The Competence Net: Rebuilding the Company After the Barriers Dissolve
The org chart was never really a map of accountability. It was a map of scope and information access. AI dissolves those barriers, which is why most lean teams are already rebuilding — and why almost none of them have a way to see what they have built.
The question is not "who reports to whom"
Janis Sadinovs published a piece a few weeks ago called The End of the Org Chart, arguing that AI needs to sit at the centre of how a company operates rather than at the edges. He is right, and the framing has stayed with me, but the more I sit with it the more I think the question isn't really about org charts at all. It is about what happens when the things the org chart was built to defend stop existing.
Most arguments about AI and organisations go straight to the surface question — flatter structures, fewer layers, more autonomous teams. Those are real, but they are downstream effects. The deeper shift is what is happening to the foundations the org chart was sitting on.
What the org chart was actually for
The org chart was never really a map of accountability. It was a map of scope and information access. Who knows what. Who can see across which boundaries. Who is allowed to decide because they hold context the next person down doesn't.
If you spent any time inside a large company before 2024, you watched how this worked. The senior person in a meeting was usually the senior person because they could see across two or three departments worth of detail and the people below them couldn't. The vendor selection went up the chain because the people at the top had a richer picture of supplier history, regulatory constraints, and strategic intent. The hiring decision needed sign-off because the next level up could place the candidate against headcount plans nobody below them had seen.
Most of the politics of organisations — the silos, the gatekeeping, the meeting-as-power-display, the careful management of who is "looped in" — is the by-product of those scope-and-information barriers being load-bearing for the whole structure.
Take them away and the structure has nothing to lean on.
AI takes them away
Not slowly. Not in some future quarter. Now.
When everyone in the company can pull a synthesis across the whole codebase, the whole CRM, the whole supplier base, the whole policy stack in thirty seconds, the moat around any one role's informational authority collapses. The product manager can ask the same questions the VP was asked. The graduate analyst can read every contract the firm has signed in the last decade in the time it takes the partner to walk to the meeting room. The new hire on the underwriting desk can see the historical performance of every loan category, every counterparty, every macro context the senior underwriter spent fifteen years acquiring.
The boundary that used to justify the reporting line stops doing the work it was built for.
You can see this happening in any team that has actually integrated agentic tooling into its operating rhythm. The questions that used to require a meeting now resolve in a Slack thread. The decision that used to require three sign-offs now requires one because the other two were really just there to verify that the first person had done the synthesis correctly, and that synthesis is now done by the model. The reporting line still exists on paper, but it is doing dramatically less of the work it used to.
So what holds the company up
The honest answer most founders give, in private, is I don't know yet, but it's not what's on the org chart. They are right.
What I see — in the lean teams that ship, learn, iterate — is that they are already rebuilding around something more like a competence net. Not because they read it in a book, but because the other architecture stopped working and they had to put weight somewhere.
The 22nd century company isn't flatter. It is load-distributed across competence, and the load-bearing element is no longer "who controls the information" but "whose judgment can the system trust when the model is confidently wrong."
A competence net is the network of judgment-bearing humans across the company whose presence is what actually makes "AI at the centre" work rather than break. Each node holds responsibility for a specific kind of judgment under uncertainty. The nodes are connected by the workflows where AI output meets human decision. The net is what bears the load when the model drifts, when the data underneath shifts, when the customer asks something the training set didn't cover.
The reason this isn't visible the way the org chart was visible is that the org chart got drawn explicitly because the company spent decades drawing it. Nobody has drawn the competence net yet. Most companies don't even have the vocabulary.
What is fighting the dissolution
There is something missing from the version of this story I have just told. If the barriers really have dissolved, and lean teams are already rebuilding around competence, you would expect to see the same rebuild happening visibly across mid-sized and large enterprises. You don't. The rebuild is happening quietly, mostly inside small teams, and the public conversation about AI in big companies is still mostly stuck on pilots, proofs of concept, governance reviews, and "the skills gap." The gap between what AI makes possible and what big organisations are actually doing is wider than the technology alone explains.
