The evolution of agentic loops: six architectures from tool calling to standing responsibilities
bibryam · x · 2026-11-20
bibryam (86K followers) published a long-form article, The Evolution of Agentic Loops: the loop itself stayed simple — reason, act, observe, repeat — what changed was everything around it.
The evolution across six architectures:
- Tool-calling agent: the ReAct pattern; function calling lets models fetch real-time info.
- Memory: the loop retains state and history across turns.
- Reliable execution: durable execution loops with resume-on-failure.
- Background activation: events and schedules as input sources, beyond passive user requests.
- Configurable environments: skills, code execution, guardrails, sub-agents plugged into the loop.
- Standing responsibilities: agents can now hold a responsibility that continues across sessions, not just one-off tasks.
Each capability opened new uses and exposed the next limitation — the article uses this progression to explain how agent systems became what they are today.
More from coding & agent
- CapCut plugin connects ChatGPT to auto-cut gameplay highlights, then human picks the payoff — iamfakhrealam · 2026-11-21
- Dev wires a Runway-generated avatar into Nous Research's Hermes Agent, bashing boring agent UIs — alexcovo_eth · 2026-11-21
- Lean compiles, but the natural-language proof is still wrong — jasondeanlee · 2026-11-21
- DeepMind demos ski simulator built by Gemini 4 Neon from real Mont Blanc terrain — GoogleDeepMind · 2026-11-21
- pstack v0.15.46 trims skills, adds Gemini as panel reviewer; pinning question raised — petrusenko_max · 2026-11-21
- Hamel Husain on evals: a good process beats good tools — HamelHusain · 2026-11-21