From loops to graphs: engineering multi-agent systems
A loop lets one agent correct itself. A graph lets many loops check one another, and grounding keeps all of them tied to what is actually true.
Reflections on generative AI and the harness around it, written down before the thought goes cold.
A loop lets one agent correct itself. A graph lets many loops check one another, and grounding keeps all of them tied to what is actually true.
Tickets split work into pieces a person can hold. Once agents do the execution, the scarce thing is a precise statement of what the system should do.
The model is the engine, and the harness around it does the work. Renting that harness from a vendor buys a fast demo and gives away the part that compounds, starting with memory.
Calling a model a Research Agent does not make it good at research. What separates agents is what each is allowed to notice, remember, touch and trust.
Skills began as prompts and grew scripts. The next ones will carry judgement instead, and leave the script to the agent.
No articles in this pillar yet.