Part of Temporal Constructs Appendix, which contains details behind the research: dates, quotations, descriptions of instruments, and sources.
General definition
The change in what the professional doctorate could attempt after the arrival, in April 2026, of an open-source agent framework that can build its own tools on hardware the artist owns. A plan written in 2025 could not have anticipated it.
Relation to the practice
The acceleration is the unintentional change of the first year, and it is recorded as an observation. For the doctorate the acceleration is both a condition and a finding. It is why the plan’s modes were worked in an order the plan did not draw, and it is evidence of the object of study moving while the study was underway. That last point is a matter for the review.
Within weeks of the arrival the agent was porting HMSL, writing KML into Trip-A-Tron, keeping a router of thirteen MIDI ports alive, and reconstructing a codebase that it had itself deleted. Assistance to a technical year became the platform the practice builds on, most visibly in CONDUCTOR, where a set of scripts became a laboratory with a command language, sixteen modules, an audit trail and a written design.
Capability made available
The possibility was specific, and none of it existed in the plan. An open-source framework that runs on machines the artist owns can be given a task and left to write the tools that the task needs. The task may be a language model writing code into a period editor, or a router opening every port of a synthesiser and a mixer. It may be a watch on the traffic of a lighting rig. It may be a port of a 1980s language to current hardware.
I’m personally very excited about this. I’ve been working with these legacy systems for a long time, and while I love them, it always took me ages before I could actually experience the initial vision I had when starting out. Now I can express what I want to achieve, and experience it and reflect on it within the same hour instead of a week later. This transformed my creative process to something much closer to what my audience experiences when they encounter my installations.
artist statement
It may be a loop in which an agent composes in KML, reads back the picture it produced, and tries again. It may be the interfaces through which the artist works with the agents themselves. The framework is open and the hardware is owned, so the practice is not renting its way of working from a supplier. The models can run locally, so the whole arrangement can be operated at a cost that a doctoral practice can carry. The same arrival produced the urgency that accompanies it. A capability can be changed or withdrawn by whoever provides it, and this is a reason to build a system that one keeps.
Change inside the archive
This archive is an instance of the change as well as a description of it. The archive was written and populated in about two days in November 2026, and every link in it resolved. Its pages were built from nine dated monthly accounts, the year’s activity log, and the owner’s own statements. Two facts about that rate are worth stating exactly.
The writing and the assembly were done by the same arrangement that the doctorate studies, with the artist deciding what was true and checking what was claimed. The rate is possible only because of the rule that the year established, that a report is not evidence until it has been compared with its source. That rule caught the fluent errors recorded on the agent’s page, and it is what allows a record to be produced at this speed without becoming worthless.
The limit matters as much as the rate. What took two days was the account of ten months of work, January to October 2026, and the accounts it draws on were themselves produced month by month while the building happened. The coding of the session history cannot be compressed in the same way, because it is work done case by case. The rehearsal of a live demonstration is the one part that is calendar time.
Reason for the record
A doctoral project about how creative agency is shared with artificial intelligence (AI) cannot fully fix its method in advance of the capability it studies. Here the capability arrived after the plan, which makes this the clearest instance of the problem in the first year. The account therefore gives the change a date. It began in April 2026, when the framework entered, and it continued through the rebuild of the system. The audit that froze the design of the system, and the year of explanation, then closed it. Two findings came out of the change.
The first is the working rule that an agent’s report is not evidence until it has been compared with the machine, which no plan anticipated. The second is the assumption about ownership, which is an argument about research independence. Work that depends on a service its provider can alter is not work that can be relied on across four years.
Unforeseen
Here the unforeseen situations, sometimes difficult, sometimes unexpectedly interesting, will be described as the project runs. Many of these unforeseen situations have already occurred and have been recorded. They will be added in the time to come.