SBCO: Self-Supervised, Verifier-Grounded Harness Optimization For Planning Agents

2026-08-10

AI Summary

Existing self-improvement methods often incur high computational costs or are difficult to apply directly in planning tasks where task competence and self-modification capabilities are misaligned. This work proposes SBCO—a verifier-guided, self-supervised optimization framework—that introduces, for the first time, a self-supervised block coordinate ascent approach into the closed-loop self-improvement of planning agents. By leveraging a fixed meta-agent and its own feedback, SBCO jointly optimizes a verifier ensemble and an execution policy, eliminating the need for expensive self-referential or population-based search mechanisms. Evaluated across two planning domains, SBCO achieves performance on par with or superior to specialized self-modification baselines while using only one-fourth to one-fifth of their computational budget, substantially improving sample and computational efficiency.

Abstract

Self-improving agents seek to reduce the human engineering effort behind AI systems by enabling them to evolve and self-improve their performance over time. Recently, methods like the Darwin Gödel Machine and the Huxley Gödel Machine have been proposed which enable open-ended, recursive self-improvement through self-reference where a coding agent edits its own code. Such self-referential self-improvement methods require that the competence required to perform the task coincides or aligns well with the competence required for self-modification which is the case for coding tasks. For domains or tasks, which do not satisfy the alignment needed, self-referential self-improvement is not available. In such cases, it is possible to adapt the above algorithms to other tasks by removing the self-referential aspect or introducing explicit self-modification of a meta-agent -- both computationally expensive, relying on population or self-modification search over many candidate agents. For planning tasks with explicit constraints, we propose a far cheaper alternative. We introduce SBCO (Self-supervised Block Coordinate Optimizer), a verifier-grounded harness optimizer in the same closed-loop, improve-from-experience family as the Gödel-machine methods, but self-supervised rather than self-referential. Given an agentic harness, SBCO learns a decomposed bank of verifiers and a harness policy via approximate block coordinate ascent, improving the agent's outputs from its own graded feedback---with a fixed meta-agent and no human labels. Across two domains SBCO matches or exceeds a customized self-modifying baseline while using 4-5.5 times less compute budget.

Problem

• self-improvement
• planning agents
• verifier-grounded
• self-supervised
• compute efficiency

Innovation

• self-supervised learning
• verifier-grounded optimization
• planning agents
• block coordinate ascent
• self-improving AI