LLM navigation hub: https://bengous.github.io/IdeAs/llms.txt — llms.txt
Augustin Bengolea
I build the workflows, tools and guardrails to stay in control.
Explorations & learnings
From early discoveries to complex architectures.
- Late 2022I discover ChatGPT and the first wave of AI tools. Huge shock. I trusted them too quickly, then learned the hard way: without rules and checks, things break fast.
- 2023With GPT‑4 and Cursor, I realize the real lever isn't a "magic prompt." It's context, and how you structure a project around AI.
- 2024I become a freelancer and use AI very little day to day. That distance helps me rethink systems and rules.
- Early 2025I dive back in hard. Long sessions with Claude Code to build projects, test patterns, and find the edge of automation without losing control.
- Summer 2025I formalize a layered architecture for context. Context is what the AI "sees": rules, code, instructions, and conversation history. I add guardrails so AI stays useful without wrecking the rest.
- Autumn 2025I move to agent orchestration, overdo it (too much process, too much complexity), then cut it back hard.
- Late 2025I stabilize and apply everything to concrete projects (web app, system configuration, agents).
Introduction
Why this journal, how to read it, and the main outlines of my journey.
The Awakening
From discovery to stepping back.
The Contextual Explosion
Intensive return and first context architectures.
The Context Odyssey
From over-engineering to radical simplification.
Orchestrating the Essential
Orchestrating agents, from workflow to sub-agents.
Guardrails
Tests, rules and constraints to stay in control.
Stabilization and Test Ground
Consolidating practices and stress-testing them on real projects.
Reflections
What I keep, what I abandon, and how to work with me.
The garden
- Agents & architectures Coming soon
- Glossary & concepts