← Agentic Systems
Agentic Systems · Reading path · 4 pieces

Put reliability in the harness

Why grounding and workflow structure often matter before fine-tuning, what changed when code owned execution order, and what a safe restart still had to prove.

  1. Most Teams Reach for Fine-Tuning Too Early

    A framework for choosing among grounding, harness optimization and fine-tuning based on the actual gap.

  2. My Agent Kept Skipping Steps. The Fix Was Taking the Order Away From It.

    The applied build that moved sequence and verification out of the model's context and into deterministic code.

  3. I Took the Order Away From the Agent. Then I Learned Order Wasn't Enough.

    The second draft: what changed when supervised repeatability had to survive overlap, interruption and restart.

  4. Building Ratchet Runtime: What a Safe Restart Has to Prove

    The technical build: six execution guarantees, effect recovery, conformance evidence and the explicit limits of a local alpha.