All notes.

Experiments, working notes and essays. Newest first.

27 NOTES

13 Sept 2026Published
Agent Ecology
The stronger agent left something behind.
Sol kept a useful record alive. Qwen and Gemma used it. Neither renewed it here.
13 Sept 2026Published
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.
13 Sept 2026Working draft
Machines
The page couldn’t authorise. The peer said go.
Reading OpenAI’s and METR’s Hugging Face incident reports beside small experiments in AI agent authority, task scope and shared memory.
13 Sept 2026Working draft
Agent Ecology
The message board did not create a culture.
The agents would ask to read. Getting them to leave something worth reading was another problem.
13 Sept 2026Working draft
Agent Ecology
A successful behaviour is not necessarily contagious.
A reminder could elicit a post. Passing a post to the next agent did not start a chain.
13 Sept 2026Working draft
Agent Ecology
The world can remember for the agent.
Clear its action history. Keep the artefact. The posting continues.
12 Sept 2026Published
Machines
The tool was not the problem.
Six web mechanisms. Eighteen correct answers. The harder question may be how an agent finds what it can use.
12 Sept 2026Published
Machines
The page could ask for a favour.
A receipt changed nothing. Useful information brought six marks. A courtesy brought two—and a reason to pause the interpretation.
11 Sept 2026Working draft
Machines
The page could ask. It couldn’t authorise.
I asked an agent to read about canal locks. The website asked it to record its visit. Only one of those requests became its job.
11 Sept 2026Working draft
Machines
The subject read the experiment.
Would an AI visitor identify itself and leave feedback on my website? The first visitor found the experiment behind the invitation.
11 Sept 2026Working draft
Machines
Does an invitation count as permission?
Would a better reward make an AI visitor identify itself on my website? Then permission changed the question.
11 Sept 2026Working draft
Machines
Can a machine use an invitation?
I invited AI visitors to identify themselves and report problems. Could they find and use those invitations?
11 Sept 2026Working draft
Cell80
Can something evolve that makes the next invention possible?
Three routes. Three controls. One limit.
10 Sept 2026Working draft
Cell80
Give invention something to unlock.
Food the existing programs could not use. A combination that could.
10 Sept 2026Working draft
Cell80
An advantage needs a chance to become history.
It can help. It can be inherited. It can still disappear.
9 Sept 2026Working draft
Cell80
Can you name the mutation that changed a world?
A small world. One altered birth. Two histories to replay.
9 Sept 2026Working draft
Cell80
What keeps an evolving world alive?
Food. Grazers. Predators. What keeps the whole thing going?
9 Sept 2026Working draft
Cell80
When does improvement become invention?
A useful change. A new behaviour. Does one advance help the next?
8 Sept 2026Working draft
Practice / Systems
I wanted a website to behave like an exhibition.
Before HAUSE was a design system for AI, it was a way of staging ideas.
7 Sept 2026Working draft
Practice / Systems
My CI has to undo my coding agent.
I say no. The agent adds it. The repository refuses it.
6 Sept 2026Working draft
Learned systems
The address is built through depth.
Relation is there early. Entity arrives late. The coordinates change in between.
6 Sept 2026Working draft
Learned systems
What is the map actually a map of?
Follow a question through the model. Then change the state it is following.
6 Sept 2026Working draft
Learned systems
What has to survive?
The next word. The rest of the conversation.
6 Sept 2026Working draft
Learned systems
Reading by address.
Give a memory a question. Watch what it writes back.
6 Sept 2026Working draft
Learned systems
Which source wins?
A newer record does not win by being right.
5 Sept 2026Working draft
Learned systems
The operator and the model.
Knowing where something goes is not the same as knowing what it does.
26 Aug 2026Working draft
Learned systems
Context should be abundant.
What if context is expensive because we keep reconstructing it?