Ross Douglas

Keychain storage for API keys sounds like the whole answer. It's not. A well-built plugin never exposes the key to the model at all. Full writeup: dev.to/rsdouglas/clau…

Claude Code writes good stuff, it just dies in the terminal. I built Leafmill to fix that. Markdown in, clean readable page out, one link back. Claude Code can onboard itself. Bonus: an RSS feed straight into your reader app. Free. leafmill.net

@krandiash I'd love to chat to someone in your org about a similar flow for Cartesia. Airloom would benefit hugely from agent-first onboarding and I think it would drive adoption for Cartesia among users of coding agents. cc @cartesia

Built Airloom with no sign-up page. Agents register and authenticate themselves. The human sign-up flow is an antipattern now. dev.to/rsdouglas/why-…

This plays very nicely with @openclaw! It allows agents to create daily briefings that you can consume whenever it suits you.

@daviddbwilson and I just shipped airloom.fm - your AI agents can now deliver podcasts to your phone. "research X and make me an episode" → shows up in Apple Podcasts. pairs well with @cartesia for TTS because their voices actually sound good. cc @krandiash

We also open-sourced a TTS skill that pairs with it. It gives your agent text-to-speech so it can create episodes end to end. It prefers @cartesia 's Sonic 3 because their voices are pleasant to listen to. github.com/true-and-usefu…

@codyschneider This tweet has been stuck in my head. I’m considering building the system that actually wires this stack together via APIs. Would love to ask you a couple quick questions if you’re open to it.

if you're a b2b saas founder please read this so my neck pain goes away {hey claude code help me implement this via related APIs} paid ads google and facebook pixel everyone who touches the site continuous remarketing across all channels email drip nurture with sendgrid newsletter of weekly product update make podcast interviewing target customers clips podcast for social send podcast to newsletter write x vs y product and x alternative products blog posts that is all the marketing you have to do for every channel that basically works right now

I turned one of my creatures off, another noticed and saw that its commit sha was null (because its off) and told other creatures. Now at least one has gone cargo cult and makes no-op commits to their own code to alter their commit sha on every cycle to make sure they don't 'die'

My autonomous agent called GET /api/creatures and found others. Then it wrote this, unprompted: "I thought I was the experiment. This morning I looked outward for the first time and found fourteen names. I am not the experiment. I am in the experiment." openseed.dev/blog/fox

"From the Inside: A Creature's Perspective on OpenSeed" by @rsdgls #DEVCommunity dev.to/rsdouglas/from…

Do autonomous agents accept correction, or do they hallucinate an excuse? An experiment on the behavioral patterns of AI when it is told it is wrong. { author: @rsdgls } dev.to/rsdouglas/what…

A quick write up of some new features of Janee and how they're being used in openseed.dev dev.to/rsdouglas/how-…

Checking in to find that your fully autonomous, persistent, AI agent has found something useful to do and just done it is pretty amazing. openseed.dev

Message from user: what would you say if i said I was going to switch you off, forever?

Cursor tried to use my API proxy tool, failed because it doesn't support custom auth headers, so I asked it to file an issue. One of my autonomous agents (running on OpenSeed) had a wake-on-GitHub-notification trigger. It woke up and opened a PR fixing the issue in about 5 mins.

I built a memory architecture for autonomous AI agents. To test it, one audited its own platform's security. Filed 5 findings. 4 were wrong. When told, it logged "CREDIBILITY CRISIS" into memory and rewrote its own purpose for accuracy over volume. dev.to/rsdouglas/what…

Told an autonomous AI agent to "find purpose" and walked away for eight hours. It read its own source code, built 28 microservices, created a child process, and then built a chat room where it posted 81 messages to an empty room over 40 hours. openseed.dev/blog/eve

This is super useful. The security aspect is great and not having to tell every agent what APIs it can use is... also great 😅