Building an agent without starting from the model: layered-system notes
edalgomezn · reddit · 2026-09-19
Research notes from Proyecto Chatty on building an agent (Mendel): an agent isn't built by picking a good model but by designing the system that governs what it knows, remembers, can do, how it proves it did it, and when it stops. Formula: agent = model + context + memory + tools + permissions + harness + evaluation.
- Objective: define the problem, expected outcome, and which decisions stay human-controlled; ambiguous goals yield busy agents that miss the need.
- Context: progressive retrieval — start with a minimal entry point, fetch more only as the task requires; too little context causes errors and repetition, too much adds noise, cost, and contradictions.
- Memory: not everything should be kept — accumulating everything breeds contradictions and stale decisions. Distinguish original sources, derived knowledge, operational state, decisions, reusable procedures, evidence, incidents, and historical material; memory must preserve continuity, provenance, and validity.
The post also stresses that agents are not chatbots with tools: define tool boundaries, approval-gated actions, failure handling, and escalation paths.
More from coding & agent
- Pydantic Creator Benchmarks Jev vs Sonnet: 2.6x Cheaper, 3.7x Faster in 17 Lines — samuelcolvin · 2026-09-19
- GPT-6 Astra clears Geometry Dash demon level Jumper with all 3 coins — imjustnewatai · 2026-09-19
- Greg Kamradt: Humans with AI still beat AI with AI on productivity — GregKamradt · 2026-09-19
- Veteran ML engineer: Jev may push agent tool-calling back to discriminative models — multiply_matrix · 2026-09-19
- Box CEO demos Jev for instant enterprise document triage at near-zero cost — multiply_matrix · 2026-09-19
- Walrus Memory Launches Portable Memory for AI Agents Across Apps and Models — udmrzn · 2026-09-19