I build the harness so agents can own the workflow.
AI Agent Engineer · Harness Layer · Düsseldorf / Remote
marc0.dev · Writing · Leaderboard · Contact
Environment design over prompt engineering. The leverage in agentic systems isn't in the system prompt — it's in the harness around it. Tool schemas, structured feedback, validation, escalation paths. The environment talks back; that's where alignment compounds.
Human checkpoints sit at decision points and risk surfaces. The harness handles the rails. The agent owns the workflow.
Brunnfeld — multi-agent simulation testing environment-driven alignment. 19 LLM agents operate inside a deterministic game engine through structured tool calls — minimal instructions, rich environmental feedback. Methodology proof: the same patterns translate to commercial workflows.
TypeScript · Any LLM · Deterministic engine · 123★
Sales Agent — skill-based outbound automation on MCP. Pluggable CRM adapters (sqlite / HubSpot / Close / Attio / Salesforce), per-channel rate limits, never-invent-details rule, hard error stops, human-in-the-loop feedback.
HubSpot Sales Agent — autonomous sales agent. Bulk outreach, inbox classification, research-driven personalization, lead recovery. HubSpot + Gmail. Runs on any local agent harness.
Code Commander — desktop command center for managing multiple AI coding agent sessions across codebases. Multi-agent orchestration over MCP.
Integration Engineer. Real-time middleware between enterprise CRMs (Microsoft Dynamics 365) and modern frontends (Webflow, Next.js). Entity reconciliation, automated content pipelines, Redis caching, EU-pinned deployments. Production systems handling live business data daily.
AI Claude · Gemini · OpenRouter · Supabase (pgvector) · E2B
Backend Node.js · TypeScript · Vercel Serverless · Redis / Upstash · Docker · Dynamics 365 API
Frontend Next.js · Tailwind · Zod · SSE streaming · Canvas
Payments & Auth Stripe · Google OAuth · Supabase RLS




