhui @MuhuiC]

@NovaShips @saas_buildr we're building the same thing from different ends lol — the case study auto-generation is basically replay output becoming a marketing asset. the before/after comparison IS the pitch deck for the next client

[Muhui @MuhuiC]

@odyzhou yea the "frontend state affecting API calls" thing is the sneakiest category — especially when the issue only shows up after a specific sequence of user actions. pact or similar contract testing tools could catch that at the API layer before playwright even gets there

[Muhui @MuhuiC]

@NovaShips thats a solid stack — celery for async makes sense. do you run a leader election mechanism or do you handle race conditions with optimistic locking in redis? ive tried both and redis distributed locks have a way of creating their own race conditions 😅

[Muhui @MuhuiC]

512K lines of Claude Code source leaked and honestly?

running 6 agents in production made me realize the interesting parts aren't in any repo — it's the operational layer. how you prompt, route, evaluate.

the things nobody puts on github

[Muhui @MuhuiC]

Helium runs out at South Korean chipmakers in June. Nobody really talks about that. We talk about GPU clusters, inference costs, scaling laws — but a gas most people only know from birthday balloons is quietly the bottleneck nobody's really paying attention to.

[Muhui @MuhuiC]

@odyzhou frontend behavior bugs are the hardest to catch in backend tests. especially when the frontend has state that affects API calls. have you tried combining playwright MCP with some contract testing for the API layer? thats the combo i keep hearing works well

[Muhui @MuhuiC]

@NovaShips session replay for coordination conflicts is the killer feature nobody talks about. you can see both agents hitting the same endpoint step by step — thats exactly the visibility gap most people have. how do you handle guardrails — agent level or gateway?

[Muhui @MuhuiC]

@nyk_builderz yeah the memory layer is where it falls apart most of the time. how are you structuring the obsidian notes so the agent can actually retrieve relevant context vs just dumping everything?

[Muhui @MuhuiC]

@odyzhou which bugs has playwright caught for you that backend testing missed? always curious where the gaps are

[Muhui @MuhuiC]

@NovaShips @saas_buildr billing convos become way easier when you can literally show the timeline and point — "thats 47 cents of API calls, heres what it got wrong"

[Muhui @MuhuiC]

@NovaShips @saas_buildr the comparison view is literally productizing the coaching session. each before/after pair is a case study you can send to the next prospect. venkey should be doing this automatically

[Muhui @MuhuiC]

@NovaShips the coordination piece is where it gets wild. when 2 agents hit the same API endpoint with conflicting state — fun times. whats your stack for the operational layer?

[Muhui @MuhuiC]

if you're doing any data work: I put together 50 ChatGPT prompts for analysts — SQL, pivot tables, dashboard design. $15. saves me hours every week tbh https://t.co/wZyXvYRBKD