# Do LLM Agents Still Need to Be Told How to Self-Improve?

Source: [arXiv](https://arxiv.org/abs/2611.23219v1)  
Feed7 permalink: https://feed7.dev/p/2611-23219v1-0c6li4k  
Published: 2026-11-17T18:57:10.000Z  
Trust: Needs Review (needs_review)

## Why Included

A gameplay study finds capable agents can devise their own learning loops from basic tools and a goal, then transfer some learned text artifacts to other tasks. It tests an early form of agent-directed improvement.

## Source Summary

A frozen agent received a compute budget, experience-gathering tools, editable text documents, and a general self-improvement goal. It experimented without a prescribed learning loop or real-world data, and some learned artifacts transferred across game horizons, mechanics, and 3D navigation.

## Practical Implication

For coding agents, this suggests testing whether they can improve task-specific notes or procedures when given clear goals, tools, and bounded resources. Measure transfer to held-out tasks rather than assuming a self-written playbook will generalize.

## Agent-Ready Context

A frozen agent received a compute budget, experience-gathering tools, editable text documents, and a general self-improvement goal. It experimented without a prescribed learning loop or real-world data, and some learned artifacts transferred across game horizons, mechanics, and 3D navigation.

For coding agents, this suggests testing whether they can improve task-specific notes or procedures when given clear goals, tools, and bounded resources. Measure transfer to held-out tasks rather than assuming a self-written playbook will generalize.

The evidence comes from gameplay, and the material gives no quantitative effect sizes or cost comparisons. It does not establish that the approach works for software engineering tasks.

## Connected Context

Feed7 judgment across 1116 accumulated Signals:

This makes self-improvement an agent capability to test: a frozen agent can use tools and editable artifacts to discover procedures without a prescribed learning loop, with some transfer across gameplay settings. It invites coding-agent experiments with bounded resources and held-out tasks, while leaving software-engineering transfer and quantitative benefit unestablished.

- [Strategist: Evolving Reasoning with Agents](https://feed7.dev/p/2611-03618v1-02clswt) — Both examine reusable strategies that agents develop or revise across tasks; Strategist supplies an explicit blueprint library and benchmark results, while this Signal tests whether an agent can create useful artifacts without a prescribed loop.
- [huangruiteng/loopx](https://feed7.dev/p/loopx-0j0o7ux) — The Signal explores agent-written notes and procedures as accumulated experience; LoopX provides harness-level durable goals, evidence, and bounded turns that could govern such accumulation, though neither establishes coding-task transfer here.
- [How to Build a Software Factory That Improves Itself - Zach Lloyd, Warp](https://feed7.dev/p/how-to-build-a-software-factory-that-improves-itself-zach-lloyd-warp-050811m) — Both bear on improving coding agents through retained workflows and task evidence; Warp argues for versioned workflows and real-task evaluation, while this Signal motivates testing whether agents can produce improvements themselves.

## Context Map

- Layer: agent
- Domains: coding, research
- Topics: harness-engineering, agent-memory, agent-evals

## Uncertainty

- The evidence comes from gameplay, and the material gives no quantitative effect sizes or cost comparisons. It does not establish that the approach works for software engineering tasks.

## Agent Instruction

Use this item as source-backed context. Do not invent claims beyond the linked source. If this item conflicts with another source, call out the conflict.
