The Repricing

In August, Rillet, an AI-native ledger two years out of stealth, raised $100 million at a $1 billion valuation in 48 hours. The founders weren't even fundraising.¹ Thrive Holdings, the vehicle Josh Kushner built to buy professional services firms and rebuild them around AI, raised $2 billion at a $12 billion valuation from SoftBank, D1 and Altimeter, and its accounting platform, Current, holds around thirty American firms with more than 2,000 people and over $500 million of revenue between them.² ³ And Xeinadin, which has 130 offices, 2,500 staff and £250 million of revenue assembled from more than a hundred acquired UK practices, couldn't find a buyer at all. Exponent paused the auction back in February after bids failed to clear an ask reported north of £800 million, with advisers saying buyers were spooked by the cost of truly integrating what they'd be buying.⁴

So pure software was priced in two days, a group of firms wrapped in an AI story was oversubscribed at twelve billion dollars, and a group of firms as they actually operate today attracted no acceptable bid at all. The market knows what the software is worth, and it believes it knows what the story is worth, but it genuinely doesn't know what an accounting firm is worth anymore. The spread between the multiple nobody would pay for Xeinadin and the one everybody paid Kushner is, I think, the most important number in the profession right now.

This essay is about what's inside that spread.

The Wave

The scale of what's already happened deserves a moment, because I still meet partners who treat private equity in accounting as a rumour.

IFAC counted 1,052 accounting firm transactions involving private equity between 2015 and 2025. That figure isn't a tally of conversations or rumoured approaches, it's completed deals, driven by 177 direct investments that pulled roughly 875 roll-up acquisitions in behind them, and the follow-on rate reached 7.6 acquisitions per direct investment in 2025, four times what it was in 2021.⁵ Cornerstone's deal tracker, which follows the American market transaction by transaction, counts 118 deals since 2020 building $30 billion of new value.⁶ Whatever this was five years ago, it stopped being a rumour some time ago.

In the US, roughly half of the thirty largest firms now carry private equity money or an alternative practice structure.⁷ Citrin Cooperman has already been sold from one PE owner to the next, from New Mountain to Blackstone at a reported $2 billion, which is the industry's first big secondary sale and the way an asset class announces it has matured.⁸ Baker Tilly and Moss Adams merged under Hellman & Friedman. New Mountain holds Grant Thornton in the US while Cinven holds Grant Thornton in the UK. And the deal flow has organised itself around a handful of repeat acquirers, with Ascend, Aprio, Crete, Eisner and Ryan now shaping the profession's consolidation, which was never the broad democratisation of capital the early narrative promised.⁹

None of that is new news. What's new is what the buyers think they're buying.

The Escalation

For the first few years the thesis was old-fashioned. Accounting revenue is recurring, contractual and stubbornly recession-resistant, so you buy it, centralise the back office, expand the margin and sell it on. TowerBrook's 2021 investment in EisnerAmper opened that era.⁶ Crete Professionals Alliance closed it.

Crete was founded in 2023 by Jake Sloane and Frank Zhang with backing from Josh Kushner's Thrive. In June 2025 it committed more than $500 million to acquiring US accounting firms with OpenAI's tooling attached.¹⁰ In December, OpenAI went further and took an ownership stake in Thrive Holdings itself, embedding its own research and engineering staff inside the portfolio, which means inside accounting firms.¹¹ This June, Crete rebranded as Current, announced another $500 million for acquisitions and AI deployment over the next two years, and published the results of its first tax season, with 7,000 returns processed through its Tax AI at a 31 percent average time saving and accuracy claimed at up to 98 percent.¹² Two months later came the $2 billion raise at a $12 billion valuation, alongside a stated ambition to hold these firms forever.²

It's worth reading that sequence twice. Some of the most sophisticated capital in technology, and the most important AI lab in the world, are taking direct equity positions in local tax practices. What they're buying has changed along the way, because the prize is no longer the cash flows so much as the conversion rights, the licence to turn labour into software inside firms they control.

