Blog

the-self-improving-stack

An editorial still life about how an agent system improves

The Self-Improving Stack: How AI Agents Get Better

The self-improving stack is the set of agent layers that can change, from prompts and skills to runtime, traces, evaluation, model training, and governance.

An editorial still life about how an agent system improves

Prompt Optimization for AI Agents: One Layer of a Bigger System

Prompt optimization searches instructions, examples, schemas, tool descriptions, and rubrics for better agent behavior while the model and runtime stay fixed.

An editorial still life about how an agent system improves

Skill Optimization for AI Agents: Training Durable Procedures

Skill optimization improves reusable agent procedures and their activation rules while keeping model weights fixed.

An editorial still life about how an agent system improves

Optimization Theory for Agent Builders

Optimization theory for agent builders is a way to choose the right mutable surface, compare candidates fairly, and keep noisy improvements from becoming production regressions.

An editorial still life about how an agent system improves

Post-Training Agents: When to Change the Model

For teams building post-training agents, changing model weights or adapters moves improvement from prompts and skills into the policy that acts across future contexts.

An editorial still life about how an agent system improves

Agent Memory: Why Retrieval Is Not Learning

Agent memory becomes a learning loop only when a past run creates a scoped, supported write, the right future task retrieves it, and evaluation proves that behavior improved.

An editorial still life about how an agent system improves

Evaluation Gates: The Rule That Decides Whether an Agent Improves

Evaluation gates turn agent evaluations into release decisions by comparing a candidate with a baseline on protected tasks, deterministic checks, cost limits, and trace evidence.

An editorial still life about how an agent system improves

Agent Traces: The Evidence an Improving System Needs

Agent traces preserve model calls, tool actions, observations, artifacts, costs, and outcomes so an improvement loop can diagnose a failure instead of tuning a score alone.

An editorial still life about how an agent system improves

Test-Time Compute for Agents: Beat Random at Equal Cost

Test-time compute gives an agent extra samples, branches, retries, or verification, but an execution-shape claim matters only after it beats a simple equal-budget baseline.

An editorial still life about how an agent system improves

Harness Evolution for AI Agents: When Prompts Plateau

Harness evolution changes the execution software around an agent when prompt and skill tuning cannot create the missing tools, isolation, checks, traces, or candidate lifecycle.

An editorial still life about how an agent system improves

Agent Runtime Topology: How Execution Shape Changes Agent Behavior

Agent runtime topology determines whether a request for parallel work, review, retries, cancellation, and evidence becomes an executed workflow or remains a sentence in a prompt.

An editorial still life about how an agent system improves

Multi-Agent Coordination: Roles Are Not Structure

Multi-agent coordination becomes real when roles have contracts, permissions, state boundaries, budgets, checks, and traces, and when the system beats a single-agent baseline at equal compute.

An editorial still life about how an agent system improves

Agent Governance: How to Govern Self-Improving Agents

Agent governance gives self-improving systems owners, authority limits, evidence requirements, approval points, and rollback paths before mutable behavior reaches users.