Ideas worth carrying into the work.
Long-form explanations of agents, execution, learning, and the systems that make a successful method durable.
What Rote Is
Rote gives any agent harness one way to reach the world, remember what worked, and carry the method forward as a Play.
AI Discovers. Rote Remembers.
Rote watches agents work in any harness, keeps the proven tool path, and turns it into a deterministic Play so teams reason once and reuse the result.
The Pleasure of Finding Rewards
A ground-up journey through agents, environments, rollouts, rewards, GRPO, GEPA, and the place durable execution traces occupy between acting and learning.
How a successful AI agent run becomes a reusable procedure
A successful agent run becomes reusable when its trace is recorded, extracted, packaged, validated, and versioned. Rote does all five and returns a Play URI.
Agent skills, workflows, prompts, and Plays remember different things
A prompt carries intent, a skill carries guidance, a workflow carries a drawn graph, and a Play carries the path that worked. Match the container to what must survive.
How to extract a reusable procedure from an agent trace
Extracting a procedure from an agent trace takes five operations: filter, reify, resolve, fingerprint, generate. Here is what each does and what makes a trace fail.
Agent procedure memory survives a model change only as a program
Procedure memory outlives a model change only when the procedure is a program rather than a prompt. A Play runs from any model or harness and stops when the API drifts.
Share an agent workflow with a teammate without sharing a credential
Share an agent workflow by sending a Play URI. The recipient inspects inputs, effects, and required credentials by name, then runs it with their own keys under a write guard.
The Rote Playoffs: Publish Plays for Repetitive Work
Publish as many public Plays as you can for repetitive tasks. Build methods that anyone can inspect and run.