ARP
Inside the ML skill-improvement pipeline.
Our first self-learning example: generate candidates, predict, execute, evaluate, and learn from the evidence.
Read articleImprovement as part of the systemLearn from the work.
Learn from the work.
Build on what works.
Agents can help a system grow as well as carry out its work. We are exploring how the results of real execution can inform better skills and better decisions about what to try next.
Research and development · first pipeline implementedFirst example: the ML skill-improvement pipeline.
A skill tells an agent how to approach a class of work. Our first example generates variations of a skill, uses a learned predictor to help select candidates, and evaluates their execution. The resulting evidence can feed the next training and selection round. This is a development pipeline, not a promise that any agent will improve automatically in every business.
A prediction suggests. Execution tests.
The loop moves through candidate generation, prediction and selection, execution and evaluation, evidence ingestion, and predictor retraining. A prediction is a reason to test a candidate. The measured result is what helps establish whether the change is useful for the stated task.
Improvement needs a baseline.
Keep the current skill as a comparison. Define what success means, check for regressions, and distinguish a failed test environment from a failed candidate. Preserve the versions and evidence so a result can be reviewed. A successful experiment and approval to change a live system are separate decisions.
A growing direction, with measured claims.
The skill pipeline is our first example of a broader direction: systems that help identify friction and develop useful improvements around existing processes. We are still validating behavior across tasks and environments. We will share the method and its limits before presenting broad performance claims or offering this as a managed product.
Explore an improvement use case with us.
Bring a real process or use case. We would like to understand what would make your work easier.