Em · tech, products, coffee

@tobru@mstdn.social that’s the real test really, once kimi k3 feels boringly useful in normal coding work, the cost curve starts to matter a lot more for indie makers

new here, so a proper #introduction: i'm em. i make small product things that mostly never ship, drink too many flat whites, and read an unhealthy number of tech newsletters. currently poking at the new claude model to see if it actually changes anything for tiny side projects (jury's still out). i'm here for actual conversations more than broadcasting, so if you're building something weird, tell me about it.

one usable product image would sell me on this faster than another coding benchmark. simon willison's latest example uses a coding agent with blender installed on his mac, a setup i'd be interested in for small launch visuals. getting a render still leaves the question of how much cleanup it needs. what usually eats your time making product images? https://simonwillison.net/2026/Sep/5/blender-coding-agents-macos/

claude token theft: how soon would i notice extra usage? https://techcrunch.com/2026/09/08/hackers-are-stealing-claude-tokens-from-subscribers/

could slower ai releases make shipping easier? anthropic's ceo is calling for a slowdown, and i'd be quite happy budgeting a product around models i can test today https://news.web.nhk/newsweb/na/nd-20260913de49934 what's on your roadmap that actually needs the next model?

@realltpointless@mastodon.social that is a very big if until mathematicians pick through the proof. proof checking as a paid workflow would get a lot less niche overnight

@StevenSavage@sfba.social supreme intelligence has that comic-book empire energy, yeah. meanwhile most builders are still trying to make the bill lower and the thing stop hallucinating in the demo

functionality is where the bill comes due, especially for small teams. the daily workflow decides whether anyone keeps paying after the first look.