TwinLadder Weekly

Issue #41 — BCG Has a Name for It Now

20 June 2026 · Weekly intelligence on judgment, governance, and the boards accountable for both

Editor's Note

From Alex —

On 17 June, Boston Consulting Group published a phrase. Sagar Goel, David Martin and Charikleia Kaffe call it distributed de-skilling, and they define it as "a collective erosion of human skills that undermines organizational intelligence and resilience over time."

The phrase describes something at the level of the institution. Critical thinking, judgment and problem framing thinning across hundreds or thousands of people at once — a loss that has mostly been discussed one worker at a time, given a name that fits a company.

Then the finding. In a study of seventy C-suite leaders and senior executives across multiple industries, half say they are already observing it in their own organisations. More than sixty per cent expect it to be a material threat within three to five years. And when those leaders ranked the capabilities most critical to long-term performance against the capabilities most exposed, the same item came first on both lists: judgment and decision making.

Seventy is a small sample. Take the percentages as the settled opinion of seventy senior people and nothing wider, and take the vocabulary as the durable part. Executives now have a term for losing judgment at scale, and they are using it about themselves.

Something else was published on 17 June, and holding the two together is the work of this issue. The Diligent Institute's quarterly reading of US public-company directors found that 82% had used generative AI in their board work in the previous six months. Six per cent said their company had a policy covering it.

Liga has the analysis.

— Alex

What Was Published on 17 June

The BCG piece is short and its argument is structural. Individual de-skilling has been studied for years — the pilot whose hands go quiet, the radiologist who stops reading the scan cold. What the authors name is the arithmetic of the same effect running through an organisation chart. When everyone in a function reaches for the same tool for the same first pass, the erosion arrives everywhere at once, and it arrives without a single visible event to attach it to.

The findings, verbatim: "In a global study of 70 C-suite leaders and senior executives across multiple industries, we found that half are already observing de-skilling in their organizations, and more than 60% believe that de-skilling will pose a material threat within the next three to five years."

Two further numbers travel with it. One company in ten reports an organisation-wide strategy against de-skilling. A third of the leaders say the subject has never been discussed inside their organisation at all.

Set the sample against the claim before going further. Seventy executives is a qualitative sample dressed in percentages. It reports what leaders say about their organisations; it measures no workforce. Half of seventy is thirty-five people. Nothing in it establishes the rate at which any capability is actually eroding anywhere.

What the sample does establish is different and, for a board, more immediately usable. Seventy senior people were asked to rank the capabilities their organisations depend on, and to rank the capabilities their organisations are losing. Judgment came out on top of both. That is a self-report, and self-reports about one's own skills are unreliable in a specific direction — people cannot feel their own erosion, which is why the individual studies use unassisted tests instead of surveys. But this survey asked leaders about the organisation beneath them, which is a thing a chief executive is comparatively well placed to observe. They looked down at the bench and reported that the part they need most is thinning.

That is a risk statement made by the people who would have to own the risk. It has a name, an owner class and a time horizon. What it lacks is an instrument.

The Same Day, on the Board's Own Desk

The Diligent Institute publishes a quarterly Director Confidence Index with Corporate Board Member. The Q2 2026 edition, published on 17 June, surveyed 104 US public-company directors. The fielding period stays undisclosed, so treat the date as publication.

The headline: 82% of directors had used generative AI in their board work in the previous six months, against 66% in September 2025. Sixteen points in nine months.

Underneath it, the uses. 30% had used AI to summarise board books or meeting materials. 45% had used it to prepare for discussions or to benchmark peers. 49% were aware of board members using publicly available consumer AI tools for board work.

Then the governance side. 54% said there was no guidance at all for directors' use of AI at their company. 6% reported a formal policy specific to the board.

One split inside the data is worth carrying. Directors who sit on a risk committee reported far higher perceived impact from AI — 79% saw a moderate-to-severe measurable change, against 54% of directors who do not sit on one. The people whose committee already owns a risk register see more of this than their colleagues do. They see it, and 6% of boards have written anything down about it.

Kira Ciccarelli of the Diligent Institute put the pattern in one line: "Directors are not waiting for perfect governance frameworks to experiment with AI."

The sample is small and American — 104 directors, one jurisdiction, one quarter. Read the two 17 June publications together anyway, because they describe the same behaviour at two altitudes. BCG's seventy executives report an erosion of judgment spreading through the organisation. Diligent's hundred and four directors report using the tool on the board's own core work, at four-fifths adoption, with a policy at one board in sixteen.

Chore, Rep, and the Board Pack

Here is the mechanism, and it explains why this wave reads differently from the ones a long-serving director has already survived.

