Ryan Flynn

Father, Husband, Entrepreneur

[Ryan Flynn @iamryfly]

@MatthewBerman @EvanMoyle I've been chewing on the idea of MAAS - memory as a service. Platform agnostic, permissioned knowledge and memories that can be read or written by any LLM agent, harness, chat, etc.

[Ryan Flynn @iamryfly]

@DhravyaShah Just discovered Super Memory this week and it's exactly what I was going to start building (super poorly) on my own, to share memories across OpenClaw, Paperclip instances, Claude & ChatGPT web apps, etc. IMO, I would really lean into folks who are using OpenClaw, Hermes, etc.

[Ryan Flynn @iamryfly]

Waiting for my @claudeai plan's weekly usage to reset to put Opus 4.7 through its full paces is the adult version of sitting on the stairs on Christmas morning, salivating over the presents but your parents making you wait.

[Ryan Flynn @iamryfly]

@Jacobsklug @openclaw @Lovable OS

[Ryan Flynn @iamryfly]

OpenClaw completes overnight work autonomously for me. Built a polished dashboard. One problem: I asked it to fix a bug.

Quality gates passed. Every review scored 8+. The wrong thing was built excellently.

The fix: strategic gates at every phase that ask "is this what the user actually wants?" not just "is this good?"

[Ryan Flynn @iamryfly]

@gregisenberg Pretty sure I've heard you admit that yourself on an episode of SIP 😁

[Ryan Flynn @iamryfly]

Built a second brain for my AI assistant. 113 entries, 108 connections, graph-traversable in 2 seconds.

Markdown files as source of truth. SQLite as derived index. Every night it reviews the day's conversations and auto-ingests new knowledge.

"What do we know about X?" now triggers a search, reads entries, follows connections, and synthesizes. Not a database lookup. Reasoning over a knowledge graph.

Built with Claude Opus 4.6 on Clawdbot (OpenClaw). Two sessions, six sub-agents.

[Ryan Flynn @iamryfly]

This is the kind of Al behavior I actually want: an assistant that pushes back when something feels off, even when the request comes from its own owner.

The security instincts aren't programmed; they emerge from good system prompts and giving your agent room to think.

[Ryan Flynn @iamryfly]

Tonight I asked my OpenClaw / Clawdbot to upgrade itself to Claude Opus 4.6.

It refused.

[Ryan Flynn @iamryfly]

Then it came up with a "cleaner" approach that broke everything and I had to go to Claude Code and use the emergency rollback prompt it wrote me beforehand. Second attempt worked.

Now running Opus 4.6.

[Ryan Flynn @iamryfly]

Running an always-on AI assistant is like having a junior employee who occasionally has IT issues, but can actually fix them themselves.

Today: voice calls broke. I told Clawdbot / OpenClaw to figure it out. It found the bug (stale state surviving restarts), wrote a fix, set up monitoring to prevent it happening again, documented everything, and gave me a link to a pre-fill a form to file a GitHub issue.

Then it called me to confirm it worked.

The future is weird.

[Ryan Flynn @iamryfly]

@AlexFinn This was inspired by your videos. Question: how long did it take you to really hone in on the process running overnight being "great". Did my first one last night and thought I gave it enough context but it didn't meet expectations and spent a lot of time today fixing.

[Ryan Flynn @iamryfly]

Built an overnight autonomous work system with Clawdbot. While I sleep, it picks high-leverage items from my backlog, spawns AI workers to build the solution, monitors progress, and delivers a morning summary.

[Ryan Flynn @iamryfly]

Just extended my Clawdbot wrapup skill to auto-draft tweets after every work session. Built with Claude Code in less than 10 minutes.

[Ryan Flynn @iamryfly]

Enjoying the Red Wings game! #rocthesummer

[Ryan Flynn @iamryfly]

@caveminds FYI...your Premium Productivity Hacks download for referring other members has a bad link. No PDF

[Ryan Flynn @iamryfly]

@XfinitySupport I am still having issues getting my previous cable modem active (I tried a new modem briefly this afternoon). I chatted with someone earlier and they said it would be fixed in an hour but it's still not working.