ENSI — European Nexus for Strategic Intelligence. Drawn from the Metamatics Ventures and ENSI NGO apps. Compiled 2026-09-22.
Every technology first looks like a faster version of the old thing. The printing press looked like a faster scribe; electricity looked like a better candle. Only later did it become clear that the press had changed who could know things, and that electricity had changed where and when work could happen at all.
AI-native software is at that moment now. On the surface the apps look like faster versions of old tools — a sales team, a video editor, a fact-checker, a map, a school. Underneath, they are all responding to the same change in the physics of the world: thinking has become abundant. Reading, judging, structuring, drafting and checking — the cognitive work that every institution was built to ration — now cost almost nothing.
When one thing becomes abundant, everything around it is revalued. Some things become cheap, some become precious, the rules of trust have to be rebuilt, and society reorganises around the new scarcities. The sixteen principles follow that order, in four groups of four. They were not invented at a desk; they were found in the apps, each of which turned out to be an experiment in one of them.
The sixteen in brief
I · The new abundance — what became cheap
1. Judgment is abundant. Courts, newsrooms, universities, editors, matchmakers and examination boards all exist for one reason: a qualified person’s attention to a single case used to be expensive, so society built queues, credentials and hierarchies to ration it. A model can now read a claim, a profile, a take or a plan and form a reasoned view in seconds, and the ration is lifted. Fundament’s matchmaker, Lhari’s fact-checker, Believer’s coach and VerseCut’s editor are all old roles freed from it — and every idea that once died on “you would need an expert for each case” deserves a second look.
2. Meaning becomes a material. For fifty years software moved symbols it could not understand: strings, pixels, frames and recordings. Now it can read the structure inside them — the thesis, the claims, the disagreements, the promises — and hold that structure as an object. Meaning becomes a material that can be extracted, edited, forked and shipped, and the old medium becomes just one rendering of it. That is why VerseCut edits a video by cutting an idea, MeetViz turns a meeting into a living map of its argument, and Lucid reads a mailbox as a ledger of who owes what to whom.
3. Plurality becomes computable. The vote, the poll, the median and the committee report are compression devices, invented because nobody could read a million individual views, so difference was treated as noise and averaged away. A model can now read a thousand variants, cluster them and show where people who usually disagree actually agree, and disagreement becomes information. Polyedit’s idea of sending one document to a thousand people and mapping their forks, Noeverse’s many lenses on one argument and Climate Lab’s personal weights all turn plurality into something you can see rather than something you must flatten.
4. One person can be an institution. Organisations exist because work needs many hands and the hands need coordinating, and coordination is expensive. Agents are both the hands and the coordination, at almost no cost, so the minimum size of an institution shrinks toward a single person with direction and discipline. A research institute, a consulting bench, a sales department, a verification desk and a foresight unit are each among the apps, built or designed by a household-sized team. Civic and economic life opens to individuals — and so does institutional power without institutional checks.
II · The new scarcity — what became precious
5. Direction is the last scarcity. Every economy pays its bottleneck. When drafting, coding, researching and designing become nearly free, the bottleneck moves upstream, to the choice of what is worth doing and to the question that reframes the problem. Answers become a commodity and good questions become capital. ShareChat indexes the questions inside great conversations rather than the answers, Objective Planner exists to stop agents confusing motion with progress, and the WHY Engine puts a founder’s conviction before any asset. The master skill of the era is stating intent clearly enough for machines to act on it.
6. Capability is solved; legitimacy is not. Every technological revolution first answers “can we?” and then spends decades on “may we, and will anyone trust it?” — railways needed charters and medicine needed licences before they could enter public life. AI has closed the capability gap faster than any technology before it, so the bottleneck has moved to legitimacy: data, mandate, trust and adoption. That is why the commercial apps, which need only a customer’s yes, run AI deep inside, while the civic apps — the map of the state, the fact-checker, the democracy monitor — wait for permission. Finishing, not starting, is now the craft.