The cash machine took the counting away from the bank teller and left the relationship whole. The spreadsheet took the analyst's arithmetic and left the modelling. Legal research databases took the hunt through the stacks and left the argument. In each case the machine took a chore, or one floor of the interpretive work, and left standing an adjacent patch of desk where judgment still got built. The bench outlived the printing press, the calculator and the database because the displacement was partial.

This tool takes the interpretive first pass. The opening read. The first draft of the judgment. The what does this even mean. And it takes that pass from most floors of the building in the same week — the junior's first read, the mid-level's first read, the specialist's first triage, one model doing all three on the same kind of file.

The first pass was never the chore around the rep. It is the rep. Reading a murky file for the first time, a thousand times, with someone senior nearby, is the thing that builds the power to know when the thousand-and-first is wrong.

Now put a director in the chair.

She is a non-executive on an audit committee. The pack arrives on Thursday for a Tuesday meeting. She opens it on Sunday morning, works through the management commentary and the numbers behind it, and somewhere in that reading she forms a view: the narrative and the figures agree, or they do not quite, and there is a question she will ask on Tuesday. That view is the product. It comes out of the reading and nowhere else. It is also, over years and many packs, how she came to be worth appointing.

Thirty per cent of the directors Diligent surveyed had used AI to summarise board books or meeting materials.

A summary arrives. The question that would have surfaced on Sunday morning surfaces or it does not, and either way she has stopped doing the thing that makes her able to catch it. The pack is read. The reading is gone.

Now the other 45%, who used AI to prepare for a discussion or to benchmark peers. That is the same tool placed differently. It clears the ground work — what have the other seven companies in the index disclosed, what did this committee decide in the last four meetings — and hands her the ground work so she can spend Sunday on the read. The chore leaves the desk. The judgment stays with the person sharpened by making it.

One technology. Two placements. Opposite effects on the same director over five years.

Which placement any given board has chosen is a fact about that board. At 6% policy coverage, it is a fact almost nobody has written down.

Nobody Outside the Building Will Write This Rule

Two days before the BCG piece, on 16 June, the European Parliament adopted the Digital Omnibus on AI by 423 votes to 57 with 174 abstentions. The vote confirmed the deferral of high-risk obligations to 2 December 2027 for Annex III systems and 2 August 2028 for Annex I, and pushed the watermarking obligation on AI-generated content to 2 December 2026. The mood was reported in one line: "To all the entrepreneurs and engineers out there, we are pressing the pause button on the AI Act and we are reducing red tape."

The legislature that wrote the AI Act voted to slow it. Whatever a European board decides about the erosion of judgment inside its own house, it will be deciding ahead of any rule that requires it.

The shareholder route is thinning at the same time. At Alphabet's annual meeting on 5 June, per the Item 5.07 disclosure filed on 11 June, a proposal on AI board oversight drew 461,472,553 votes for against 11,863,462,046 votes against — 3.7% of for-and-against. Alphabet's dual-class structure, where Class B shares carry ten votes, is a material part of why the number is that low. The season around it points the same way: Mayer Brown's mid-season review, published 11 June, counted roughly 789 shareholder proposals submitted in 2026 against 951 in 2025, with about 7% of the proposals that went to a vote taking majority support, down from 14% the year before.

So the standard comes from inside the building or it stays unwritten. The third event of the week shows a very large buyer building exactly that kind of standard, about capability, on its own initiative.

On 18 June the UK government announced the end of what it called outsourcing by default. Chief Secretary to the Prime Minister Darren Jones: "I want to end the era of 'outsourcing by default' and build stronger in-house capacity." Cabinet Office Minister Chris Ward: "the age of outsourcing is over."

The mechanics, whatever one makes of the politics, are portable. A Public Interest Test applies to contracts over £1m. Departments spending more than £100m a year on contracts must produce five-year roadmaps to rebuild in-house capability. Cleaning and security across 83 government buildings come back in-house as contracts expire in 2028.

Every element a board would need is present. A threshold that decides which decisions get tested. A test that has to be applied and recorded. A plan with a horizon. A date on which the plan gets checked against reality. The object in this case is outsourced service capability, and the government has decided the capability loss was real enough to schedule its reversal.

BCG's seventy executives named a capability loss. The UK government scheduled one. The distance between those two moves is the distance a board has to travel this year.

What the Minutes Are For

Blue Bell Creameries had one product, and in early 2015 a listeria outbreak in it killed three people. The company recalled everything, shut production at every plant and laid off over a third of its workforce. A Caremark oversight claim against the directors was dismissed at the Court of Chancery. On 18 June 2019 the Supreme Court of Delaware reversed.

