A priming ecology cannot be bought finished or set up from a description. It is built collectively, for each organization, company, project or collective, from its own ways of knowing and with the people who will carry it, and it is theirs from the start.
This page says how a priming ecology is built collectively, and what stays with you afterward. What a priming ecology is, and where it can matter, is in the short version.
Groups that work across fields are bringing language models into how they analyze, plan and decide, often faster than they can agree on what should be handed over. This page is for people in organizations and companies, research projects, public and civic work, and collectives who are considering building a priming ecology, or who want to know what building one would ask of them. It is also for those who are not sure a model belongs in their work at all. Finding that out is part of the work, and the answer can be no.
How a collaboration runs
The first conversation
It begins with a conversation about the work, and about where a model does not belong in it. Some tasks do not need one, and some are better done by a smaller one. Where the work concerns personal data, or a decision an authority makes about someone, the model belongs in the work around the case and never in the case itself. The conversation is also about where each of you stands: there are more positions toward AI than for or against, and letting the default decide is not one of them.
Three phases follow. They overlap and return, and none of them is ever quite finished; a fourth runs through all three.
Mapping
With the people concerned, and mostly without a model: which knowledge practices bear on the work, whose questions are being asked, which tensions should stay open, who is missing, and what the model should never be asked to do, with someone named for each limit. The phase leaves a short written account that the people it concerns can read and correct before anything is built on it. You can stop here and keep what you made.
Building
The first texts are written with you, in your own language, and tried in work you actually do. We start from what your place already has that does the same work, such as a meeting format, a review routine or a way of keeping notes. The time someone has, week after week, to maintain the texts decides how much is built. Every change I propose is visible, and the people whose texts these are can refuse it.
Maintaining
This phase does not end. Someone reads what comes back for drift, the model's slide back toward its defaults, revises the texts when they stop doing what they were written for, and retires what no longer earns its place. I propose to step back when your own people have changed or struck something without me, can say why, and can turn down a proposal from me.
Who bears it
This runs through all three: who maintains the texts, with how much time and with what recognition of the skill involved, who reads what comes back, and who answers for what is carried forward. Maintaining a priming ecology is skilled work, easily mistaken for administration, and it needs hours that someone has agreed to give.
What stays with you
What is built is yours from the start. After the mapping, that is a short written account of your positions, practices, tensions and limits, with a named owner for each limit, and you can stop there and keep it. If you go on, it is also texts in your own language that your people can read, argue about and change, and a named person, role or group who answers for what is carried forward. Whether what you have built is a polyphonic priming ecology is a dated claim that your own people, and those its outputs reach, can contest; no one confers it as a status, and that includes me.
Conditions and limits
These hold throughout, whoever builds the priming ecology.
Anyone concerned can keep the model out of their own part of the work, at any time, without giving reasons. A group can end the model's part in its work as a whole, and you can end the collaboration at any time, without giving reasons.
Someone who was interviewed, recorded or quoted can ask that their words be taken out, and with them any conclusion that rested only on those words.
For every output carried forward, someone answers by name or role, and the people it reaches have a route back to that person. The model is not used to stand in for a decision: “the system says” does not replace “we decided”.
What may be placed in the model's context, and what goes to the company that runs it, is decided before anything is sent. That includes the priming ecology's own texts.
A maintainer's reading cannot check itself; a second reader who did not write the texts is one way to check it.
Where people cannot decline the model's part in their work, what is built is not called a priming ecology.
There is no guarantee: whether it helps can only be tried in the work, over time. The work also has costs: energy and water each time the texts are sent, the load on the places where the data centers stand, the labor the model rests on, and, with a commercial model, dependence on a company that can change it without notice. Whether what it changes is worth what it costs is weighed by those who bear the cost and those who are affected, not by the maintainer alone.
In different contexts
What a collaboration looks like depends on where it happens. Choose a context to see where it can help, an example, and what can stay with you.
In organizational change
Where it can help
Where interests and ways of knowing conflict, power is unevenly shared, and a question has to be reframed before people can work out what to do together. The aim is to strengthen how people think and decide together, and to widen who takes part and on what terms, not just to speed up existing habits.