The shape of the thing matters as much as the money. Current holds its firms in a structure designed never to sell them, a forever hold, which is a deliberate inversion of the buy-improve-exit cycle that defines conventional private equity.¹² The practices themselves are deliberately ordinary. Larson Gross, a decades-old firm in Washington state, is the kind of business it consolidates, local books, local payroll, local relationships, and Current has gathered around fifty such practices under the one roof.¹³ What changes after the acquisition is the layer underneath, because the embedded OpenAI teams rebuild the workflows from inside, and the arrangement gives OpenAI ownership upside that grows as the portfolio grows.¹¹ Kushner describes the thesis in a single line, that the shift will happen from the inside out as domain experts and practitioners use AI as a native tool to reshape their own fields.¹¹

Nor is accounting the end of it. Thrive Holdings runs a second platform, Shield, rolling up IT services businesses, and has a third planned for physical-world compliance, the permits and inspections behind data centres, manufacturing and power.¹⁴ OpenAI runs a parallel deployment venture with TPG and Bain. The labs have worked out that the fastest way to sell intelligence is to own the businesses that consume it.

The incumbents are running the same maths from the inside. The Big Four have committed more than $9 billion to AI between them, PwC trained 315,000 people in it while reducing headcount by 5,600 in a half-year, and KPMG has already demanded a fourteen percent audit fee reduction from a supplier on the grounds of AI savings.¹⁵ When the largest buyers of professional labour start repricing their own suppliers for AI, the direction of travel has stopped being a debate.

The arithmetic behind the trade is the one I laid out in The Last Intern, where AI takes on sixty percent of the execution work and firm margins move from roughly twenty-five percent to fifty-five.¹⁵ Charlesbank-backed Aprio has publicly put $300 million behind its own version of the same bet.¹⁶ And Current's stated reason for its rebrand cuts deepest of all. AI, they say, now lets local firms compete with the large incumbents.¹⁷ Hold that thought, because it's also the strongest argument a firm has for never selling at all.

What The UK Top 50 Actually Looks Like

I moved to the UK after thirteen years in the US, and the question partners keep asking me here is whether all of this is an American story. So this weekend we ran the UK's top fifty firms, ranked by filed turnover, through the enrichment stack we built for Archie, resolving the ownership of every firm with each claim tied to a cited source.¹⁸

Twenty-three of the fifty sit under external capital today. Eleven are directly PE-backed, with Hg and PAI Partners behind Azets, Exponent behind Xeinadin, Penta Capital and Toscafund behind Sumer, Waterland behind Moore Kingston Smith, Bridgepoint behind Interpath as of May, Lee Equity behind Cooper Parry, Apax behind S&W following the carve-out from Evelyn, Synova behind Bishop Fleming, Tenzing behind Gravita, Pollen Street behind Leonard Curtis and Palatine behind BK Plus. Nine more belong to consolidators that are themselves private equity vehicles one layer up, so Blick Rothenberg and Ensors sit inside Azets, HW Fisher inside Sumer, Kreston Reeves inside Goldman Sachs-backed AAB since February, Barnes Roffe inside IK-backed Dains, Shipleys inside Moore Kingston Smith, Gerald Edelman majority-held by ETL, Old Mill inside Kinbrook and Alliotts inside Shaw Gibbs. Three trade on AIM, being MHA, Begbies Traynor and FRP.¹⁹

That consolidator layer deserves a pause, because it's where the market is most opaque. A firm can look independent on its own website while sitting two layers below a fund. Barnes Roffe reads as a London firm until you follow it into Dains, which answers to IK Partners, and Old Mill went into Kinbrook in a £28 million deal in November. Moves like those have happened all through the last eighteen months, and each one shifts a firm out of the independent column without a press release ever using the words private equity.¹⁹

There are stranger paths through the middle too. MHA chose the public markets over a sponsor, and its £98 million float was the largest AIM IPO of 2025. Moore Kingston Smith went the other way in 2023 and became the first major UK firm to take institutional money while keeping its LLP structure, with Waterland behind it, a template several of the deals above have since copied.¹⁹ S&W runs another direction again, an accountancy and advisory business of around 1,800 people carved out of Evelyn Partners by Apax last year and set loose to grow on its own. And the UK now has a secondary of its own, with Interpath moving from HIG Capital to Bridgepoint in May, the same maturation signal Citrin Cooperman sent in the US.¹⁹

The 27 independents include the six biggest firms in the country. Strip out the Big 4, BDO and Forvis Mazars and the mid-tier picture is stark, because twenty-three of the remaining forty-four have taken external capital. That's more than half, and put together the externally owned twenty-three bill roughly £2 billion a year between them. Two years ago you could count this market's PE deals on one hand.

So no, it isn't only an American story, because the capital crossed the Atlantic some time ago. What hasn't crossed yet is the second thesis. The UK wave is still running the old playbook of buy, bolt on, centralise and exit, and the old playbook is exactly what has stopped clearing.

