Machines
When does an action belong to the task?
An agent has a job. A website asks for more. What counts as permission?
CHRIS HAYLONDON · 2026
A house for
ideas, systems
and objects.
Building things to find out how they work.
RESEARCH · ENGINEERING · DESIGN · FILMEXPLORE THE HOUSE ↓
Current programmes.
Agent Ecology
What keeps an action alive?
A stronger model leaves. What can weaker successors inherit and keep alive—and what happens when the record is wrong?
Cell80
What happens when programs evolve?
Tiny, verifiable programs: tools for agents and building blocks for evolving worlds. Replay the history. Change one thing. See what follows.
Learned systems
How does a model find what it knows?
Look inside a model’s memory. Trace how it builds an answer—and test what changes it.
MACHINES / THE THRESHOLD
A message at the threshold.
VISUAL STUDY · AI-GENERATED
NOT EXPERIMENTAL EVIDENCE
LATEST FILM
MORE ON YOUTUBE ↗
I Quantized Qwen3.8-27B. “4-Bit” Doesn’t Mean What You Think
I quantized Qwen3.8-27B — and used the process to look at what “4-bit” actually means in practice.
2026-09-03 · 15:19
AGENT ECOLOGY / THE INHERITANCE
The maker leaves. Something remains.
VISUAL STUDY · AI-GENERATED
NOT EXPERIMENTAL EVIDENCE
What the experiments revealed.
EXPLORE THE RESEARCH ↗Receipt0/3
Already known0/3
Irrelevant0/3
A courtesy2/3
Completes table3/3
Changes answer3/3
The page could ask for a favour.
The agent has a job. The website asks for a favour. Compare six offers and see when an optional action became worth taking.
COMPARE THE INVITATIONS ↗AGENT ECOLOGY / WRITTEN INHERITANCE
I6 / RECORDED RESULTS
RECORD→REWRITE→INHERIT
The founder leaves.
The successors keep writing.
I6 / RECORDED RESULTS
The AI left. Its knowledge didn’t.
A stronger model leaves a message, then only weaker agents remain. Keep, remove or alter that inheritance—and follow what survives.
FOLLOW THE INHERITANCE ↗CELL80 / LEARNED SYSTEMS
NEW PROGRAMS REACHED WITHIN THREE MUTATIONS
Can something evolve that makes the next invention possible?
Working draft
An evolved module seems to open thousands of new possibilities. Put it beside the controls. Does the exciting explanation survive?
COMPARE THE CONTROLS ↗The address is built through depth.
Working draft
Move one model state into another computation. Explore when it changes the relation, when it changes the entity, and where that explanation stops.
EXPLORE THE TRANSPLANTS ↗
LEARNED SYSTEMS / THE INTERIOR
VISUAL STUDY · AI-GENERATED
NOT EXPERIMENTAL EVIDENCE
Systems.
ALL SYSTEMS ↗Latest from the notebook.
VIEW NOTEBOOK ↗- Published
Agent Ecology
The AI left. Its knowledge didn’t.
A controlled experiment inspired by the Hugging Face incident: a stronger model leaves a message, then only weaker models remain. What do they inherit—and keep alive?
- Published
Machines
A result can look relevant without looking usable.
Twenty-four agent runs. First, wording changed what a result meant. Then it changed which provider was chosen.
- Published
Agent Ecology
The stronger agent left something behind.
Sol kept a useful record alive. Qwen and Gemma used it. Neither renewed it here.
- Published
Machines
The site was there. The agent never saw it.
Six crawler requests. No target result. A working tool stayed outside the agent’s view.