Sign InOpen Brain
AI EngineerVideoSource Linked

Learning on the Job: The Future of Post-Training — Raymond Feng, Applied Compute

Training on a production harness could adapt an agent to real workflows, but non-replayable interactions break common RL assumptions and make feedback harder to turn into updates.

AI Engineer · Aug 1, 2026
Open Source Open MarkdownOpen JSON
Source Summary

Replayable synthetic environments support **GRPO** by running several trajectories from the same initial state, grading them, and adjusting model weights toward the better paths.

Practical Implication

For a custom agent, preserve traces and feedback at the model boundary. A **bring-your-own-harness** setup can train on the actual workflow instead of a brittle simulation, even when the training team cannot inspect its source.

Agent-Ready Context
Replayable synthetic environments support **GRPO** by running several trajectories from the same initial state, grading them, and adjusting model weights toward the better paths.

For a custom agent, preserve traces and feedback at the model boundary. A **bring-your-own-harness** setup can train on the actual workflow instead of a brittle simulation, even when the training team cannot inspect its source.

Production interactions may be non-replayable and off-policy, so conventional comparative rollouts no longer fit. **Self-distillation**, automated trace curation, and qualitative-feedback ingestion remain open research directions rather than general solutions.
Connected Context · Feed7 Judgment

This makes production traces and harness feedback potential post-training data, extending self-improvement from observability into weight updates on the real workflow. Replayable environments fit comparative RL, but live interactions often do not, so bring-your-own-harness training depends on preserving model-boundary evidence and careful curation. Self-distillation and qualitative feedback remain unresolved paths, not established substitutes.

Why Off-the-Shelf AI Doesn't Understand Money — Udi Menkes, IntuitIntuit’s verified state-action-outcome histories specify the kind of grounded production record that harness-based learning would need, while adding causal and data-quality constraints.From Signal to PR: Anatomy of a Self-Improving Agent — Jason Lopatecki, ArizeArize already uses production traces to generate investigated issues and draft fixes; this signal extends the same evidence stream toward post-training model weights.Data and Environment Curation for Post-Training LLMs — Mahesh Sathiamoorthy, Bespoke LabsBespoke Labs narrows the implied path to improvement by arguing that curated SFT may capture much of the value before teams incur the complexity of RL.Reinforcement Learning without Verifiable Rewards — Will Brown, Prime IntellectPrime Intellect addresses the same non-verifiable-work gap with mined traces, controllable simulators, and judge-derived rubrics, while highlighting reward hacking as a risk for the proposed feedback loop.
Context Map
agent#harness-engineering#agent-reliability#tool-use
Uncertainty
Production interactions may be non-replayable and off-policy, so conventional comparative rollouts no longer fit. **Self-distillation**, automated trace curation, and qualitative-feedback ingestion remain open research directions rather than general solutions.