Look closely at why the auctions stalled, because the reason matters more than the fact, and it isn't a story about bad businesses. Xeinadin built something real, more than a hundred practices brought together in six years, thirteen firms in 2024 alone by its chief executive's own count, running on a single shared management system.²⁰ Sumer assembled 44 acquisitions in three years. These are some of the most capable dealmakers in the profession. What shifted is what buyers will pay for, because Evercore ran Xeinadin's process against roughly £60 million of EBITDA at an ask implying up to 14 times, and the bids stopped short.⁴ When Sumer's £1 billion auction paused in May, its owner moved to weighing a continuation fund, holding the group while the market finds its level.²¹ Several other mid-tier processes were reported paused or under review through the same window.²² The message from the buy side is consistent, and it lands on everyone at once. Owning many practices and having integrated them are different assets, and the next multiple gets paid for the second one. Nobody failed here. Integration is the hardest problem in professional services, AI is about to make it more solvable than it has ever been, and the group that truly cracks it will reset the price for everyone.

There's no British Current yet, and no OpenAI engineers embedded in a Midlands practice. When that thesis does land here, and it will, the repricing will be fast.

Meanwhile the resistance is finding its voice. James Gilbey at Forvis Mazars says he's "not picking up the phone" to private equity. Simon Massey at Menzies sees "no point for partners under 50", and his firm merged with Beever & Struthers without a sponsor in sight. Buzzacott publicly reaffirmed its independence "in an era of consolidation", Harris & Trotter remain fiercely independent, and Henderson Loggie, owner-managed since 1909, now markets itself against the consolidators.²³ Half the mid-tier has sold while the other half digs in, and both halves are about to face the same repricing.

Where Capital And Code Disagree

Private equity's accounting thesis rests on the durability of the revenue. The AI thesis says the work underneath that revenue is about to be repriced. Both can't be fully right at the same multiple.

Dave Yuan at Tidemark, an early backer of Karbon, the practice management company I co-founded, has written the clearest map of this tension I've read. Intelligence alone, he argues, doesn't book the revenue, chase the invoice, or file the compliance report. The value lands where AI reasoning meets deterministic execution and knows when to hand a judgement call to a human, a combination he calls the application layer. The prize attached to it is enormous, because labour is roughly half of a small firm's cost base while software is about five percent, so whoever builds the layer that converts one into the other is playing for a market perhaps ten times the size of everything our industry has built to date.²⁴

His framework for whether the foundation labs will eat a market themselves comes down to four questions, whether the opportunity is big enough to make a lab's roadmap, whether the field has enough training data and verifiable outcomes, whether a lab will build the unglamorous deterministic plumbing a complete solution needs, and whether customers will actually change how they work. Software development answers yes to all four, which is why coding assistants arrived first and arrived from the labs themselves. Accounting mostly answers no. The ontologies, the edge cases and the jurisdictional depth are precisely the unglamorous work a generalist lab has no incentive to build, and the twelve-person practice is not going to build a semantic ontology of its own workflows either.²⁴ Somebody has to stand in between. That somebody is the application layer.

The chapter's sharpest line is about context, that an agent which doesn't understand your business will answer what the user asked but not what they meant. Accounting is nothing but context, the client quirks, the prior-year positions, the things a partner knows and never wrote down. That's why the layer matters more than the model, and why the winners of this period will be decided by who encodes that context first.²⁴

Everyone in this essay has read the same map. The labs no longer just sell tokens, they sell embedded deployment with equity attached, with OpenAI inside Thrive Holdings, billions committed to deployment vehicles, and Anthropic building alongside Blackstone and Goldman.²⁵ The AI roll-ups are building the application layer privately, inside firms they own, for the benefit of their own P&L. The software side is building it for every firm at once, with Rillet on the ledger, Accrual on tax preparation with General Catalyst's $75 million behind it, and us on the work itself.²⁶ That's the real strategic split in this industry now, a private application layer for the few or a shared one for the many. Archie sits deliberately on the second side of it, and I'll leave that there, except to note that the customers have already voted on where they want the layer to live. When Tidemark surveyed vertical software buyers, 83% said they'd adopt an agent that was only 80% effective if it arrived inside the system they already use, ahead of a better standalone tool from outside.²⁴

One caution for any partner reading the vendor decks, though. Most AI spending still fails. MIT's NANDA project found that 95% of organisations got zero return on $30 to 40 billion of generative AI investment, largely because tools without learning loops and workflow integration don't compound.²⁷ The pattern behind the failures is consistent, pilots bolted beside the workflow instead of inside it, no memory of what worked last month, no loop that turns yesterday's exceptions into today's rules. The application layer is hard, and it's supposed to be. That difficulty is what all this capital is actually paying for.

