AI Notes — Juanje Ojeda
Notes and articles on AI agents, memory systems, learning, and complex systems — from a software engineer who studies how adaptive systems learn.
I’m a software engineer at Red Hat. On the side, I research complex systems, neuroscience, and how agents (biological, social, or artificial) learn and adapt. These notes are where those interests converge.
Start here
- Intelligence Is Beyond the LLM — If what makes a system intelligent is its capacity to adapt, then an LLM on its own is not particularly intelligent. What gives it flexibility is everything around it.
- What If Your AI Agent Could Actually Learn? — On giving local agents memory, identity, and the ability to forget.
- Why File-Based Memory Works — The ideas behind giving AI agents memory are showing up everywhere. Here’s why they work and where they come from.
- When AI Agents Interact, Nobody Controls What Happens Next — Four studies show that multi-agent AI systems produce the same patterns as teams, markets, and ant colonies.
- Outside-In Development: Spec → BDD → TDD for AI Coding Agents — Without structure, AI coding agents build the wrong thing, skip testing, and never know when to stop. A methodology that fixes this at the source.
Core Concepts
Short, self-contained reference notes on systems thinking and cognitive science:
- Complex Systems — Understanding the difference between complex and merely complicated.
- Agents — Who are the active actors of an adaptive system?
- Behavioral Flexibility — Defining intelligence as adaptability rather than raw processing power.
- Emergence — How global behaviors and properties arise from bottom-up local interactions.
- Influence vs. Control — Why we cannot dictate outcomes in complex systems, only guide adaptation.
- Self-Organization — How order and structure emerge from bottom-up local interactions.
- Strong vs. Weak Agents — Classifying agents along the spectrum of adaptability and memory.
- Stigmergy — How decentralized agents coordinate indirectly by leaving traces in their environment.
- Extended Cognition — The theory that our minds extend into the physical and digital tools we use to think.
- Four Memory Zones — The architectural structure of file-based agent memory systems.
- Maintenance Cycles — How agents distill daily episodic logs into structured semantic knowledge.
- Offline Replay — Simulating variations of past experiences to extract abstract, flexible skills.
- Progressive Disclosure — Loading knowledge on-demand to scale memory without context congestion.
- Controlled Forgetting — Why discarding old or low-value information is essential for memory to scale.
- System Memory — The enabling condition that allows systems and agents to retain history and adapt.
- Constraints as Tools — How boundaries and rules shape bottom-up self-organization.
Themes
Explore by topic: AI agents, memory, complex systems, multi-agent systems, learning, harness, emergence, methodology.
About this site
This is a digital garden, not a blog. Articles are interconnected and evolve over time. Use the graph view, backlinks, and search to explore connections between ideas.
Built with Quartz.