Don't Automate, Obliterate: The Second Reengineering Reaches Professional Services

On September 29, Long Lake, the two-and-a-half-year-old AI holding company backed by General Catalyst, closed its $6.3 billion take-private of Amex GBT, the world’s largest corporate travel manager, whose roots go back to the travel business American Express started in 1915. It paid with venture equity and bank debt, the way a private equity firm would. It doesn’t sell software. It buys service businesses and rebuilds how they run with AI.

In practice, Long Lake sends its own engineers and change managers to work inside the company it buys, and runs it on its own AI platform, Nexus. At Amex GBT, General Catalyst writes, bookings, changes, cancellations and disruptions are still largely handled by hand. That is the work Nexus is meant to take over, and it’s the same kind of coordination work law and accounting firms run on.

Two weeks later, CNBC reported that Wall Street is losing patience with AI spending that has no clear results. Big tech and AI companies are now hiring forward deployed engineers, who sit inside customer companies and wire models into workflows. US job postings for this role went from fewer than 200 in 2023 to more than 7,300 in the first nine months of this year.

The fight is moving from the models to the workplace, and we have seen this script before.

A slogan that never got old

In 1987, the economist Robert Solow wrote that you can see the computer age everywhere but in the productivity statistics. Three years later, Michael Hammer answered in Harvard Business Review with a title that became the slogan of an era: “Don’t Automate, Obliterate.” By automate he meant using computers to make old processes run faster, and by obliterate, tearing those processes down and designing the work again.

Over the next decade, three forces reshaped large companies: consultants redesigned processes, companies rebuilt around ERP, and outsourcers took over the back office.

All three are back, with one twist. Last time, firms mostly stayed in charge and hired outsiders to help. Outsiders bought firms then too: by 1999, American Express had acquired 35 CPA firms, but the work itself stayed much the same. This time, outsiders are buying firms to rebuild the work itself, or building new ones to replace them. Forward deployed engineers play the consultants’ role, and the big consultancies sell them too: Accenture is training 1,000 of them with Google. Where companies once rebuilt themselves around ERP, Long Lake and Thrive Holdings, spun out of Josh Kushner’s Thrive Capital, buy companies outright and rebuild them. And a batch of YC startups are building AI-native accounting firms, law firms and insurance brokerages from scratch.

Not B2B, but B replacing B

Billow (YC S26) describes itself as an accounting firm that replaces human accountants with AI agents, and its clients never touch any software. At Last Accounting Company (YC S26), AI agents do the work and a licensed accountant signs off. General Legal (YC W26) is a law firm for startups that quotes a fixed price for each matter up front. Panta (YC W26), a commercial insurance brokerage, set out to sell AI tools to other brokers, then got its own license instead, because the bottleneck wasn’t how fast a broker could work, it was that the state of every case lived in one person’s head.

These companies aren’t suppliers to the industry, they are its competitors. This is exactly the line Hammer drew in 1990. A tool that makes each professional a bit faster is automation. A firm that takes the coordination off people entirely is obliteration.

Clayton Christensen called this strategy integration, and explained when it wins: while a technology is still too rough to drop into someone else’s workflow, the company that owns the whole process captures the value, and the one selling a single piece gets squeezed.

The idea isn’t new. In 2014, a16z’s Chris Dixon called it a full stack startup: instead of selling technology to incumbents, build the whole service and go around them. What’s new is where it works. Professional services held out partly because so much of the work was coordination that only people could do. ERP could take over fixed processes that could be written as rules, but this work has no fixed steps: each matter is worked out back and forth with clients, counterparties and other firms, and never fit into one set of rules. Until now, handing that layer off meant handing it to another team of people, such as an outsourcer. AI has now come far enough to take it on itself.

Which profession gets restructured first depends less on what AI can do there than on what its rules allow. In most US states only lawyers can own a law firm, Arizona and Utah being the main exceptions, so an AI-native law firm stays in lawyers’ hands and a holding company can’t simply buy one, though private equity is already reaching in through a side door: management services companies that run the business side for a fee without owning the firm. Accounting is more open: a firm can split off its audit practice and take private equity money into the rest, and a NASBA task force counts more than 50 CPA firms already set up this way. Insurance brokerage has no rule like that on who owns the firm, and private equity backed buyers have made up about 70% of US and Canadian agency deals over the past decade, according to OPTIS Partners, an insurance brokerage M&A advisor.

The coordination layer goes first

Inside the firm, what goes first is the execution and coordination work that junior people used to do. Final judgment and sign off stay with senior licensed people.

Stanford payroll data covering millions of workers shows that in the occupations most exposed to AI, the number of employed workers aged 22 to 25 fell about 11% between November 2022 and June 2026, while the same age group in the least exposed occupations grew about 10%. In those same jobs, workers aged 35 to 49 grew about 10%. The authors point to one mechanism: AI substitutes for codified knowledge, the kind you learn from textbooks, and complements the tacit knowledge senior people build through practice. The pattern holds even after excluding tech companies and computer occupations. In professional firms, most coordination work sits with these same junior people.

Reengineering later became a byword for layoffs, but the restructuring didn’t stop. This time the players avoid the word from day one. Long Lake stresses it has done no layoffs at the companies it owns. The language got softer. The direction didn’t.