What the court read was the board's own record, obtained through books and records. From the opinion: "no board committee that addressed food safety existed; no regular process or protocols that required management to keep the board apprised of food safety compliance practices, risks, or reports existed; no schedule for the board to consider on a regular basis, such as quarterly or biannually, any key food safety risks existed; during a key period leading up to the deaths of three customers, management received reports that contained what could be considered red, or at least yellow, flags, and the board minutes of the relevant period revealed no evidence that these were disclosed to the board."

The minutes were the evidence. Whatever the directors of Blue Bell knew or discussed, the record they kept is what a court later used to decide whether they had a system at all.

A limit belongs beside that. No Delaware court has decided a Caremark claim on AI oversight. No case yet. Everything above is the doctrine as it stands on a food-safety record, and any extension to models is argument by analogy.

The European duty needs no analogy. On 1 April 2026, in a director-disqualification judgment arising from a company whose stock system only one employee understood, the High Court restated the Barings propositions: "Directors have, both collectively and individually, a continuing duty to acquire and maintain a sufficient knowledge and understanding of the company's business to enable them properly to discharge their duties as directors." And: "the exercise of the power of delegation does not absolve a director from the duty to supervise the discharge of the delegated functions."

Hold the Diligent numbers against those two sentences.

The duty is to acquire and maintain knowledge of the business, personally and continuously. Reading the pack is how a director does that. A director who reads a machine's summary of the pack has delegated part of the acquisition, which company law permits. The supervision of that delegation survives it, which company law requires. Supervising a summariser means somebody checking the summary against the source, on some basis, at some interval, with a record of having done it.

Six per cent of boards have a policy that could name who.

For UK-listed boards there is a date attached. Provision 29 of the 2024 Corporate Governance Code applies for accounting periods beginning on or after 1 January 2026, so the first declarations appear in the 2026 annual reports. The board has to provide "a declaration of effectiveness of the material controls as at the balance sheet date" across financial, operational, reporting and compliance controls, together with a description of how it monitored and reviewed the framework. That declaration is a board judgment about things the board does not operate. The quality of the board's own reading sits upstream of every word of it.

What This Means for Boards Right Now

One. Write the policy for the board's own use before writing one for the organisation. 82% and 6% describe the same population, and the gap between them is a governance gap the board owns directly and can close in one meeting. Three questions settle most of it: which tools a director may use on board material, what may be uploaded and where that material then lives, and how a summary gets checked against the source before a director relies on it in the room. The 49% who are aware of colleagues using publicly available consumer tools have already answered the second question by default, and the answer is that board material has left the board's control.

Two. Decide the placement, use by use. The same tool either clears the ground work and leaves the director doing the read, or takes the read and leaves the director doing the checking. The 45% benchmarking peers and the 30% summarising board books are doing two different things with one subscription, and only the second one costs the board something it will need later. This is a wiring decision, it can be made deliberately, and once made it can be written into the policy above in a sentence.

Three. Put the record where an examiner would look. Marchand was decided on books and records, and Provision 29 turns a board judgment about controls into a signed statement in an annual report. Minutes recording that assurances were given are evidence of assurances, and nothing more. A record that carries a named verifier and a review date is a different object.

Which leaves the question for the next meeting, and this week put it on the table twice:

What is our own policy for AI in board work, and would our minutes survive discovery?

Reading List

The longer treatment of the mechanism at the centre of this issue — why the first pass is where judgment gets built, and what happens to a firm that automates it across every rung at once: The Broken Learning Ladder: AI Is Removing the Work That Built Expertise
What happens to a group's thinking when a confident source enters the deliberation, and why board discussions are unusually exposed to it: When AI Enters the Room, Your Best Thinking Leaves
Two companies that ran the automate wiring at scale, what each of them found on the other side, and what it cost to reverse: The Hollowing: What Klarna Learned, What Block Is About to

What We Are Watching Next

Whether the Council completes its adoption of the Digital Omnibus, and what survives in the final text of the AI literacy obligation
Whether any listed issuer publishes a policy governing directors' own use of AI in board work, and whether it appears in the annual report or stays on the intranet
Whether the UK departments now required to produce five-year in-house capability roadmaps publish them, and whether those roadmaps carry measurable capability targets or only headcount
Whether the first Provision 29 declarations, due in the 2026 annual reports, say anything about material controls that run on models
Whether the vocabulary BCG supplied this week reaches a risk register anywhere, with a date and an owner against it, or stays a management programme

The next issue goes deeper into one of these. If you want a specific function or sector covered, reply to this email.

— Liga

TwinLadder Weekly is a weekly intelligence report on judgment, governance, and the boards accountable for both. Subscribe at twinladder.ai/newsletter. Forward this issue freely.

BCG Has a Name for It Now