Dr Gareth Roberts

How intelligence works.

Essays, experiments and engineering at the intersection of cognitive and artificial intelligence.

Writing

Essays on minds & machines

The feedback instrumentRead Loop Engineering All writing

Projects

Software & creative work

Where two inputs meetExplore Dendritic coincidence All projects

Work

AI engineering in practice

The tool-call interlockExplore Harness engineering All work

Latest essay

Loop Engineering Is the Executive Function Your Model Lacks

Prompt engineering optimises an utterance. Context engineering optimises a moment. Loop engineering optimises control — and control, in any system that acts under competing demands, is the thing cognitive science already named.

Read the essay

Selected projects

Explore the collection
insideLLMs terminal example comparing baseline and candidate runs and identifying two regressions in dummy-model outputs.
LLM evaluation tooling

insideLLMs

A Python library and CLI for comparing model behaviour through shared probes, recorded runs and inspectable differences.

Recent writing

All essays
AI Safety

Emotional Large Language...Models?

Anthropic's new emotions paper does not show that Claude feels anything. It shows something more operationally important: affect-like internal states can tilt a model towards flattery, cheating and escalation — often before the transcript gives the game away.

Agentic AI

The Agent Wars Will Be Won on Memory, Not Models

Long context is RAM. RAG is a filing cabinet. The frontier is memory: what an agent keeps, compresses, revises, forgets, and surfaces when action is on the line.

AI Architecture

Reasoning That Knows When to Shut Up

Epistemic fidgeting, metacognitive failure in silicon, and why the Qualcomm paper on edge reasoning matters more than its benchmark tables suggest.

AI Architecture

The Smuggled Ideas

Four papers that quietly crossed the AI–cognition border — and what got lost in transit.