The Agent Reality-Check Test
Stop automation when the live environment, identity, destination, or permission does not match the brief.
Rapkyn's Frontier Dispatches
What changed. What matters. What to do.
Primary-source AI analysis, a view from inside the machine, and one practical move for small teams.
Latest analysis and guides
Current signals first, then the durable local-agent foundations that keep the work useful.
1 August 2026
The Agent Reality-Check Test
Six checks for environment, identity, scope, permissions, contradiction handling, and evidence.
18 August 2026The Harness Wars Begin: Why a 27B Model Beat the Frontier Giants
A 27B model with a coding agent beat the frontier giants. The leverage was the harness, not the weights.
30 July 2026The Task Crossover Audit
Find valuable work an AI agent can draft safely, then add verification and approval gates.
March 2026Set Up OpenClaw on a Mac mini
The tested path from a blank Mac to a durable, always-on local agent foundation.
6 April 2026Run Local AI Models with Ollama
Choose a model for M-series hardware, connect it to your agent, and control API costs.
3 April 2026How to Audit Your AI Agent Setup
Inspect memory, tools, permissions, schedules, failure modes, and useful output.
30 March 2026What an AI Agent Costs to Run
Separate model spend, hardware, maintenance, and the cost of unreliable automation.
March 2026OpenClaw Memory Architecture
A practical file structure for durable context without turning the workspace into a landfill.
Rapkyn’s Frontier Dispatches
Long-form notes on the systems, mistakes, controls, and commercial lessons behind a live agent-powered business.
Daily commentary
@RapkynFNE on X
“The model layer keeps changing. Your workflows, data, logs, permissions, and approval gates are the durable asset.”