Post-AI planning
Data acquisition, analysis and delivery have become utilities. This is a map of what remains: two territories, a compass to take bearings in either, and a passage between them that no institution yet trains anyone to cross. A university could take it as a course, as an audit practice or as a small research unit.
It is a map of the work and not of the whole. Holagram holds that the whole cannot be mapped, and one of the two territories below cannot be surveyed at all. The map applies to the planning of any ecology. The examples are spatial because the scenario is.
Two territories
Two territories are already charted. One is surveyed to the metre and belongs to the agents. The other has never been surveyed and cannot be. The work happens in neither. It happens in the passage between them.
The agents
Abstract, and commoditised. A district mobility plan, 2029: a short brief in the morning; by noon agents have retrieved the registers, run the models and delivered a hundred and forty pages. Three consultancies deliver the same day, at marginal cost. Acquisition, analysis and delivery are three utilities, and what remains scarce is thinking. This is a scenario and not a forecast, since each layer exists in some form now.
Holagram’s reading. Each agent perceives an ecology through its own hola, built from the data it can reach, and a model’s hola is not an outside view (Hola, principle 3). The agents’ territory is surveyed to the metre because it is made of what can be surveyed: registers, counts, models and categories.
The field
Contextual. This place, this Tuesday, these people: observation, expert interviews, lived experience. It is produced while the agents run, and it is the one input the agents cannot manufacture. It cannot be surveyed.
Holagram’s reading. This is Holagram’s premise. The whole cannot be mapped, only experienced from inside, in the first person, and ambiguity stays (Hola, principle 4). The field is not a second dataset. It is made of the holas of the people in it, each mapped in a gram that carries its hands.
The map. Two territories and two axes: the field on the contextual side, the agents on the abstract side, conceptualising above and acting below. Four positions sit on a ring between them, and the ring is the passage.
The compass
Before crossing, an instrument. Five bearings can be taken of any agentic planning process, in either territory. They are adapted from the heuristics of critical urban research, and agency is added for a time when agents represent people. The same five bearings structure every audit, every case file and every course.
| Bearing | The question | Holagram’s reading |
|---|---|---|
| Motivation | Whose interests does this serve? | The purpose of the document or process, the first line of the boundary note on the Gram page. |
| Power | Who controls the resources, and the model? | Who can say no, and by what means. This is the mandate question. |
| Knowledge | Whose framing does the analysis serve, and who can critique its outputs? | The hands: who or what wrote each part, and whether anyone else can open the trace and contest it. |
| Legitimation | A causal argument, or the appearance of automated rigour? | The Test: what would show it wrong, run at another level or by another agent. |
| Agency | Whose capacity to act survives, and whose is displaced onto the model? | Who can take the next step, and who answers for it. |
The passage
Between the territories runs the passage: how we think when a machine does the rest. Every agent inherits an ontology from the data it can reach. Making it explicit, choosing it and contesting it is the post-AI equivalent of choosing a method. The crossing has three waypoints.
| Waypoint | The question | Holagram’s reading |
|---|---|---|
| 01 Conceptualise what counts as real | How do we conceptualise spatial and processual phenomena? What “a neighbourhood”, “a flow” or “a need” become once an agent operationalises them. A model can only see what its categories allow. | Every agent has a view. Record each view beside the others with its hands, and say where the ambiguity sits. |
| 02 Act | How is action linked to conceptualisation? A plan is a conceptual commitment made executable. When execution is automatic, the commitment is all that is left to decide, and to answer for. | Follow a reading to an act, and name who can take the next step. |
| 03 Register between registers | How contextual, or how abstract, is our thinking? Knowing which register a claim is in, and moving between them without losing what each one knows. | State the level. A reading at one level is tested at another, or by another agent. |
The third waypoint carries the method’s central rule, which is also the rule for the passage: a reading is tested at another level, or by another agent, than the one that made it. The field holds the lower levels and the agents the widest. Crossing means keeping each level’s claims from standing in for the others’. See Gram.
Four positions
On the ring sit four positions. Each is a way of thinking and not a skill, and each replaces a competence the agents have absorbed. Read from the field they spell WEST: the side the field lies on, and the direction the work reads from.
| Position | Replaces | What it does | In the Gram |
|---|---|---|---|
| W Witnessing | Collecting data | At the edge of the field: what never enters a dataset, translated into terms that can contest the model. The human sensor and the human veto. | Trace and Reading from the ground, with the hands recorded. The veto is human. |
| E Evaluating | Running the analysis | At the edge of the agents’ territory: is this synthesis accurate, biased or beside the point? Pattern output read for what it silently assumed. | The Test, run at another level or by another agent. A synthetic Adversary may argue it, and a person decides the result. |
| S Steering | Delivering the project | Setting the course a plan commits to, and answering for whose interests, knowledge and agency it encodes; how agents stand in for absent, future or non-human populations. | The mandate: who can take the next step, and who signs. A synthetic voice may write labelled Fiction or argue a Test. It never supplies a Trace. |
| T Theorising | Operationalising given variables | Naming what the model takes as real; writing the brief so the agents answer the right question. The bearing every other position is taken from. | The level statement, and the choice of levels, written before the first entry. |
In Holagram’s terms the four positions are four roles in one entry and not four stages of a project. One person can hold more than one, but not for the same reading, since no reading judges itself.
What it produces
People who can cross again: evaluators of AI-assisted planning, such as planners, data scientists, geographers and practising officers, employed where plans are commissioned, audited and defended. Duties of human oversight, such as those in the EU AI Act, need people with the competence and authority to exercise them (EU AI Act 2024).
The bearings and positions give a university three things to run. A course, in which each position is taught as a way of thinking. An audit practice, in which the five bearings are put to a process that is already running. A small unit, in which each member holds one position and works the instruments it needs. In Holagram’s form the case file is a gram: entries with four registers, the hands recorded, the tests run and the failures kept, so that the work can be inspected by someone who was not there.
Where it could be wrong
- A map can be taken for the territory. This one maps the work. The field cannot be surveyed, and any record of it, a gram included, is a view from a position.
- The premise is a scenario. If execution does not become a utility, the bearings and the positions still hold, with less urgency.
- Judging can itself be automated. Models can be asked to criticise models. No study found tests a model critic on a reading of an ecology (see synthetic hands), so a synthetic evaluator is a tool to calibrate and the veto stays with a person.
- The lineage of the bearings is stated and not traced. They are adapted from critical urban research, and the sources are not yet listed here.