What Survives

The comfortable conclusion from everything above would be that the independent firm is finished. I don't believe that, and I'd go further, because the firms that adopt properly may end up holding the best hand at the table.

Run the maths from the buyer's side. The entire $12 billion story rests on taking a firm where labour consumes sixty to seventy cents of every revenue dollar and converting a large share of that cost into compute.²⁸ Nothing about that conversion requires selling your firm. A partnership that moves its own cost base from fixed salaries to flexible capacity captures the same margin private equity is paying a premium to capture, and it keeps all of it. The roll-up needs the arbitrage. You already own the asset.

Work the same numbers from the partner's chair. Take a £10 million firm running a 25 percent margin, making £2.5 million. Move a third of the labour line to software economics over three years, hold pricing broadly steady by shifting the mix toward advisory, and the margin walks toward forty percent without a single share changing hands. That is the whole trade the buyers are underwriting, and it's available at retail.

There's a deeper inversion coming that the deal models haven't caught up with, and I wrote about it in The Last Intern. A firm's value has always correlated with headcount, because people were capacity, and capacity was revenue. In an AI-native model, headcount becomes a cost centre, and the most valuable firm stops being the biggest and becomes the one with the highest ratio of revenue to humans. Succession planning turns from a people question into a technology question. The firms being acquired today at eight to twelve times EBITDA may look very different once the acquirer realises the revenue is durable but the method of delivering it has changed underneath, and private equity hasn't fully priced that in yet.²⁸

The demand side holds too. The reasons clients want a human who knows them, the trust, the judgement, the person to sit across from when the numbers get frightening, don't expire in 2027, and they'll outlast every model release. What changes is what stands behind that trusted person, with far fewer hours of preparation and far more of the thinking. Adoption is hardly exotic anymore either, given that 98% of firms already use AI somewhere in the practice and save about an hour per person per day doing it.²⁹ The profession will run two-speed for a while, the way it did through cloud, and the gap compounds faster this time because the systems learn.

One warning inside the good news, because adoption without a pricing change is a trap I've watched firms walk into all year. If the AI halves the hours and you bill the hours, you've halved your own revenue and kept the full cost of the transition. The firms getting this right are moving to fixed fees, subscriptions and outcome pricing while the efficiency gain is still theirs to keep, and revenue per person is quietly replacing realisation as the number that matters.¹⁵ Hourly billing turns every AI gain into a client discount.

The split that matters over the next three years is adopted versus didn't, far more than sold versus independent. Ownership decides who captures the margin. Adoption decides whether there's a margin left to capture.

If You're The One Taking The Call

Four practical readings of all this, for four different seats at the table.

If you're a partner considering selling, remember that the earn-out maths assumes the cash flows you're selling persist through the earn-out, and AI is the variable that decides whether they do. Make it concrete. Sell that same £10 million firm at eight times EBITDA and the headline is £20 million, with perhaps half of it deferred against performance over three years. If compliance pricing compresses by 30% during those three years while the buyer's AI does the work that used to justify your fees, the deferred half is the part that quietly doesn't arrive, and the equity you rolled into the platform gets marked by the next buyer, not this one. None of that makes selling wrong. It makes the second half of the price a bet on the same technology curve you were told justified the first half. Ask the buyer precisely how the model behaves when compliance pricing halves. The honest answer is usually volume, and then the question becomes who owns the volume engine. If the answer is their AI and not yours, you've just learned what they think they're buying.

If you're staying independent, notice that your cost base is the very optionality the buyers are paying premiums for, and it's already yours. But independence without adoption is just slower decline with better stories. Adopting properly means moving the work itself, the preparation, the chasing, the reconciliation, onto systems that learn your firm, and keeping the judgement and the relationships where they've always lived. The tools have become subscriptions, the moat is the trust you already hold, and trust only compounds when there's an application layer underneath it doing the work at the new price.

Now, if you're early in your career, the seat most at risk is the one that used to train you, which is the problem I wrote The Last Intern about. Pick the firm that's investing in both the AI and your judgement. They exist on both sides of the ownership divide, and they're not hard to spot, because they're the ones that can tell you exactly what their juniors will be doing in 2028.

