🌉 After 9 years, I’m thrilled to be back in San Francisco for @ycombinator S24. Yesterday marked the start of the batch retreat in Sonoma, and I've already met incredible companies, batch mates, and YC partners. I'm excited and looking forward to lots of learning during this retreat - Unconferences, group sessions, and more!

I'm trying to assign one agent to oversee coding sessions across my projects. Not to write more code, but to spot stalled work, check PRs and tell me which decisions still need me. Anyone already doing this?

Finally it’s here

Landed

Fyi

Wow, he actually did it.

The new updates to the Claude Code desktop app are fantastic. It's great to see so many of the previous issues resolved. Incredible work @ClaudeDevs @claudeai

Damn, I’m barely halfway through the usage.

I’m starting to prefer using AI agents like Dots and Devin through Slack. Am I the only one? Where do you prefer talking to your agents: Slack, their own apps, or the terminal?

I’m currently exploring Cue Agents by @ManusAI and am impressed by the initial experience. Looking forward to sharing a more detailed update soon!

So far, so good with the AI subs: Claude, Codex and Devin. Getting more done, spending less, and still have plenty left on the usage limits. What’s your current setup?

Opus 5.5 Medium might quietly be the best deal in AI coding right now. High no longer feels anywhere close to “unlimited” for me. Really hoping Medium stays this generous for a while longer. Anyone else already defaulting to Medium over High?

Here for TOKEN2049/F1 🇸🇬? Visit these food places: > Ya Kun (Far East Square): Kaya toast set > Ah Tai (Maxwell Food Centre): Chicken rice > Han Kee (Amoy Street Food Centre): Sliced fish soup > Outram Park Fried Kway Teow (Hong Lim Food Centre): Char kway teow > Kok Sen (Keong Saik Rd): Big prawn hor fun > Nylon Coffee (Everton Park): White coffee

I’m sticking with ChatGPT Dots for now. As I'm juggling multiple products, I can ask it from my phone to investigate CI failures, coordinate fixes and follow through. It also flags things I’d miss, like arrival instructions for an upcoming event. Having that context across work and life is what makes it useful to me. Anyone using Muse or Grok Bot this way? How much babysitting does it need?

You can only keep 2: > ChatGPT Pro 200 > Claude Max 20x > Devin Max Which one are you cancelling, and why?

Been trying ChatGPT Dots and I’m impressed. It proactively surfaced my upcoming in-person event, including the organizer’s directions. I didn’t ask. I played around with OpenClaw quite a while ago, and this feels much closer to a real assistant. Haven’t tried Grok bot or Muse yet.

Try dot.com

Question for people who use Codex + Claude Code heavily every day: What does each one currently do better for you? Not benchmarks. Actual work. I'm finding the differences in workflow more interesting than the differences in raw model quality now.

Still surprised by how little Opus 5.5 seems to be eating into my Claude Max usage. I've been using it pretty aggressively for coding and expected the meter to move much faster. Curious if this holds once I start throwing more parallel agent work at it.

Opus 5.5 made me realise something. I'm not really comparing models anymore. I'm comparing the whole setup around them. Model quality, remote control, subagents, limits, context, how much babysitting it needs. The harness matters a lot now.

If you could only pay for ONE right now: Codex or Claude Code? After Opus 5.5, this is a much harder question for me than it was a few weeks ago.