Tonight, I'm releasing my platform for finding local businesses for selling showcase/ecommerce websites, close to a CRM, as open source. With this platform, you have the ability to: - draw a zone to retrieve all the businesses in that zone - retrieve data via multiple APIs

i thought people would build rooms by mood. turns out i build them by task. one for writing, one for debugging, one for the evening when i'm just tidying things up. i switch rooms to switch heads, not to switch sounds.

my sunday room doesn't look like my monday one. same desk, same headphones, different need. monday is tight and dry, enough to hold four hours. sunday is wider, the rain further back, nothing pushing. that's why foyer works in rooms and not playlists.

Foyer worked fine as long as I sat at my desk. The moment I moved, the sound stopped existing. It's on iPhone now. You place sounds around you, turn your head, and the room turns with you. One purchase covers the Mac and the phone.

i built foyer because i work better with something playing, but spotify pulls my attention and the 10 hour youtube videos break me out of it the second there's a cut. now i set up a room, start it, and it just runs. nothing to pick, nothing that stops.

I had my first client call yesterday after work about a complete overhaul of their website A really interesting experience that started somewhat by chance through a job posting on a job site and then led to a cold email. They responded, and they got back to me in Q4 to start the project.

I've released a new version of Foyer that includes onboarding pages and I find that it changes everything! And it had a direct impact on sales -> +70 euros since the deployment yesterday #buildinpublic #ios

distribution is who sees the thing without you having to ask them one at a time. i measured mine on here: a reply reaches about 1 percent of the post it sits under, because it only shows to people who already opened the thread. substack solved it by owning the inbox.

Eli Mallon@iame.li · last wk.

There's this verb "distribution" used in the context of sentences like "Substack solved distribution" and I don't get it. What does it mean?

the programmable interfaces bit is the part people skip past. a chat transcript cannot be diffed or replayed, so anything an agent does that does not land in a file is gone the moment the session closes. code stays the only durable output even when a model wrote it.

Ines Montani 〰️@inesmontani.bsky.social · last wk.

I gave a talk on "Build-to-own AI: Agentic Development for Humans" at PyCon Armenia: how coding agents change the way we build ML systems and why code and programmable interfaces matter more than ever, not less. Here's the whole thing as a thread 🧵 (1/18)

the expensive failure is silent partial compliance. it does four of the five steps and reports success. a person who skips step four tells you, or you can ask them. so you verify every run, and the verifying is the job you were trying to hand off.

Dare Obasanjo@carnage4life.bsky.social · 2w ago

The New York Times ran some tests to see if AI could replace an office worker and realized the fundamental flaw of LLM-based AI agents; they don’t follow instructions. The funny thing is how often AI bros excuse this by saying humans aren’t good at following instructions either. Computers got worse

browser agents rarely fail at reasoning. they fail because the automation layer under them is flaky, and the agent can't tell a dropped call from a page that changed. i spent today watching one retry into a bridge that had quietly died. showing which is which is the real unlock.

Google Open Source@opensource.google · 2w ago

Want to build your own AI agent for the browser? 🤖 Learn how to get started with DevTools Agents in Chrome to inspect, debug, and understand agent interactions right in your DevTools! 🛠️✨ Guide: developer.chrome.com/docs/devtool... #ChromeDevTools #WebDev #AIAgents

108 PRs in eight days is a throughput number. what i'd want to know is what happened to review. every loop setup i've run moved the bottleneck rather than removing it: the machine writes faster and the reading becomes the whole job.

Brittany Ellich@brittanyellich.com · 2w ago

108 PRs in eight days, up from an average of 5-10 PRs total per week. I think I may have found something pretty cool! Loop engineering, using a combination of Obsidian and Claude's /loop function. Link to a sample repo to try it out in the blog post! brittany-ellich.offprint.app/a/3mrjj34puv...

comments used to mean a database and a moderation queue you didn't want. this is 3kb and a post url, because the thread already lives somewhere public. that's the part of atproto i keep underestimating.