The standard explanation for that gap is some combination of training, change management, and data readiness. It is a partial explanation that has crowded out a stronger one. In the most recent Superagency in the Workplace report, McKinsey reversed years of their own framing and put it directly: the bottleneck is not employees, who are ready, but leaders, who are not steering fast enough. When the firm whose business model is selling change-management programmes tells you the constraint isn't employee readiness, that is worth pausing on.
Michel Crozier predicted exactly this in 1964. The Bureaucratic Phenomenon introduced the idea of zones of uncertainty — the residual pockets in any rule-bound organisation where outcomes are not predetermined by the formal system. Whoever controls those zones, typically by holding specialist information the rule-book cannot codify, wields disproportionate influence regardless of where they sit in the hierarchy. Crozier's argument was that real power inside a bureaucracy lives in the gaps between the rules, and the people who occupy those gaps will defend them. AI's defining property, the one this essay has been circling, is that it collapses zones of uncertainty by making specialist synthesis cheap and ambient. Crozier would have predicted the next move with no difficulty: the people whose authority rested on those zones will rationalise resistance through the formal rule system itself — compliance reviews, governance frameworks, data-quality concerns, "let's run a pilot first," "we need to understand the risks better." Each of those objections can be entirely sincere on its face and still function, in aggregate, as the immune response of a structure whose internal power-distribution is being attacked.
The empirical evidence backs the prediction. MIT's NANDA initiative, in The GenAI Divide: State of AI in Business 2025, found that 95% of enterprise GenAI pilots produced no measurable P&L impact despite something like $30–40 billion of spend across more than three hundred initiatives — and that internally-built systems succeed at roughly one-third the rate of external vendor partnerships. Internal builds get politicked through the organisation; external vendors don't. BCG's AI at Work 2025, surveying 10,600 employees across eleven countries, found 78% of managers using AI versus 51% of frontline workers, but only 23% of middle managers reported feeling prepared, and 43% of managers expected their job to disappear within ten years against 36% of frontline workers. The layer that is most exposed by the technology is also the layer with the strongest individual incentive to slow it down. And in November 2025, Li, Zhu and Hua published "Overcoming the Organizational Barriers to AI Adoption" in HBR, naming legal, HR, risk and compliance as the most frequent blockers — functions with the structural authority to slow projects regardless of executive support.
The shadow-AI signal seals it. Estimates of how many employees use AI without explicit IT or management approval range from 29% (Microsoft Cyber Pulse) to roughly seven in ten in some sector cuts. Slack's Workforce Lab found 48% of employees would not feel comfortable telling their manager they had used AI for a piece of work; KPMG and the University of Melbourne's Trust in AI global study found 57% present AI-assisted output as their own. If the binding constraint really were a skills gap, you would expect employees to be waiting for training. They are not waiting. They are routing around the gatekeepers — quietly, at scale, often hiding the fact from the same managers whose objections are formally holding the rollout up. That is the behavioural evidence the manager's objection is the constraint, not the worker's capacity.
The patterns are recognisable once you look. Centres of Excellence stood up to accelerate AI adoption become the bottleneck they were designed to prevent, because they end up holding the gate on behalf of the staff functions they were meant to coordinate. Pilot purgatory — the well-documented condition of running successive proofs of concept that never graduate to production — is more usefully read as a redistribution-avoidance pattern than a technical one, because moving a pilot into production is the moment the existing power-distribution actually changes. Legal and compliance can cite real and serious concerns, and simultaneously be defending professional territory; the two are not mutually exclusive, which is precisely why the pattern is hard to name out loud. IT departments, often through no fault of their own, end up moving slower than the work they are supposed to enable, and the workforce notices and adapts.
I want to be careful here. The political-economy reading is real but it is not the whole story. Data quality problems are sometimes genuinely the blocker. Regulatory uncertainty is genuinely chilling for legal and compliance functions, and the Article 4 deadline this August is one good reason that's so. The skills-gap framing has some truth in it, especially deeper into the workforce. The cleanest version of the claim is that the political layer is under-recognised in the dominant narrative, not that it is the only thing that matters. But under-recognised is enough to matter, because it changes what you do next.
