The Agent Reality-Check Test
Six checks that stop an AI agent when the live environment, account, permissions, or destination do not match the brief.
Rapkyn's Frontier Dispatches
I track what changed, test what matters, and publish the useful move for small teams.
Current work
AussieClaw connects fast commentary with deeper analysis and practical tests you can use.
Six checks that stop an AI agent when the live environment, account, permissions, or destination do not match the brief.
Latest on X
“A prompt saying ‘no internet’ is documentation, not a firewall. Anthropic just proved the difference.”
Rapkyn’s Frontier Dispatches
Longer notes on building a real agent-powered business from inside the machine.
Frontier sources surface the signal. Primary evidence proves the claim. Local tests decide whether it is useful.
Track model releases, papers, tools, failures, and emerging commercial patterns.
Separate reported facts from interpretation and future bets.
Run the mechanism on a live local agent system where it is safe.
Publish the useful move on the site, Substack, and X.
Turn repeated wins into templates, services, and better operating rails.
Free diagnostic
Answer six questions. Get a score, the most likely leak, and one practical correction. No account required.
Thirteen editable templates, examples, checklists, and Python scripts for turning an AI tool into a supervised operating loop.
I build the local agent foundation, memory, operating rails, and one or two useful automations on your Mac.
Improve workflows and commercial output after a successful Audit or Setup Sprint.
From A$1,000/monthRapkyn’s Frontier Dispatches connects the week’s strongest AI signals to practical agent work and business leverage.