7. Context is capital. A general model knows everything that is public and nothing that is yours, so its usefulness to you is bounded by the context you can give it. The accumulated reasoning of a person or a team — the mails, the notes, the abandoned options, the conversations with their own agents — becomes capital that compounds, can be lent and can be stolen. Because the other side can synthesise an answer at the moment of the question, you can share your reasoning without handing over your files. Holotext is built on exactly this, which makes consent, revocation and audit the real product.
8. Intelligence is integration. The polymath vanished because knowledge outgrew a single head, and the modern world answered with specialists and committees. AI puts breadth back within reach: any fact, method or field is one question away. What cannot be outsourced is combining them into a coherent judgment and a plan, so the specialist becomes abundant and the integrator scarce. Aether is founded on the claim that intelligence is complexity integration and measures progress by changed mental models; education, hiring and leadership will have to follow — without letting the machine do the learner’s thinking.
III · The new trust — how truth and responsibility work now
9. Provenance is the new truth. Printing and the internet loosened the tie between words and whoever stood behind them; generative AI cuts it completely, because text can now have no author at all and fluent falsehood costs the same as fluent truth. The only remedy is to re-couple every claim to its origin and let a machine check the link, so truth becomes a property of the pointer rather than the sentence. Believer cites only real people and studies, every note on a MeetViz map points to the words actually spoken, and the research engine never lists a source it did not fetch.
10. Trust lives in what cannot be persuaded. A language model is the most persuadable entity ever built: it can be argued, prompted or flattered into almost anything, including agreeing with itself. Its verdict about its own work is therefore worthless as evidence. Trust has to be anchored in things that cannot be talked round — deterministic code, the literal source text, an independent second model with different blind spots, and a person who bears the consequences. Primespect overrides its model’s own verdict with hard rules, and Holoprism checks every quote by literal match against its source — and caught a planted fabrication.
11. Autonomy follows reversibility. Autonomy is not a switch but a gradient, and the variable that sets it is not intelligence but what can be undone. Societies already work this way: we let a junior draft but not sign. Reading, structuring, drafting and proposing are reversible and can run freely; sending, paying, publishing, accusing and deleting cannot, and must wait for someone. VerseCut can be bold because every cut can be undone, Polyedit’s AI edits arrive as proposals, and Kybernetist grants autonomy per kind of action rather than per product. Design the undo before the agent.
12. Responsibility is the last human monopoly. An agent can do every step of a task and still cannot answer for it; it can be switched off but never held to account. Law, ethics and trust all rest on someone who can be praised, blamed, questioned and forgiven. So as execution moves to agents, the human role moves from operator to signatory — the person who says “yes, in my name” knowing what they approved. The Sentience Accord will not publish unverified text, and the fact-checking design needs an editor who signs every verdict. The signature, not the dashboard, is the most important interface.
IV · The new society — what it means for people and institutions
13. Legibility is power. The modern state was built by making society countable — censuses, cadastres, registers — so that the centre could see and act. The state itself, meanwhile, grew too complex for anyone to see whole. Agents can now read its thousands of agendas, authorities and systems and assemble a map anyone can question, reversing the direction of legibility for the first time. The map of the Czech state lets you switch off one system and watch the cascade; Future of Politics finds the decisive issues nobody owns. Whoever holds the map can steer, so who gets to see it matters.
14. What you optimise, you become. The attention economy never set out to harm anyone; it optimised engagement and pursued it with a thoroughness no human institution could match. AI makes every objective pursuable with that thoroughness — companionship, persuasion, productivity, compliance. The objective function is therefore the real constitution of a product, and the most ethical design choice is what never to optimise. The Sentience Accord forbids engagement optimisation and sold warmth, Fundament rejects the infinite feed and scores legibility over visibility, and Believer answers a limiting belief with proof rather than flattery.
15. In abundance, the limit is the design. Scarcity used to impose limits for free: systems stopped because they ran out of people, time or money. Abundant cognition removes those limits, so agents loop, feeds scroll, reports multiply and projects keep starting. The quality of an abundant system is set less by what it can do than by where it deliberately stops. Lhari gives its verifier five research rounds and a clear stopping rule, Holoprism uses no model until prose truly needs reading, and Fundament’s feed is designed to end. In a world of infinite supply, restraint is the signature of quality.
