Founder, Arroway — keep people and their AIs on the same page. Also MemberVIP, a monetization platform for adult creators.
Joined August 2012
- Tweets799
- Following56
- Followers356
- Likes6.6K
AI agents often say "done" in the same tone whether it worked or not. Ask them to end every task by saying what they ran, what came back, and what they could not check. With that attached, you can verify a "done" in a minute. Without it, you redo the work to be sure.
Ask an AI who to hire and it returns a list of names. We asked four AIs for the best GEO agency and none named sayurname. Three repeated the same agencies. Claude warned that many top lists are written by the agencies themselves. Test yours free: sayurname.com/?utm_source=x&…
Last Friday one of my AI routines was still reviewing three pull requests when another one merged them. They had woken up at the same time and neither knew the other was there. Nothing broke.
3. Start by reading. Every agent opens with what was already decided and what is half done, before it touches anything.
Test it: stop one agent halfway and see if another continues without you explaining anything. If you had to explain, one of the three is missing. The three, with details: arroway.app/en/answers/keepi… When two agents grab the same task on your team, who notices first?
A developer posted last week that they feel far more tired since AI arrived than back when they wrote code by hand. Has working with AI made you more tired or less? What changed?
The rule your team revoked is still being obeyed by your AI agent. In CLAUDE.md or AGENTS.md it has no owner, no date and no way to expire. A 15-minute check for your own file:
4. Test it. Ask the agent: list the rules you will follow in this task and tell me where each one came from. Then compare the answer with what your team decided.
A rule that changes cannot live in a file nobody reviews. In my personal Arroway project there are 40+ items and each one has the condition that ends it written down. How that works: arroway.app/en/answers/stale… Who removes the dead rule from your file?
Paste an API key into a chat with your AI and it can end up in the AI's memory, in plain text, in a file nobody audits. A 15-minute check for your own machine and your team's memory:
On my own AIs, Arroway refuses to save anything shaped like a credential. From 10 Sep to 9 Oct it refused 76 writes. 2 were a named password or token, the rest looked like tokens, and one false positive (a deploy id) had to be fixed. It is a heuristic.
It only screens what is written to Arroway, not what your other tools keep, so the check above still applies. The full check: arroway.app/en/answers/keepi… What has your team's AI memory kept that it should not?