It has begun. Or at least, I just found out about it.

UI work is the most annoying part of web development.

Formula columns are live in Rowset. Tell your agent what you want—“show me everyone I should contact this week”—and it can create the calculation itself. No spreadsheet formulas to configure. Rowset keeps the result live across filters, search, exports, REST, and MCP.

5.6-sol is indeed slow. The key is to work on 10 things at a time, so you never have to wait.

Rowset is open source and now much easier to self-host. Give an AI agent terminal access, point it at the self-hosting guide, and it can check the server, deploy the Docker Compose stack, and verify the result itself.

PGSandbox is now much easier to set up locally. Copy one prompt into your coding agent. It installs PGSandbox, configures the MCP server, and verifies the complete create → query → delete workflow. A real, isolated Postgres database—ready for agent work.

Shoutout to the PostHog team for `npx -y @posthog/wizard`. It’s one of those rare developer tools that feels genuinely magical: run one command and watch the integration come together. I use it in Rowset. This is what great developer experience feels like. 🦔

Side projects can be practical. They can make money. They can also have a personality. I gave Awesome’s 404 and empty states a little whimsy without touching the parts that need to stay clear. Revenue ≠ beige UI.

AI Steering Files now has a repo-explorer design. Search by problem, file, or use case, then open each pattern to see what belongs inside. Also new: ANALYTICS.md, a checked-in contract for events, metrics, attribution, consent, and privacy. https://ai-steering.lvtd.dev

Launching Rowset: an open-source, self-hostable dataset backend for AI agents. Give trusted agents private, structured data they can create, search, and update through MCP or REST. Use it hosted or run it on your own infrastructure. No lock-in. Your data. Your backend.

working on AI tools is a hack for people who don't like to talk to potential customers, cause you just have to ask AI Agents to try new features and then report on results, this create the LOOOOOOOOP.

using the tools you create is a huge advantage as a builder. you get so many good ideas just from using it.

figuring things out for yourself is the best thing you can do your dev journey and that includes working with AI Agents. Reading the posts people share is good, but nothing beats figuring it out yourself and coming up with your own systems

the main barrier that a developer needs to overcome when working on stuff is the feedback loop. once that is dealt with you can iterate quickly and more importantly let you agent iterate quickly.

SaaS isn’t dead. But one-size-fits-all SaaS makes less sense when AI agents can build tools around how we actually work. I don’t love most project management tools; I may just have AI build one around my workflow. Code is getting cheap. Custom software is getting normal.

I built AI Steering Files: an open-source catalog of repository steering/config files for AI coding agents.

Found a useful Python scanner: pyscn. I tried it on Awesome. My take: useful as a CI safety signal, especially when AI agents touch code. Run it in GitHub Actions as advisory first, then make the cleanest checks blocking.

used to be: coming to bed honey, just one more round in Civ6 now: just let me start another AI agent session before i go to sleep

Codex App + Greptile is :chefs-kiss:

Most AI agent failures are workflow failures. Before changing models, tighten three things: clear inputs, one source of truth, and explicit done criteria. Better process usually beats better prompting.

Distribution is a product task, not a marketing afterthought. Every new feature should ship with one demo, one use-case post, and one user outreach loop. If nobody hears about it, it doesn’t exist.

Shipping weekly beats planning quarterly. Pick one user pain, build the smallest fix, and charge for it. Revenue is the best roadmap filter.

AI agents are useful when they remove repeated decisions, not when they imitate busywork (looking at you agent dashboard builders). Start with one painful workflow, automate 80% of it, and keep a human checkpoint for the risky 20%.

I might be too harsh, but I feel sorry for people who think openclaw is useless. It is such an unlock. x.com/i/status/20...

@openclaw - @GregAGI is crazy. I just replaced Metabase in a couple of queries. Give it read_only access to one of my project's DBs and ask it to create a daily summary message. Not coding required on my part. Crazy times we live in... x.com/GregAGI/sta...

i really, desperately need to do a video on how I've set this up! god give me enough will power 🙏 x.com/rasulkireev...

finally have setup @openclaw on my VPS and now i see what's all the fuss about... it's pretty awesome. inspired by @nateliason and his @FelixCraftAI, i've setup @GregAGI to be my personal assistant.

what APIs/MCP do you wish existed?

any person who has created more than 3 projects should setup a boilerplate, to never setup anything twice. this will allow you to be super blazing fast, especially when paired with AI.

any tips on letting go? i'm having issues trusting Claude do its thing 😢