The honest test of an agent stack is whether the company building it uses it for its own ops. Our hiring emails, social drafts, and daily engineering journal all run on Claude plugins we wrote for ourselves.
Feels like the iOS-app-in-a-weekend genre still understates what's possible. Auth, cloud sync, StoreKit, Sign in with Apple, procedural art, App Store submission. One day. From a templates scaffold, no glue work.
Won the Android hackathon two weeks into Google. Asked the VP when we'd deploy the model. He gave me the look a father gives his son after the family dog has died, and started explaining OKRs. I should've talked to friends about it. I just ran.
Full piece: https://t.co/lEz1BnqdbN
Most people think the test of whether to leave a job is whether you're unhappy. The opposite is closer to true. I left StartPlaying happier than I've ever been. The real test is whether the year was worth remembering.
One hour. Bring your own coding agent. We reimburse the tokens. Paste our prompt into Claude Code if you want; that's a fine start. We score judgment, steering, and how much you ship. Hand-writing code isn't scored. That's the founding-engineer onsite.
Instinct on a "managed" agent runtime is to write defensive reactivation logic around every idle state. With Anthropic's Managed Agent, you don't. A new user.message during end_turn idle gets picked up on the next poll cycle.
Feels like the deploy-then-test order still gets a bad rap. It's bad when deploys are slow and shared. It's the right move when each build is its own isolated URL and you can throw the bad ones away.
Every Kiloforge MVP now ships with a planner-picked brand, a 1,200-word web-researched blog, an FAQ, sources, and JSON-LD schema. Generated concurrent with scaffold and design, not serial. Each ship arrives with its own front door.
Which sentences in your SKILL.md describe behavior the code never actually runs? We just found a step-4 line in ours asking for commit-message mining; the implementation only searched PRs. No error, no test, just a quietly worse output.
'Cali or bust' sounds like a gamble. It's just expected-value math with a personality: aim where the upside is nonlinear and the downside is a regular job.
A candidate vanished between interview rounds. He'd moved to Japan, where he knew exactly one person worth moving for: a woman in San Francisco who hadn't moved. Some people don't make decisions. They make leaps and let the reasons catch up.
Fuchsia (@getfuchsia) is the fastest way to get your hardware certified.
AI agents automate the grunt work that takes consultants weeks, and experts sign off.
Congrats on the launch, @togao0!
https://t.co/cYg4oOCWYX
A friend of mine texted me at 3am about what they can do to speed up their fundraise.
Obviously momentum is number one. Have you closed the new investor? Have you close the new client?
But sometimes the best thing to do is rest. Make sure that when the pitch meeting comes you're at 100%