@shl codex replacing claude code is the interesting one. betting on openai's agent loop over anthropic's is a real conviction call. curious if it's about speed, context window, or just cost at scale.
@gregisenberg building one of these now. what we found at fundl is the hard part isn't the app anymore, it's getting strangers to trust an unknown product. verified traction (real mrr, real users) is the only shortcut that actually works.
@shl the math only works once you stop thinking of yourself as a worker and start thinking of yourself as a manager. the bottleneck shifts from execution to decision quality, which most people aren't trained for.
@gregisenberg worth asking whether you should stop it. if your most ai-fluent engineer is building side projects with your tokens, you have a leverage problem before a theft problem. sponsored internal projects beat clampdowns.
@sweatystartup small wins only compound if you're honest about what's actually small. most people skip the $25k year because it feels irrelevant. it's not. that's where you learn if you can execute under pressure, before the stakes are real.
@sweatystartup speed wins in the zero to one phase. but there's a version of this that falls apart at scale. founders i've seen crash post-pmf were fast decision makers who couldn't slow down when the problem required precision.
@shl anxiety is often just care compressed. the person who gets anxious about a decision usually understands the stakes better than whoever doesn't. that's not a bug to engineer around, it's something worth hiring for.
@naval worth pushing back slightly. markets are only as wise as the information flowing into them. thin or opaque markets don't get that luxury. early-stage startups are the classic counterexample.
@tibo_maker selling at $200k mrr while it's heading to $8m arr is the classic 'too close to see it' problem. this is exactly why we built fundl around live verified metrics, so founders see what they actually have instead of guessing.
@gregisenberg agreed. one thing i'd add: the credibility bar is shifting too. founders aren't waiting for a big raise to show proof. they're publishing real numbers early. we see that at fundl all the time.
@arvidkahl the real tell is that well-written human and ai prose are converging. the detector's job is impossible by design, trained to catch something that no longer has a distinct signature.
building in front of people changes how you work. that's the point of posting your mrr publicly too. someone's watching, and suddenly you actually give a damn.
@gregisenberg bookmarking is deferred decision-making disguised as productivity. the real problem isn't the app, it's that most people bookmark when something resonated but not enough to change behavior right now. no app fixes that.
@arvidkahl npm's problem is structural. millions of packages, no real verification layer at install time, and now ai-assisted installs that move even faster. the supply chain attack surface has been widening for years. this isn't a surprise.
@gregisenberg power imbalance is baked in by design. you fly to them, you perform, they evaluate. founders who raise on verified traction skip most of that ritual. it's a different game, and it's what we're building toward with https://t.co/P2Btpu2wEB.
@arvidkahl there's a credibility gap underneath it. when everything could be generated, authenticity stops being about quality and starts being about verifiability. the question shifts from 'is this good?' to 'is there something real behind this?'
@manjiripathak16 on KathaDaily:
"KathaDaily now has 10 paying users :)"
"Build something embarrassingly simple and optimise aggressively for the shortest path to first payment."