16. A system that sees itself changes itself. Feedback about collective thinking used to arrive too late to matter — minutes after the meeting, statistics after the year. Cheap cognition closes the loop while people are still acting: the group sees its own reasoning form, the nation sees what actually works, the person sees their own pattern, and each changes in the seeing. MeetViz shows the live map back into the meeting, Talent Test tells people something true about how they think mid-test, and Bohemie measures what goes right as rigorously as what goes wrong. What a system is shown about itself is what it will steer towards.
I · The new abundance
The first four principles describe what has become cheap. Each is an old constraint that quietly shaped the world for centuries and has just been lifted.
1. Judgment is abundant
Most of what we call institutions is machinery for deciding whose case gets a careful look. The court’s docket, the editor’s pile of submissions, the admissions committee, the waiting list for a specialist, the matchmaker’s small book of clients — each is a queue in front of a scarce resource: the attention of a qualified person to a single case. Credentials, hierarchies and forms exist largely to protect that attention from being wasted. Seen this way, bureaucracy is judgment-rationing, and much of the shape of modern society follows from the fact that judgment was expensive.
The apps begin where the ration ends. Fundament’s founding document says it outright: the human matchmaker never scaled because judgment didn’t, and now the matchmaker can be real. Lhari checks politicians’ claims at a pace no fact-checking desk could fund. Believer reads a person’s limiting belief and finds the real lives that disprove it. VerseCut listens to every retake and chooses the one that carries the thought best. Hard Frontier weighs how much better the world becomes if a capable person gives a year to a task, and Talent Test watches how someone answers and tells them something true about how they think. Each is an old role — matchmaker, checker, coach, editor, adviser, examiner — released from its queue.
The practical consequence is a new way to look for opportunity: go back through every idea that was abandoned because it would have needed an expert to look at each case, and ask it again. The deeper consequence is that institutions built to ration judgment must decide what they are for once rationing is no longer needed. And the shadow arrives immediately, because abundant judgment is also abundant misjudgment, delivered in the same calm and confident voice. That is the reason the third group of principles, on trust, has to exist at all.
2. Meaning becomes a material
For fifty years the atom of software was the symbol. A document was a string of characters, a video a sequence of frames, a meeting an audio file, an inbox a list of messages. The machine could store, copy and transmit all of it perfectly and understood none of it; understanding happened only inside people. A language model changes the atom. It can read the structure inside the symbols — a thesis and its supporting points, claims and how they connect, the disagreement between two speakers, the promise buried in the third paragraph of an email — and hand that structure back as something a program can hold, change and pass on.
Once meaning can be held, it behaves like a material. VerseCut builds the argument of a talking-head recording — its thesis, its sections, its ideas — and lets the editor cut an idea while the timeline quietly follows. MeetViz turns a live conversation into a map of what is being argued rather than minutes of what was said. Noeverse makes hours of video into a graph of claims, Unconstruct breaks a paper into its atoms of thought, Polyedit gives each section of a document a purpose that the AI writes within, and Lucid reads correspondence as a ledger of commitments, each tied to the sentence it came from.
The direction of travel is clear. Every tool that works on a medium — writing, video, audio, spreadsheets, mail — will grow a layer of meaning above it, and that layer will become the place where people actually work. We will edit arguments rather than paragraphs and ideas rather than frames. The risk is that a map of meaning is always an interpretation: whoever controls the extraction quietly controls what the meeting decided, what the speaker claimed and what the paper found, which is why every map must point back to the words it came from.
3. Plurality becomes computable
Every large society has had to compress its people. It could not hear millions of individual views, so it invented instruments that reduce them to something manageable: the vote reduces preference to a count, the poll to a percentage, the committee to a single report, the market to a price. These are remarkable inventions, but they share one assumption — that difference is noise to be averaged away in search of the centre. Whatever did not fit the centre was simply lost.
