The Robots Were Supposed to Take Our Jobs a Decade Ago
What Actually Happened Tells You Everything About AI
By Daniel Lawrence
In 2015, Martin Ford published Rise of the Robots: Technology and the Threat of Mass Unemployment. It won the Financial Times and McKinsey Business Book of the Year. It argued, with serious evidence and serious credentials, that a wave of software automation, what most of us came to call RPA, would hollow out the white-collar workforce within a generation. Accountants were on the list. So were paralegals, radiologists, junior analysts, and most of the back office of the financial services industry.
I was working in enterprise automation when that book came out. I read it. I took it seriously. So did the boards I was advising. So did most of the people in my industry. The forecast wasn’t fringe, it was mainstream, well-evidenced, and delivered by people who had genuinely thought hard about what the technology could do.
A decade on, almost none of it came to pass.
This matters now because we are watching the same script being staged again. Last week at Accountex London, a panel of leaders from IFAC, ICAEW, and ACCA spent an hour answering questions about whether generative AI will replace the accountant by 2030. The doom-stories have a new technology, the same structure, and many of the same headline writers. And underneath all of it sits a question that nobody in the room quite asked, which is the only question that actually matters.
If the last revolution didn’t do what we said it would, what does that tell us about this one?
Two schools of thought
I hold two views on this in genuine tension, and I’m going to lay them both out honestly because I think the profession deserves the actual argument rather than a confident one-liner.
The first school of thought is that the doomers are eventually right, just early. Ford’s book wasn’t wrong on the technology, it was wrong on the timeline. RPA didn’t deliver mass white-collar unemployment in 2020 because RPA turned out to be more fragile, more limited, and more expensive to maintain than the breathless coverage suggested. But the trajectory of capability is real. Generative AI is a genuine step-change in what machines can do with unstructured information, which is most of what knowledge workers actually do. On a long enough horizon, the historical pattern of “technology evolves the work rather than ending it” might simply break. Past performance is not a guarantee of future returns, as accountants of all people should know.
The second school of thought is more interesting. It’s that the doomers weren’t just early, they were wrong about why. The reason RPA didn’t deliver mass unemployment isn’t because the technology fell short of what was promised. In most cases, the technology delivered exactly what its vendors said it would. The reason the unemployment wave didn’t arrive is that the deployment wave didn’t arrive either. Bots were fragile, yes, but more importantly, the organisations rolling them out hit a wall called reality.
Citizen-developer automation generated technical debt that often consumed more than half of the internal automation team’s time, a cost that almost never appeared in the original business case. Processes that looked clean on a process map turned out to be held together by tribal knowledge, undocumented exceptions, and the judgement of someone who’d been doing the job for fifteen years. Governance frameworks lagged the deployment by years. Change resistance was real and rational, because the people being automated had usually figured out that the bot couldn’t handle the edge cases they handled every day.
The technology was capable of more than the institutions deploying it were capable of absorbing. That’s not a story about RPA. It’s a story about us.
The Accountex panel got it nearly right
To their credit, the leaders on stage at Accountex were not catastrophising. The ICAEW shared something genuinely revealing from their 2026 evolution-of-mid-tier-firms research: only 17% of firms say they can accurately assess the impact of AI on their people, and 34% admit they have no idea how AI will affect their headcount. Eighty per cent believe the role is shifting from compliance toward ethical judgement and advisory work.
Read those numbers carefully. They don’t describe a profession being eaten by AI. They describe a profession that hasn’t yet developed the instruments to measure AI’s effect on itself. That’s a very different problem, and it has a very different shape.
Michelle Cardwell from IFAC made the point that the spreadsheet, predicted in the 1980s to kill the profession, instead expanded it. The panel’s read on the next five years was that AI will elevate the human work, not erase it: more multi-disciplinary advisory, more non-accountants joining firms in data science and sustainability, more fluidity between practice and industry. As a forecast, that is roughly the right shape. The profession will evolve, not evaporate.
But the panel didn’t quite name why the evolution-not-evaporation pattern keeps repeating. It’s not that accounting work is too human for AI to handle. It’s that human institutions absorb new tools slowly, partially, and on their own terms, and the deeper the tool reaches, the more friction it generates on the way in. That friction isn’t a bug. It’s the immune response of an organisation discovering that the new technology requires it to redesign work that’s been stable for a generation.
What I'd tell a firm planning for 2030
Don’t pick a school of thought. Pick a posture.
Assume the doomers might be right on the long horizon, not because they probably are, but because the cost of being wrong about that is asymmetric. Plan for a five-year window in which the work changes substantially, the headcount mix changes substantially, and the entry-level role looks unrecognisable. That much is going to happen regardless of which school of thought turns out to be correct.
Then assume the second school is right on the short horizon. Which means the firms that win the next five years won’t be the ones who buy the most AI. They’ll be the ones who do the unglamorous, uncomfortable work of redesigning around it: rebuilding the training pathway when the bookkeeping ledger isn’t where junior judgement gets formed, governing the technology before it embeds itself in places no one can later untangle, measuring impact rather than guessing at it, and refusing to mistake tool adoption for transformation. The 17% who can measure AI’s impact on their people are not the leading edge of the profession because they bought the best tools. They’re the leading edge because they did the harder work of figuring out what to measure.
A decade ago, the robots were supposed to take our jobs. They didn’t. The reason wasn’t that the technology failed. The reason was that the institutions deploying it bumped into themselves on the way in, and the firms that came out of that decade strongest were the ones that treated the bumping as the actual work.
The technology in front of us now is more capable. The institutions deploying it are the same institutions. The question of whether the doom-stories come true this time is, in the end, not a question about the technology at all.
It’s a question about us.
Where the real risk sits
This is where I want to be careful, because I don’t want either school of thought to turn into an excuse.
If you take the first view, that the doomers are eventually right, the temptation is to brace for impact and treat AI as an existential threat to be defended against. That posture produces defensive purchasing, AI-washing scepticism turned up to eleven, and a kind of strategic paralysis dressed up as prudence. It also produces firms that arrive at the destination late, having spent the journey trying to prove the destination didn’t exist.
If you take the second view, that the issue is us, not the tech, the temptation is the opposite. It’s to assume the friction will protect you. That because RPA didn’t deliver mass unemployment, AI won’t either, and the profession can absorb this wave the way it absorbed the spreadsheet and the cloud accounting platform. That’s a more sophisticated form of complacency, but it’s still complacency.
The real risk, in my reading, is neither of those. The real risk is the fat middle the Accountex panel actually flagged, the structural problem of firms hollowing out the entry-level base while keeping the management tier intact, because AI and offshoring handle the transactional work that used to be how trainees built the sixth sense for when a number looked wrong. Gareth John was right to call that out. It isn’t a doom scenario. It’s a slow erosion of the apprenticeship model that built the profession’s professional judgement in the first place.
That problem is not solved by AI, and it is not caused by AI. It’s caused by firms making local rational decisions, “we don’t need three trainees in the bookkeeping team anymore”, that aggregate into a sector-level problem five to ten years out, when the partners who were never trainees turn out not to have the instincts the work still requires.
This is, in microcosm, the entire pattern. The technology doesn’t do the damage. Our decisions about how to deploy it do.
Daniel Lawrence is the Founder and CEO of Bots For That, an AI-native company serving the UK accounting and bookkeeping profession. He has spent over a decade deploying enterprise automation in regulated industries, including finance and accounting services.