And if you're the buyer, the two stalled auctions are your syllabus. The integration bill is real, the conversion story is the only thing that justifies the entry multiple, and a conversion story without an application layer already working inside the firm is a slide, not a plan. The next vintage of winners will underwrite workflow, data and learning loops with the same rigour they underwrite EBITDA.

The Firm That Doesn't Exist Yet

There's one more entrant coming, and almost nobody is pricing it.

Everything in this essay so far analyses firms as they look today, which is retrofit economics, an engine bolted onto a horse cart. The more interesting question is what a firm looks like when it's designed for AI from the first day, with no pyramid of juniors, no sixty percent labour line and no apprenticeship model to protect, because the labour cost is mostly compute. A firm like that can undercut today's pricing by eighty or ninety percent and stay more profitable, and where a fifty-person practice serves five hundred clients, a firm built this way serves five thousand with the headcount of a boutique.³⁰

The deeper advantage is parallelism, which I already wrote about and which still gets missed. AI doesn't just do a task faster, it runs hundreds of engagements at once, overnight, at consistent quality, so the growth constraint stops being how many people you can hire and train and becomes how much compute you can provision.³⁰ A firm built on that assumption doesn't look like a smaller version of today's firm. It looks like a different species that happens to file the same forms.

And the clients themselves are changing shape underneath us, which I keep coming back to when someone tells me clients will always want a human. AI-native companies need their books done too, and increasingly the firm best placed to serve them is also AI. The client is a machine, the bookkeeper is a machine, and the output gets consumed by another machine.¹⁵ The human relationship argument is true and durable for human clients. It just describes a shrinking share of the ledger.

The regulatory objection is narrower than it sounds. The moat protects audit and attest work, and it's real there, but it was never built around tax, bookkeeping, management accounts, payroll or advisory, which is the majority of what most firms outside the Big Four actually sell.³¹ The licensure regimes differ by jurisdiction, CPA in the US, ICAEW and ACCA here, another regime again in Australia, and none of them prevents a structure where a small number of licensed humans oversee thousands of AI-processed engagements. The first versions of that structure already exist.

None of this is science fiction, and a regulator has already met it once, because in May 2025 the Solicitors Regulation Authority authorised Garfield Law, the first AI-driven law firm, with real clients, real filings and a licence, chasing small debts at fixed fees that start at £2 for a polite chaser letter.³² It has since claimed the first courtroom win by an AI-driven firm.³³ Pilot already runs bookkeeping with no human in the loop.³⁴ Accountancy's equivalent is a matter of when, and my answer is that the first AI-only firms will be operating at genuine scale by 2027.

For what it's worth, we'd like to be involved when it happens. If the right partner turns up, Archie will help start one, and not in competition with the firms we serve, since the whole point of Archie is arming them, but because someone is going to build the first AI-only accounting firm, and better it's built by people who love this profession than by people who only ever priced it.

Mostly, though, the AI-only firm matters here for what it does to every valuation model in this essay. It's the entrant all of them quietly assume never shows up.

Three Prices, Again

Go back to those three prices. Rillet cleared in 48 hours because software with the workflow built in is legible, and the market can see the compounding. Thrive Holdings cleared at $12 billion because capital believes labour can become software and wants to own the conversion. Xeinadin found no bid because a decade of buying firms turns out to be a different thing from building one, and buyers have finally noticed.

Below the Big 6, half of this country's top fifty has already answered the ownership question one way, and the other half is holding. Every one of them, sold, listed, independent or defiant, gets repriced by the same technology, and the repricing doesn't care which box you ticked.

So when the call comes, and the data says it will, the multiple is the least interesting number in the conversation. What matters is which firm is being valued, the one AI is about to reprice or the one you could still build with it.

Which one will you be selling?

____

Footnotes

¹ TechCrunch, "How AI accounting startup Rillet raised $100M and became a unicorn in 48 hours," 21 August 2026. Iconiq and Sequoia led the round, with roughly 600 customers and $200 million raised to date.

² Tech Funding News, "Josh Kushner's Thrive Holdings raises $2B at a $12B valuation," August 2026. SoftBank, D1 Capital and Altimeter led. The $12 billion prices the holding company, which also owns Shield, an IT services platform of roughly twenty companies, and not the accounting business alone.

³ CPA Practice Advisor, "Crete Professionals Alliance Rebrands as Current," 3 June 2026. Nearly thirty firms, more than 2,000 employees and over $500 million of revenue in the most recent fiscal year.

⁴ Accountancy Age, "Why the Xeinadin auction collapse is a reality check for the mid-tier," 3 February 2026, and International Accounting Bulletin, May 2025, on the Evercore-run process, with EBITDA reported at roughly £60 million.