An example (fictional)
A university gives its staff a general-purpose model with its default settings and a short policy saying that the model must support their judgment and not replace it. Nothing else has been decided, so the default configuration now shapes how hundreds of people's drafts and assessments are first framed. What the default should be is a decision, and it can be taken as one: who chooses the instructions the model is given, who reads and revises them, how a change is dated and signed, how staff can decline without cost, and whether to put a model in front of staff at all.
What can stay with you
A written account of the organization's positions, practices, tensions and limits, with a named owner for each limit. Standing instructions for the model that belong to a function and not only to a person, so that they outlast changes of staff.
In companies
Where it can help
Where shifting demands on sustainability, faster and less predictable change, and the expectation to answer for more than profit raise questions no single department or discipline can answer. That calls for people with different backgrounds and experience working together, among them people with artistic experience who take part as colleagues, not as guests brought in for a workshop. The work is meant to keep the tensions between the company's goals in view, and it is done with the people who will carry the result.
An example (fictional)
A manufacturer of building materials has to cut the emissions of its products while developing a new line of reusable components. Engineering, sales, sustainability and procurement each have their own figures and their own sense of what counts as progress. Before any model is involved, the people concerned map where these goals pull against each other and whose knowledge is missing: the installers who handle the components on site, the people who live near the plant. The priming ecology then keeps those tensions in the texts the model reads when it drafts analyses and proposals, so that a proposal that looks good on one function's figures is read against the others before anyone decides.
What can stay with you
A written account of where the company's goals pull against each other, with a named owner for each tension. Texts the model reads that keep several functions' questions present in analyses and proposals, maintained by someone who has time set aside for it.
In research across disciplinary boundaries
Where it can help
A priming ecology can help shape the question, and read interviews, field notes, archives or literature with questions that pull attention in different directions, so that a reading shows what is explicitly said, what the material leaves uncertain and whose perspectives may be missing. Counts and people's own words are worked with side by side, neither standing in for the other. Building the priming ecology is part of the research, done by someone on the team.
An example (fictional)
A team asked to map a network of funded initiatives first maps its own work, together with people from the initiatives, and writes the funder's values as fields of tension instead of sorting the initiatives by them. The model reads reports and drafts; the team makes every placement. Each placement carries the date it was made and is read by the initiative before the funder sees it, and one initiative's refusal to be placed is reported as such.
What can stay with you
The priming ecology as part of the research, maintained by someone on the team. Readings that mark which statements were read in the material and which were inferred, so that an inference does not travel as a finding. A way to notice early when the work only confirms what the team already thought: someone who did not write the prompt judges the result, and the same question is put to the model with plainer instructions to see what differs.
In public and civic work
Where it can help
Where those affected are many, some of them are never in the room, and the way the work is done can change who comes to be in it. The mapping asks who is missing before any text is written.
An example (fictional)
A municipal planning office invites residents to comment on a plan for their neighborhood, and a model groups several hundred comments into themes for a report to a political committee. Someone signs the list of themes as theirs, having read a sample of the comments against it. The report says what the summaries were made from, what the recordings did not capture and which comments fitted no theme. Residents have a name and an address to say “that is not what we said”, and a resident who did not want their comment read by a model can say so, and the comment is still read.
What can stay with you
For each output carried forward, a person who answers for it and a route back to that person for those it reaches. Reports that say what they rest on and what did not fit, signed by someone instead of saying “the analysis shows”. A way for people to decline the model's part that does not leave them unheard.
What I bring
Two things are needed that organizations tend to underrate. One is a processual orientation: a trained habit of seeing situations as still forming, not as finished and settled, and the tools to keep doing so under pressure, in actual meetings, analyses and texts. The other is experience across fields that has been lived and not only read. Mine comes from growing up between cultures and from work across the history of religions, philosophy, music, improvisation and artistic research, inside the arts and outside them, and from building and maintaining a priming ecology of my own. In the places I have worked, these competences were seldom all present at the start, and they grew through practice more than through reading.
You can also do it without me
None of this means that only a few people can do the work. Much of it needs no model at all: people can say their positions aloud, map the knowledge practices of a place, question renamings, and have work read by someone who answers for it. You can build a priming ecology yourselves, with people who bring such an orientation and experience, starting from what your place already has.
Get in touch
If you are considering building a priming ecology, or want to find out whether one belongs in your work, write to me.
What a priming ecology is, with four short scenes: the short version.
The essay itself tells the four scenes in full and explains how a priming ecology works, why it is built as it is and who answers for it: read the essay