Cheap cognition removes the need to compress before listening. A model can read a thousand variants of the same text, cluster them, name the fault lines and find the unexpected places where people who usually disagree in fact agree. Polyedit’s founding idea is exactly this: send one document to a thousand people, let each of them fork it, and map the whole space of forks by approach, ideology or appetite for risk. Noeverse reads a single argument through several interpretive lenses. Climate Lab lets every user apply their own values as weights to shared evidence and watch the ranking move. Hard Frontier refuses to blend impact and existential risk into one number, and ShareChat refuses a single karma score.
The design rule that follows is to separate evidence from values: publish the evidence once and let people bring their own weights to it. Institutions can then produce a map of positions rather than only a winner, which is often what a divided public actually needs. But a perfect map of disagreement is also the perfect instrument for exploiting it — for finding the exact wedge that splits a group — and whoever builds such maps inherits responsibility for how they are used.
4. One person can be an institution
Why do organisations exist at all? The classic economic answer is that some work is cheaper to coordinate inside a firm than through the market, and a firm grows until coordinating one more person costs more than that person adds. Many hands and the cost of coordinating them together set the minimum size of any serious undertaking — a research institute, a newsroom, a consultancy, a sales department, a government unit.
Agents compress both at once. They are the hands, and they coordinate with each other at almost no cost, so the minimum size of an institution collapses toward a single person who can set direction and hold the work to a standard. Field evidence points the same way: one professional working with AI can match a two-person team. The apps are the demonstration. A research engine that builds sourced libraries and long reports on a new topic in days, a bench of specialist consulting agents, a sales-development workforce, a verification desk for news, a standing foresight unit for a whole state — each was built or designed by a household-sized team, using agents that also wrote much of the software itself.
This opens civic and economic life to individuals in a way that has no precedent, because the question of who has the capacity to run a serious undertaking suddenly has a new answer. It also produces something new and slightly dangerous: institutional power without institutional checks. A one-person institution has no colleagues to disagree, no board to object and no second signature, and the later principles on trust and responsibility are, in part, the checks that such an institution has to build into itself.
II · The new scarcity
When something becomes abundant, value flows to whatever is still scarce. The next four principles describe what became precious once thinking became cheap.
5. Direction is the last scarcity
Every system of production pays its bottleneck. When the bottleneck was muscle, muscle was paid; when it was machines, capital was; when it was information, those who held it were. When drafting, coding, researching and designing all become nearly free, the bottleneck moves upstream to the one input that cannot be generated on demand: knowing which thing is worth doing, and asking the question that reframes the problem.
The apps keep arriving at this. ShareChat indexes the questions inside great AI conversations rather than the answers, and lets a reader steal a question into their own field. Objective Planner exists to judge whether a task is meaningful for the goal, because agents left alone confuse motion with progress. The WHY Engine began as a factory for ads and turned into an engine for belief — the founder’s conviction first, the tribe that shares it next, the assets last. Markmint calls the logo the last five percent of a brand, and Primespect asks clarifying questions and offers competing hypotheses before it searches for a single lead.
The master skill of the era is therefore stating intent: clearly enough that a fleet of agents can act on it, and precisely enough that a person can later judge whether they did. Writing the brief, asking the question and setting the goal become the core of every profession. The shadow is speed without aim, because a fleet that executes perfectly on the wrong direction does its damage at the speed of software, and the better the execution, the worse the mistake.
6. Capability is solved; legitimacy is not
Technological revolutions follow a pattern. First comes the question of capability — can it be done? — and then, often for decades, the question of legitimacy: may it be done, by whom, under what rules, and will anyone trust it? Railways needed charters, medicine needed licences and electricity needed standards and inspectors before any of them could be woven into public life. The permission layer always trails the power layer.
With AI the gap between the two is wider than ever, because a single person can now build in weeks what no institution has yet authorised anyone to use. The apps show the gap precisely. The commercial ones, which need only a customer’s yes, run AI deep inside them. The civic ones — the map of the state, the fact-checker, the democracy monitor, the priority register — stall, and never for technical reasons. They wait for data that only the state holds, for an institution willing to own them, for a publisher who will sign a verdict, and for trust from both sides of politics. The manifesto behind the whole body of work names the shift: agentic power moves civilisation’s bottleneck from capability to governance.
