By: Michael Arden

Professional services careers were built on years of research decks and first drafts, the same work AI now finishes in seconds. Daniel Cohen-Dumani argues the apprenticeship isn’t dying. It’s about to accelerate.

For generations, a professional services career began the same way: years of grunt work. Research memos nobody read twice. First drafts rewritten by someone senior. Long nights assembling briefing decks. The work was tedious by design, and it was also the apprenticeship, the way judgment got built one unglamorous task at a time. That work is now precisely what AI does in seconds, which raises a question the industry has barely started to answer. If the machine does the learning material, how does anyone learn the job?

“I think AI will do 80 percent of the work,” says Daniel Cohen-Dumani, who spent 30 years in the industry, built and sold a consultancy, and now runs an AI company serving the firms he came from. “But the other 20 percent is the whole job. Judgment. The relationship. Empathy in a hard conversation. That stays human.”

The paradox is that the 20 percent has always been learned by doing the 80. Take away the research and the first drafts, skeptics argue, and you take away the ladder. Cohen-Dumani’s answer is counterintuitive. The apprenticeship does not disappear; it accelerates, but only in firms that treat their own knowledge differently.

His company, Experio, builds what it calls organizational memory: an intelligence layer, IQ1, that makes a firm’s decades of engagements, precedents, and hard-won lessons reachable in seconds, with every answer cited to its source. In that world, he argues, a first-year associate arrives with the firm’s memory attached. “A first-year associate can know what the partner knew, the moment they need it.” The basics stop consuming the first five years, and juniors get into real client rooms earlier, where the 20 percent is actually taught.

The obstacle, he says, is rarely the technology. It is a culture where knowledge is power hoarded one career at a time. Every firm has the veteran who answers questions the way an oracle does, and whose value has always been in being the only one who knows. “Hoarding used to be job security,” he says. “In the AI era, shared knowledge is the moat. Hoarded knowledge is the risk.” Senior partners, he adds, also carry a quieter burden under the old model. They are the firm’s only search engine, fielding the same questions for the hundredth time in their careers.

The excuse he hears most often for waiting is data readiness, the plan to start once the files are cleaned up. He translates that plan bluntly: never. Thirty years of shared drives, inboxes, and documents named for their fourth final version will never be tidy, he says, and that is what real work looks like, not a failure of the firm. The old generation of knowledge systems demanded pristine inputs because the machines were limited. The current generation is precisely good at reading through mess, connecting scattered context, and finding the needle without anyone alphabetizing the haystack first. Firms waiting for clean data, he warns, are waiting for a day that does not come, while the mess keeps growing.

Leadership has its own reckoning coming. When work that billed eight hours takes seconds, firms that price by the hour face what Cohen-Dumani calls an existential question rather than a productivity upgrade. Become more efficient and make less money, or change what you sell. The firms that get ahead of it, he argues, will redeploy junior time toward outcomes and client exposure rather than protecting the timesheet.

He has little patience for what passes as preparation at many firms. Buying licenses is not an AI strategy, he says, and neither is a task force, a policy memo, or a pilot that never ends. His test is simple. If a firm removed every AI tool tomorrow, would anything about its business model, pricing, or staffing plan need to change back? If the answer is no, the strategy was theater.

For the young professionals themselves, his advice draws on his own first job, as a junior consultant in Switzerland with a fresh computer science degree. Everything durable he learned in those two years was about people, not technology. Listening before solving. Earning trust before advising. Choosing useful over impressive. The tools of his trade have turned over five or six times since, he notes, and every technical skill eventually expired. The human ones never did.

He pairs that with a warning about the machines doing the other 80 percent. Verify everything. “An answer without a source is a guess wearing a suit,” he says. In work where an error becomes a liability, he argues, the winning firms will be the ones that made verification effortless, not the ones that trusted the software most.

His skepticism extends to the loudest promises in the field. Cohen-Dumani is publicly doubtful that general intelligence is imminent, having, as he puts it, looked under the hood of the latest models and found extraordinary pattern machines rather than minds. That assessment shapes his workforce advice. Leaders who believe the machine will eventually figure out their industry on its own tend to wait, while the specific, unglamorous work of making AI reliable inside a real business goes undone. He also does not expect the frontier laboratories to do that work for anyone. Some problems live so deep inside one industry, he says, that a general-purpose model never reaches them. Someone who knows the terrain has to go build there.

By 2030, Cohen-Dumani expects the shape of the industry to look plainly different, with smaller teams doing what larger ones did, juniors operating with veteran memory from their first week, pricing tied to outcomes, and expertise held as a permanent firm asset instead of walking out the door with each retirement. None of it strikes him as radical. Every piece, he says, is already underway in firms that moved early.

The retirement problem, in particular, is one he thinks the industry consistently underprices. When a forty-year partner leaves, nobody remembers why the firm structured an engagement the way it did, or what went wrong on a similar deal a decade ago, or which approach quietly saved a client millions. The knowledge is not lost in a file, he says. It is lost in a person. Firms that make their memory permanent, he argues, stop paying twice for the same lessons, and stop losing their most expensive asset one send-off party at a time.

His closing advice works for a managing partner or a new graduate equally well. The tools will keep changing. Bet your career, and your firm, on the parts that don’t.