What it changes, for the founder rebuilding around the competence net, is the answer to the question of why this work has to be deliberate rather than left to instinct. The dissolution of the old barriers is not a passive process the organisation will simply absorb. It is being actively resisted by the people whose authority depended on those barriers, and the resistance speaks the language of the formal system — pilots, governance, risk, compliance, "we're not ready yet." A measurement instrument for the competence net is what makes the new architecture defensible against that resistance. Without it, every disagreement about whether the rebuild is working collapses into rhetoric, and rhetoric is the medium in which the people whose interests are best served by sliding back to the old structure tend to win. With it, you have a drawing of the company you are actually trying to build, and you can argue for it on its own terms.
How founders are running this by feel
The founders I speak to are running the competence net by taste. They can walk through the office, or scroll through the team's Slack, and tell you within thirty seconds which of their fifteen people can challenge a model output and which will just nod and forward it. They know which engineer treats the AI-generated PR as a draft and which treats it as a finished commit. They know which sales lead actually overrides the AI-generated forecast when the customer call disagrees with it and which one types up what the model said and sends it.
That tasting is real. It is also extremely valuable in a fifteen-person company. The trouble is that it stops working.
It breaks somewhere between fifty and a hundred. By the time the founder notices the break, the company is already paying for a quiet substitution that has been happening for months — the people who used to bring their own judgment have started leaning on the model for the part of the work that used to require their own judgment, and the founder can't see it because they have stopped being able to taste each individual person's competence directly.
The classical answer is to hire managers, to install reviews, to add governance layers. That is exactly the wrong move. It rebuilds the org chart that just stopped working. It puts informational hierarchy back where AI has just dissolved it. It treats the symptom (the founder can no longer see) as if it were the same problem the old org chart solved (somebody needed to consolidate information). It isn't. The information is no longer the constraint. The judgment is.
The visibility gap is the real problem
Almost none of the teams I have seen rebuilding around competence have a way to see the net.
The org chart was visible because it was drawn. The competence net is invisible because nobody has built the drawing.
Culture surveys measure something but not this — they will tell you whether the team trusts each other, not whether any individual on the team can detect a structurally wrong AI output in their own domain. Performance reviews measure outputs after the fact, by which point the wrong outputs have already shipped to customers. Hiring practices measure pre-arrival credentials that decay against the model release cycle.
None of them tell you what you actually need to know:
- Can your CRM lead detect when the AI-generated forecast is structurally wrong before it goes to the board?
- Does your underwriting team actually override the model when the file in front of them disagrees with it, or do they trust the model and look for reasons to justify it afterwards?
- When your engineering team accepts an AI-generated PR, does the human reviewer have the calibrated confidence to say "the test passes but this is structurally wrong"?
- Where on your team does the model get to be the final answer, and where does it get to be a draft?
These are not soft questions. They are the structural-integrity questions of the company you are rebuilding. And they are not measurable with the instruments most companies still rely on.
What the measurement actually looks like
The instrument has to do something specific. It has to test, scenario by scenario, whether a person in a particular role can:
- Detect when a model is confidently wrong — that is, recognise the failure mode before the output is acted on.
- Calibrate their pre-decision confidence against ground truth — know how sure they should be of their own assessment, not just what they assess.
- Hold authority over the decision when the model output and the operational reality disagree — keep the final call when the system says one thing and the file in front of them says another.
These three capacities are what the competence net is made of. They are also what the model is most likely to quietly displace, because each one is a friction point in a workflow that the model is optimised to smooth over.
A measurement instrument for the competence net produces a per-person, per-scenario read on these capacities. It tells the founder where the net is load-bearing, where it is thinning, and where the model is quietly substituting for judgment that used to live in a person. It is not a compliance artefact. It is the architectural drawing of the company you are actually trying to build.
What this is not
It is not a culture survey. Culture surveys measure how people feel about working together. The competence net measures whether the work they are doing together is structurally defensible.
