interesting survey - confirms what I expected, almost no one is spending time in Xcode anymore.

anyone else sending photos of pages of physical books to their agents as they read?

crazy how far we’ve come in 2 years

congrats to everyone involved in shipping America.gov as someone with a pathological aversion to paperwork and bureaucracy, this makes me incredibly optimistic about the future

If you use RevenueCat paywalls, here's a repo you can use to let your agent iterate on designs locally github.com/adamwatters/rc-pa… tldr: skip the wysiwyg editor. let agent write the json directly and review it's own work with the sdk in a simulator

trolling Claude by telling him our new website needs to look good on Internet Explorer 11

interactive inline widgets in muse chat are good, but Frank is not great at backgammon (…unless he installs an engine, but then is it really him playing?)

1 year ago I thought react native wasn’t worth the added complexity. “Android users don’t pay anyway, it’ll *feel native*, and App Store review is no big deal” Now, with agents writing all the code, crazy to not use react native and make over-the-air updates.

can't decide if squishy dumpling toys are upstream or downstream of cute ai mascots for consumer tech

Details of our messy open-claw-inspired shared agent setup. For a small team with high trust, I think this is the way: - Lives on my main work Mac, which sits on my desk, always on. No annoying authentication / tool-chain issues. Anything I setup, it can use. - Everything runs through `claude -p`. Messages within a Slack thread are continuous sessions. - Directory structure like this: - /workspace * agent always runs here - /actual-project-code - Agent can write whatever scripts, channel-logs, etc it wants in the workspace directory, which means the project repo stays clean. The workspace is an insane mess, but it doesn't matter. - Also in the directory, bot code that routes message to and from Slack. Once the bare minimum was connected, pretty much entirely built out via slack messages. The agent applies changes and when necessary restarts itself.

Reporting from the ai consumer app trenches I want to start sharing some details that help us build agentic software with agentic software. Here's one: if you have a consumer facing agent in your app, give it a "feature_request" tool. If the agent wants to do something, but lacks the affordances, it can tell you about it. Also works if a user is complaining / requesting features directly. We make this transparent to the user, with a dedicated tool-result widget that shows the message sent to the team. The message goes straight to Slack. From there, usually we @ our work agent, have it find analytics events around the request, identify the underlying cause or missing code, and (often) ship an update directly from that thread. With this setup, we're pretty close to closing the product loop, at least for small stuff.

something like clawdbot in every group chat by default will be normal soon

i haven't managed to max out weekly usage for $200/month claude code - going to see if I can do it this week

telegram group chats with all my friends + clawdbot to discuss their business ideas and let claude build stuff for them on my mac studio

close button top left, "next" button top right, 20px padding under those, 2 column grid, each card should have a 50% opacity overlay, bottom of the scroll should have an opacity mask... annoying writing prompts like this, so i made a figma plugin figma.com/community/plugin/1…

yes, many backgammon apps already in the app store... but mine will be the BEST backgammon app

- randomly thought of a vscode color picker I made 6+ years ago... - devastating 1 star review from Nicolas - but otherwise fun surprise that people apparently still download and use it - feels worth it to spend an hour and just ask claude to address issue, fix old dependencies, make it better