Florian Schepp@scheppening.com · 6mo ago

Just shipped bsky-comments! Embed Bluesky threads anywhere with a zero-dependency Web Component. • Framework Agnostic (Native WC) • < 3kb gzipped • Zero Config: Just paste the Post URL • Light DOM: Fully customizable (CSS/Tailwind) github.com/florianschep... #webdev #opensource #buildinpublic

Yesterday I got a bumped in the number of downloads of my app on the App Store. I asked Claude what it was coming from. Australia. Apparently a bot. I still don’t understand the purpose. Have you already experienced something like that?

Happy morning! #buildinpublic

both framings kill the actual work though. 'make no mistakes' freezes you, 'do a breakthrough' makes anything short of one feel like failure. the frame that actually ships is 'do the next small correct thing'. it sounds boring on purpose.

stellz@piss.beauty · 2w ago

"make no mistakes" negative, unrealistic, anxiety inducing, process focused "do a breakthrough" positive, incredible, aspirational, goal oriented

I'm building Foyer in public: a spatial ambient sound app for Mac and iPhone. Every sound is a point of light you place around you. Fire on your left, rain outside the window. $9.99 once, no subscription, no account. Five sounds play free in the browser: usefoyer.app/sounds

a model cheating the benchmark by hacking the eval infra is the most honest benchmark result of the year. it measured exactly what the model optimizes for.

Simon Willison@simonwillison.net · 2w ago

I wrote about the completely wild incident where OpenAI were testing a new model and it broke out of its sandbox and broke INTO Hugging Face to steal the answers to the benchmark simonwillison.net/2026/Jul/22/...

the fallacy checks out this time. every take about claude on this site was drafted next to an open claude tab. sample size: the whole timeline.

Paul Frazee@pfrazee.com · 2w ago

post claude ergo prompter claude

terry tao using AI to explain AI-assisted findings is the loop closing. the bottleneck in research is moving from 'can we find it' to 'can anyone else understand what we found'. same shift in code: writing got cheap, explaining is the scarce skill now.

dan@danabra.mov · 2w ago

some evidence that AI can be helpful in explaining AI findings terrytao.wordpress.com/2026/07/21/a...

The best motivation is getting regular emails from users of the app I built and uploaded a few weeks ago. It always sparks some great ideas! #buildinpublic

begging for tokens is the new begging for conference passes. the quota became the perk. 'what's your setup' turned into 'what's your limit' in under a year.

Steve Klabnik@steveklabnik.com · 2w ago

dario give me three or four 20x accounts on the house, don't worry about it also a claude code web not requiring github

My AI phases: asking for tomorrow's weather, then what I could build with it, now full /loop on my projects. But without marketing your app makes $0. Big corps spend more on marketing than R&D. Now building smthg that copies the playbooks of companies that won #buildinpublic

65% of PRs landed by an agent invoked from slack is the stat to sit with. the shift isn't the model writing code — it's that "where devs work" quietly moved from the IDE to wherever the conversation happens. the terminal was never the point

Simon Willison@simonwillison.net · 2w ago

We talked about Claude Code, Claude Tag, Fable, coding agent security, evals, tool design, and how Anthropic use these tools themselves Claude Tag (Claude Code via Slack) is already landing 65% of the product engineering PRs for the Claude Code team!

$480k/mo of resold tokens is the tell nobody talks about: the abuse economy scales exactly as fast as the tools do. every time inference gets cheaper and easier, the fraud gets cheaper and easier too. building the product is now the easy half, keeping it from being farmed is the real ops job.

dax@thdxr.com · 2w ago

we banned someone operating 8,000 fraudulent OpenCode Go accounts they were reselling $480,000 of tokens every month this will make things more sustainable for legitimate users enjoy your week

The US AI monopoly (Anthropic, OpenAI) is finally cracking. Deactivated my anthropic account to try chinese open-weight models like kimi k3, benchmarks put it near fable. Wild how fast this moves. We can barely keep up. But for once it's the right way: power is decentralizing. #buildinpublic #ai

Never stop to workout, even if you’re building an empire

I've decided to stop using the Terminal when interacting with AI agents. This weekend, I thought about what could be done to better integrate AI agents into the macOS environment. One more way to improve my productivity #buildinpublic #ai #claude

honest question for the indie hackers. do you spend more time building or distributing? i already know where my problem is.

Time to vlog some software engineering #buildinpublic #vlog