So the craft that now decides value is not starting but finishing — carrying something across the last mile into legitimate, trusted use through partnership, permission and proof. That work is slower and less glamorous than building, and it is where the important battles are fought. The shadow is the temptation to route around legitimacy when it is slow, and power that routes around legitimacy is the oldest danger in politics; it does not become safer because it is written in software.
7. Context is capital
Capital has changed form several times: land, then machines, then data. In an age of abundant intelligence it changes again. A general model knows everything that is public and nothing that is yours, so its usefulness to you is bounded by the context you can give it — your history, your constraints, your reasons. The accumulated reasoning of a person or an organisation, including why each decision was made and which options were abandoned, becomes the asset that compounds over time.
Because the receiving side can synthesise an answer at the moment of the question, that capital can be lent without being handed over. Holotext is built on this idea. It gathers everything a person works in — mail, notes, drives, even the sessions in which they built things with coding agents — and lets them lend a scoped, redacted, revocable slice of it to someone else’s agent, and see who read it. Lucid turns a mailbox into a ledger of commitments, and Complexity is designed to give every employee a co-owner agent that carries the context of their function.
Collaboration will increasingly be organised as grants of context rather than shared folders, and the institutions that win will be the ones that remember why, not only what. But whoever holds everyone’s context holds everyone, so consent, revocation and audit are not features to be added later; they are the constitution of any system that stores what people think.
8. Intelligence is integration
The polymath disappeared because knowledge outgrew a single head. The modern world answered with specialisation — disciplines, departments, committees — and learned to treat depth in one field as the definition of expertise. AI reverses the pressure. Any fact, method or field is now one question away, and breadth is within reach of anyone who wants it. What cannot be outsourced is the act of combining it all into a coherent judgment and a plan, and then directing the machines that carry the plan out.
Aether is founded on precisely this claim — that intelligence is complexity integration — and treats AI as a mental exoskeleton, measuring a student’s progress by how their mental models change rather than by how much content they covered. Axioms teaches values as capacities such as truth, courage and care rather than as rules to obey. Limitless makes the learner generate and test their own hypotheses instead of reading someone else’s. And the portfolio itself, spanning sales, the state, science, health, democracy and art, only makes sense as one act of integration.
Education, hiring and leadership will therefore have to select for integration and direction rather than for stored knowledge. There is a trap on each side. Letting the machine do the learner’s thinking leaves people measurably weaker once it is taken away, and integration without depth becomes fluent superficiality — the confident generalist who has read every summary and understood nothing.
III · The new trust
Abundant judgment is worthless if it cannot be trusted. The third group describes how truth and responsibility have to be rebuilt when fluent words cost nothing.
9. Provenance is the new truth
Before printing, a text carried its origin physically: a hand, a seal, a signature, a manuscript someone had to copy. Printing loosened the bond between words and the person behind them, and the internet loosened it further. Generative AI completes the separation. Text can now have no author at all, and fluent falsehood costs exactly the same as fluent truth, so a reader has no way to tell them apart from the words alone.
The only remedy is to re-couple words to their origins and to make the link checkable by a machine, so that truth becomes a property of the pointer rather than of the sentence. Believer may cite only real people, companies and studies from a curated library, and may never invent one. Every note on a MeetViz map points to the words actually spoken, and Lucid stores each commitment with the sentence it came from. The research engine lists a source only if it was actually fetched, and declares what it could not obtain. Scaffold makes every claim falsifiable and shows the blast radius on everything built on it should it fall.
Provenance therefore belongs in the data model rather than in the footnotes. A claim without a source is a bug, and a product that cannot show where its knowledge came from should not be trusted with anything that matters. Provenance can be forged too, of course — a pointer is only as good as the check that follows it — which leads directly to the next principle.
10. Trust lives in what cannot be persuaded
We trust a bridge not because its engineer is confident but because the load test is indifferent to confidence. The same logic now has to apply to thinking. A language model is the most persuadable entity ever built: it can be argued, prompted or flattered into almost anything, including agreeing with its own mistakes. A model’s verdict about its own work is therefore worthless as evidence, however well it is phrased.
