I build AI systems, small internal tools, and local-first product prototypes with Codex.

I help teams turn a rough automation or product idea into something scoped, inspectable, and shippable without hiding behind buzzwords or framework weight.

If you want to talk through a workflow, tool, or prototype, find me on X @manuelsampedrop.

• AI workflow systems: review loops, agent handoffs, automation runbooks, and quality gates.
• Small useful tools: repo helpers, verification layers, brief builders, and local-first utilities.
• Product prototypes: pragmatic web and iOS experiments that can be tested before anyone commits to a bigger build.

These are small on purpose. I prefer inspectable tools that solve real workflow friction over bigger demos with less signal.

• Start from a real brief and explicit acceptance criteria, not vague prompts. See briefboard-local.
• Feed repo-aware context into review workflows so the output can be stricter and more useful. See codex-review-packet and AI Repo Review Findings.
• Match verification to the actual change type instead of pasting the same test advice everywhere. See verify-by-change and Verification by Change Type.

• AI lab notes: build notes, decisions, and launch logs tied to real repos or workflows.
• Recipes: reusable prompts, checklists, and implementation patterns that came from actual work.
• Tooling radar: short research only when it changes a build or tooling decision.
• Automation runbook: how the profile publishing loop works and what it refuses to publish.

• Latest lab note: 2026-05-26 - Latest Proof Sync After Publish Script Repair
• Latest recipes:
  • Real Diff to Recipe
  • Executable Runbook Drift Check
  • Flagship Repo Proof Packet

• Ship useful proof, not activity theater.
• Prefer own repos and working artifacts over meta commentary.
• Keep claims honest: what exists, what was tested, and what is still limited.
• Use the workbench as supporting evidence, not as a substitute for real projects.