/uses · July 2026
The setup behind the loop.
I keep tools that help me see the work clearly, make better calls, and execute without losing the evidence. The stack changes. The job is stable.
01
Desk
The desk should disappear. It just needs to keep repos, decisions, and proof close.
02
Builders
I use models by job. One model can ship. Another can challenge the call before it hardens.
CodexInvestigation, implementation, and verification. The default place for turning a decision into working code.Claude CodeTaste, framing, careful review, and second-order risk when the shape of the work matters.GeminiLong-context compression when the evidence is too wide for one clean pass.GrokJudgment pressure. Useful for direction, scope, and finding where a plan is too soft.
03
Memory
Agent work compounds only when the next session starts from the real state.
agent-session-resumeReconstructs the actual work state across Codex, Claude Code, Antigravity, and OpenCode before the next move.KontinuoA compact checkpoint: what changed, what was verified, what remains open, and the exact next action.AgentCanvasA local evidence surface for making a repo easier for an agent and a person to inspect together.SkillsReusable operating judgment for jobs that repeat: where to look, what proof counts, and when to stop.
04
Guardrails
Speed is useful when the control points stay visible.
Second-model reviewA different model challenges consequential decisions before they ship.Focused agentsSplit research, implementation, and review by ownership, then bring the evidence back together.Runtime proofRun the route, hit the endpoint, open the simulator, inspect the workflow. A build alone is only a checkpoint.
This page changes as the setup earns or loses its place.