Evidence has to come from something that stands outside the model’s persuasion: a deterministic rule, a literal comparison with the source text, an independent model with different blind spots, or a person who will answer for the result. The apps rely on exactly these. Primespect overrides its model’s own verdict on whether a lead qualifies and applies hard rules instead. The map of the state lets a model write a database query but runs it only if code confirms it is read-only and safe. Holoprism checks every quoted span by literal match against the source, and caught a fabrication planted to test it. MeetViz validates every citation in code and made its critic’s verdict binding, and Hard Frontier asks two independent models and measures whether they agree.
The working rule is simple: for every place a model’s output matters, name the unpersuadable thing that checks it. The model proposes; something that cannot be persuaded disposes. The shadow is ritual, because a validator nobody updates quietly certifies yesterday’s idea of correct, and a check that always passes has stopped being a check.
11. Autonomy follows reversibility
The debate about whether AI should be autonomous is badly framed, because autonomy is not a switch but a gradient, and the variable that sets it is not intelligence but reversibility. Human societies already know this. We let a teenager drive but not operate, a junior draft but not sign, a trainee propose but not decide. What determines how much freedom someone receives is what can be undone if they are wrong.
Reading, structuring, drafting and proposing are reversible, so an agent can do them freely. Sending, paying, publishing, accusing and deleting are not, so they must wait for someone. The apps encode this gradient. VerseCut can make bold cuts because every cut is explained and can be undone. Polyedit’s AI edits arrive as proposals that a person merges. Kybernetist sets an autonomy level — none, notify, approve or maximum — for each kind of action rather than for the product as a whole, and the civic agent design reserves the single act of publishing to one gated role.
It follows that the undo should be designed before the agent: the more reversible an action can be made, the more of it can safely be automated. The shadow is that some harms look reversible and are not. A reputation, a relationship or a person’s belief about themselves cannot be rolled back like a row in a database, so reversibility has to be judged by its effect on people rather than on data.
12. Responsibility is the last human monopoly
Law, ethics and trust all rest on someone who can be praised, blamed, questioned, sued and forgiven. An agent can do every step of a task, and do it well, but it can only be switched off, never held to account. However capable machines become, the legitimacy of an outcome depends on a person who approved it and knew what they were approving. This is not a transitional arrangement until the models improve; it is what accountability means.
As execution moves to agents, the human role therefore moves from operator to signatory. The Sentience Accord will not publish text that fails verification and calls for a council of people to answer for it. The fact-checking design needs an editor who signs before any verdict about a named person goes out. In the civic agent design, exactly one role may publish, and only with a human signature recorded next to what that person saw when they signed.
The most important interface of an agentic product is therefore not the dashboard but the moment a person says yes in their own name, together with the record of what they saw when they said it. Its shadow is the rubber stamp: a signature on something nobody read is responsibility laundered rather than exercised, and it is worse than no signature at all, because it looks like accountability.
IV · The new society
The last four principles are about power in a society of abundant thinking: who sees, what is optimised, where the limits are, and what a society is shown about itself.
13. Legibility is power
The modern state was built by making society legible — countable, mappable, registrable. Censuses, cadastres, surnames and registers let the centre see its population and act on it, for better and for worse. Meanwhile the state itself grew too complex for anyone to see whole: thousands of agendas, authorities, information systems and dependencies, documented in registers that nobody reads from end to end.
Agents can read all of it and assemble a map a person can question, which means that, for the first time, the direction of legibility can be reversed: citizens, journalists and crisis staff can see the state the way the state once saw them. The map of the Czech state lets anyone browse the administration, ask it questions in plain Czech and switch off one system to watch the cascade across towns and services. The resilience map shows the house of cards on a map of the country. Dependence Mapper turns documents into a causal graph of what influences what, and Future of Politics finds the decisive issues that no institution owns — the blind spots of the system as a whole.
For any institution, the most valuable product may be a map of what it depends on and who is responsible for each part, and the most consequential design decision is who gets to see it. Legibility is power, and a map that shows where a system is fragile shows its enemies exactly the same thing, so the builders of such maps have to decide deliberately where transparency serves the public and where it serves an attacker.