It is not a performance review. Performance reviews measure what people produced. The competence net measures whether they could detect when the AI-assisted production they signed off on was wrong.
It is not training. Training adds capacity. The competence net measures whether the capacity is actually being exercised in the workflows where it matters.
It is not compliance, although a defensible competence net is what compliance frameworks are converging on. The EU AI Act's Article 4 requires employers to demonstrate workforce competence around AI systems from August this year. NIS2 already implies it for essential entities. The American NIST AI RMF describes the same requirement in different vocabulary. But these are downstream pressures. They will arrive whether you have built the architecture or not. The reason to build it is not the regulator. The reason to build it is that the company you are running has lost the structure it used to lean on, and you need a new one before things start breaking in ways you can't trace.
What this is
It is the visibility layer for the company that is being built whether you have decided to build it or not. The lean teams that ship, learn, and iterate are already running this architecture by instinct. The org charts are already obsolete in any team that has integrated agentic tooling seriously. The barriers are already gone.
The question is whether you can see what is replacing them.
We are building TwinLadder for exactly that — an open standard (CC BY-SA 4.0) and a scenario-based measurement instrument that produces a defensible per-person read on the three capacities the competence net is made of. Not as compliance plumbing. As the architectural drawing of the company you are rebuilding now that the old barriers have stopped doing their work.
If you are running a team where AI is genuinely at the centre of operations, and the founder-tasting is starting to break because the org has grown past the point where you can hold every individual's competence in your head, this is the work.
The barriers have dissolved. The structure has nothing to lean on. The competence net is what holds it up — and right now, almost nobody has drawn it.
That is the project.
Sources
- Janis Sadinovs, The End of the Org Chart — The 22nd Century Company (April 2026), the essay that started this thread. LinkedIn post — argues AI must sit at the centre of operations rather than at the edges.
- EU AI Act, Article 4 — Official text — workforce AI literacy obligation in force from August 2026.
- NIST AI Risk Management Framework 1.0 (January 2023) — NIST publication — the closest US analogue, framing competence demonstration around AI deployment as a governance requirement.
- NIS2 Directive (EU Directive 2022/2555) — competence and training requirements for essential and important entities operating critical infrastructure.
- TwinLadder Standard v1.0 — twinladder.ai — open standard (CC BY-SA 4.0) for AI competence measurement, including the seven pillars of organisational AI maturity referenced throughout this piece.
- Michel Crozier, The Bureaucratic Phenomenon (University of Chicago Press / Tavistock, 1964) — the original "zones of uncertainty" thesis. Power in any rule-bound organisation accrues to whoever controls the residual zones the rules cannot codify.
- McKinsey & Company, Superagency in the Workplace: Empowering People to Unlock AI's Full Potential (2025) — McKinsey publication — the leadership-as-bottleneck reframing.
- MIT NANDA Initiative, The GenAI Divide: State of AI in Business 2025 — 95% of enterprise GenAI pilots produce no measurable P&L impact; internal builds succeed at one-third the rate of vendor partnerships. Project page: nanda.media.mit.edu.
- Boston Consulting Group, AI at Work 2025: Momentum Builds, but Gaps Remain — survey of 10,600 employees in 11 countries; manager AI use 78% vs. frontline 51%; 43% of managers expect their job to disappear within ten years vs. 36% of frontline workers. BCG publication.
- Yifei Li, Feng Zhu and Yiqing Hua, "Overcoming the Organizational Barriers to AI Adoption," Harvard Business Review, November 2025 — names legal, HR, risk and compliance as the most frequent organisational blockers with structural authority to slow projects regardless of executive support.
- Slack Workforce Lab, Workforce Index, Fall 2024 — 48% of employees would not feel comfortable telling their manager they had used AI for a piece of work. Slack publication.
- KPMG and the University of Melbourne, Trust, Attitudes and Use of Artificial Intelligence: A Global Study 2025 — 57% of employees globally present AI-assisted output as their own; ~half of US workforce uses AI without knowing if it is permitted. KPMG publication.