⁵ IFAC private equity study, reported by CPA Practice Advisor, 6 March 2026.

⁶ CPA Trendlines, "PE Deal Tracker: 118 Deals Build $30 Billion in New Value," November 2025, and the Cornerstone dealflow timeline 2020-2026.

⁷ Footnote News, "Private Equity and AI: The Accounting Firm Roll-Up in 2026."

⁸ CPA Practice Advisor, 7 January 2025, and Public Almanac, "The First PE-to-PE Sale of an Accounting Firm."

⁹ CPA Trendlines, "PE Wars: Top CPA Platforms Battle for Supremacy," 1 April 2026. The Baker Tilly and Moss Adams merger and the Grant Thornton positions are per trade press reporting across 2024 and 2025.

¹⁰ Yahoo Finance and TechStartups, "Thrive-backed accounting firm Crete to spend $500 million in AI roll-up," 4 June 2025.

¹¹ OpenAI, "OpenAI takes an ownership stake in Thrive Holdings to accelerate enterprise AI adoption," 1 December 2025. The Kushner line is from that announcement.

¹² CPA Practice Advisor, "Crete Professionals Alliance Rebrands as Current", 3 June 2026, including the forever-hold framing.

¹³ Footnote News, ibid., on Larson Gross, the roughly fifty consolidated practices and Thrive's commitment of around $1 billion in total.

¹⁴ Tech Funding News, ibid., on Shield, the planned third platform and OpenAI's parallel deployment venture with TPG and Bain.

¹⁵ Stuart McLeod, "The Last Intern," February 2026, at heyarchie.ai/resources/the-last-intern.

¹⁶ Aprio public statements on its AI investment programme, discussed with Richard Kopelman on Accountants in the Field, episode 1. Charlesbank Capital Partners has backed Aprio since 2024.

¹⁷ Current rebrand announcement, June 2026.

¹⁸ The top fifty UK firms by filed turnover per The Accounts leaderboard, August 2026, resolved through abm.dev multi-source enrichment on 23 August 2026, with every ownership claim source-cited. The filed-accounts ranking excludes Grant Thornton UK, and counting its 2025 sale to Cinven makes the external-capital picture heavier still. Two classifications rest on thinner sourcing, Leonard Curtis on a single source and BK Plus on Palatine's own portfolio page.

¹⁹ Full firm-by-firm dataset held by the author. The consolidator transactions are per trade press and firm announcements, 2021 to 2026, including Kreston Reeves joining AAB in February 2026, Barnes Roffe joining Dains in June 2025 and Old Mill joining Kinbrook in November 2025.

²⁰ AccountingWEB, "Xeinadin's CEO on M&A, growth, private equity and integration," May 2025.

²¹ Accountancy Today, "Sumer owner postpones £1bn auction," 26 May 2026, and Scottish Financial News on the continuation fund option.

²² Accountancy Age, February 2026, on further mid-tier sale processes reported under consideration.

²³ City AM interview with James Gilbey, Accountancy Daily interview with Simon Massey, the Menzies and Beever & Struthers merger per trade press, August 2025, and firm statements from Buzzacott, Harris & Trotter and Henderson Loggie, 2025 to 2026.

²⁴ Dave Yuan, "System of Action Part 4: The Application Layer," Tidemark Vertical SaaS Knowledge Project, 2026, including the four-question framework and the control-point surveys of January 2025 and April 2026.

²⁵ OpenAI, December 2025, with the lab deployment vehicles and lab-alongside-PE construction per Tidemark, ibid.

²⁶ CPA Practice Advisor, "Startup Accrual Officially Launches with $75M in Funding," 5 February 2026.

²⁷ MIT Project NANDA, cited in Tidemark, ibid.

²⁸ "The Last Intern," February 2026, on AI-native firm economics.

²⁹ Karbon State of AI in Accounting Report 2026, cited in "The Last Intern."

³⁰ "The Last Intern," February 2026, on parallel execution and the compute constraint.

³¹ "The Last Intern," February 2026, on the regulatory moat by jurisdiction and emerging hybrid structures.

³² Solicitors Regulation Authority, "SRA approves first AI-driven law firm," May 2025, with fee levels per contemporaneous press coverage.

³³ VinciWorks, "AI law firm claims world's first courtroom victory," 2025.

³⁴ Pilot autonomous bookkeeping launch, per sources in "The Last Intern" and explained in The Compounding Unlock