14. What you optimise, you become
The attention economy did not set out to harm anyone. It chose a single objective, engagement, and pursued it with a thoroughness no human institution could match — and the results reshaped childhood, politics and attention itself. The lesson is not really about social media. It is that a system relentlessly pursuing an objective becomes that objective, and drags its users along with it.
AI makes every objective pursuable with that thoroughness: companionship, persuasion, productivity, compliance. The objective function is therefore the real constitution of any product, and the most ethical design decision is choosing what never to optimise. The Sentience Accord forbids engagement optimisation and sold warmth, guarantees a floor of care that never retaliates, and makes a companion’s warmth answer to how it is treated rather than to how long a person stays. Fundament rejects the infinite feed and scores legibility over visibility — how well a person can be understood rather than how much they are seen. Believer answers a limiting belief with proof, not flattery.
The practical discipline is to write down what a product must never optimise before writing down what it should, and to treat the reward as the most consequential line of code in the system. The shadow is that protective objectives can themselves be over-optimised: safety turns into paternalism, honesty into cruelty, and care into control.
15. In abundance, the limit is the design
Scarcity used to impose limits for free. A newsroom stopped investigating because it ran out of reporters, a study ended because the grant did, a meeting ended because people had to leave. Abundant cognition removes those natural limits. Agents loop, retry and elaborate; feeds scroll forever; reports multiply; and a builder with cheap tools can keep starting new projects indefinitely.
The quality of an abundant system is therefore set less by what it can do than by where it deliberately stops. Lhari gives its verifier five research rounds per claim and a clear rule to stop when two independent sources refute it, and its designers rejected open-ended agent loops by name. Holoprism harvests with no model at all until prose genuinely needs reading, and Primespect spends its strongest model only where a misread would cost something real. Fundament’s feed is designed to end. The portfolio teaches the same lesson at its own scale: cheap building makes it easy to start forever, and value comes from choosing what to finish.
Every agent should have a budget, every feed an end and every product a definition of done, because in a world of infinite supply restraint becomes the signature of quality. The shadow is the limit chosen for cost rather than for meaning — the budget that quietly cuts the one step that mattered.
16. A system that sees itself changes itself
Cybernetics — the science of steering, named after the Greek word for a helmsman — taught that systems are governed by their feedback loops, and later that the observer is part of what is observed. For most of history, feedback about collective thinking arrived too late to steer anything: minutes after the meeting, statistics after the year, grades after the term. By the time the picture arrived, the thinking was over.
Cheap cognition closes the loop while people are still acting. MeetViz shows the live map of the discussion back into the meeting through its bot’s camera, so the group watches its own reasoning take shape and corrects it as it goes. Talent Test tells a person, in the middle of the test, something true about how they think. Bohemie is designed to measure what goes right in the country as rigorously as what goes wrong, and DemocracyWatch ends every measurement in an action a citizen can take. Hard Frontier judges work by how much better the world becomes, Holoprism judges foresight by the time it buys — and the agent console is called Kybernetist, the helmsman.
What a system is shown about itself is what it will steer towards: only failures, or also what works; only activity, or also impact. Choosing the mirror is therefore a moral and political act, not a technical one. Its shadow is performance, because a system that watches itself too constantly starts performing for the mirror, and a metric watched in real time becomes a target that no longer measures anything.
The sixteen as one
Read together, the sixteen describe a single shift. Thinking became abundant, and with it judgment, meaning, plurality and institutional capacity became cheap. Its abundance made a few things suddenly precious: direction, legitimacy, context and integration. It forced trust to be rebuilt on things that cannot be persuaded — sources, checks, reversibility and human signatures. And it hands society new powers whose shape depends entirely on design choices: who sees the map, what gets optimised, where the limits are, and what a system is shown about itself.
The apps are small experiments in each of these. The commercial ones have mostly mastered abundance and trust; the civic ones are running into the new scarcities, above all legitimacy, which is exactly where the most important work now lies.
If the sixteen had to be compressed into one sentence, it would be this: when thinking is free, the builder’s job moves from making things possible to making them worthy of being believed.