Who keeps the savings

Since the 1990s, outsourcing firms built on staff in India and the Philippines grew big on IT, back office and customer support work. Now they face what the industry calls AI deflation, clients who know AI can do much of the work and ask for lower prices. HCLTech, one of India’s largest IT outsourcers, estimates that about 40% of the industry’s revenue faces 3% to 5% deflation from AI. Their customers are procurement teams who know how to push on price, so the savings go straight back to the client.

Professional services works differently. The people inside understand the work best, but as long as nobody cuts prices, no firm has a reason to go first. In a 2026 survey by the legal software company Clio, 75% of small US law firms already use AI, yet 86% of solo lawyers and 78% of small firms have made no pricing changes, against about half of mid-sized and large firms. Meanwhile, US legal services prices are up about 8% from last year. If AI is saving time, it isn’t showing up in the bill.

Where no buyer pushes, it’s outsiders who break the unspoken agreement. In law, that means startups opening their own lawyer-owned firms with fixed prices. In accounting and insurance, it also means holding companies that buy the whole firm, though they may keep the savings as margin instead of cutting prices, and a broker paid mostly by insurer commission doesn’t set the price the client pays anyway. So far, the savings stay with whoever owns the firm, and reach clients only when a buyer or a competitor forces the price down.

The next target is experience

The first wave of AI went after coordination and textbook knowledge. The next target for capital is experience itself. Christensen’s conservation of attractive profits explains why: when one layer of a business becomes cheap and interchangeable, the profit tends to move to the layers next to it. AI is making software cheap to build, so more of the profit should move to what software can’t easily copy: the client, the license and, for now, the experience.

When you buy a firm, you don’t just get its clients, you get years of case records and decisions. When you embed engineers on site, you are writing senior people’s judgment into a system. Palantir keeps a gross margin around 85% on very custom work, partly because it books the cost of pilots at new customers as sales and marketing, but the bet is clear: what it learns in each deployment goes into the platform, not into headcount.

Bank tellers once had machines on their side too. James Bessen studied what ATMs did to tellers. Between 1988 and 2004, the number of employees per urban US bank branch fell from about twenty to thirteen, but because branches got cheaper to run, banks opened more of them, and total teller jobs rose a little. Then banking moved onto phones, and the branch became the thing to cut. BLS data show US tellers falling from about 600,000 in 2008 to about 330,000 in 2025, with another 13% decline projected by 2035.

For now, AI works in favor of experienced professionals. That protection lasts only as long as their experience stays theirs.

Risks and limits

Most AI deployments still lack verified returns, CNBC notes, and money follows returns. The AI-native firms are tiny, and licensing and liability cap their growth. Incumbents could also win, since they are buying the same tools and already have the clients, licenses and insurance. Nobody has shown yet that a roll-up like Long Lake gets better with every firm it buys, rather than just cheaper. Whoever wins may not even own the client, if more people start asking an AI before they call a professional. And cheaper services might mostly bring in people who couldn’t afford a lawyer or an accountant before, growing the market instead of shrinking the jobs. That last one is my guess, but it’s the version I’d most like to be true.

The evidence has limits too: the Stanford data covers all AI-exposed jobs rather than professional firms and doesn’t measure coordination directly, a “pricing change” in the Clio survey can mean a new fixed fee rather than a lower bill, client files come with confidentiality rules, and Long Lake’s no-layoff record at Amex GBT is only weeks old.

Restructure yourself before someone else does

Felix Rieseberg at Anthropic recently wrote “How to be more ambitious”. I agree with him, but there’s another side to it: if people inside an industry don’t redefine their own work, outsiders will do it for them.

Software engineers went through this first. AI learned to code largely from code engineers had published themselves, yet many of them didn’t believe it could touch their jobs. Even today, some engineers wave off code written by AI. Now capital is pushing the same force into professional services, through vertical tools like Harvey, forward deployed engineers inside companies, and AI-native firms that replace the incumbents.

For professionals, this is a prisoner’s dilemma. If every firm held back, the experience would stay with the profession. But any firm that holds back just loses ground to the ones that don’t, outsiders get the experience anyway, and every deployment shows the vendors a little more of how your job is done. I don’t have a way out of that. As an engineer, I’ve been on both ends: my profession’s code trained the models, and now I build one of the tools. But if the change is coming anyway, the real choice is who sets the terms, and the best terms I can see are to keep the client, the responsibility and the call.

Insiders still have the asset outsiders find hardest to copy, a deep understanding of the work. I think there are three ways to use it.

First, separate coordination from judgment. Look at where your hours go: how much is chasing, filling gaps and going back and forth, and how much is the judgment your clients pay for. The chasing is what outsiders take first.

This is also what struck me most while building Pingbo. For most professionals, much of this coordination work never makes it into the firm’s practice management system, it stays scattered across the inbox: an email waiting for a reply, a document that’s still missing, a promise to talk again next week. The inbox is where work arrives, but it has no idea what’s still open or who is waiting on you. That is most true in small firms, the same ones slowest to change how they charge.

Second, don’t just get faster, change what you charge for. In law, people have predicted the end of the billable hour for decades, and it still feels safe, because most small firms haven’t moved and prices are still rising. What’s different this time is that AI has cut outsiders’ costs enough to quote a fixed price on bespoke work, not just forms. If you charge for outcomes, the time AI saves stays with you, at least until competitors reprice the same work.

Third, systematize your own experience before someone else does. Write down how you decide, the questions you ask on a new matter and the red flags you check, and build that into your own team’s tools. If it’s going to become a product anyway, it should be your product.

Holding on to the way things are done today just hands control of the restructuring to someone else.