In a recent post, The Four Stages of AI in Business, I outlined the four stages of AI in a given industry.
Stage 1: no AI.
Stage 2: humans doing the work, assisted by AI.
Stage 3: AI doing the work, reviewed by humans.
Stage 4: AI doing the work, no humans.
Then in another post, The AI Chasm: Why Most Businesses Won't Survive the Jump to AI-First, I laid out my case for why most businesses won't be able to make the jump from Stage 2 to Stage 3. I didn't get much pushback on that one, but I did get a lot of messages asking the same two questions: which industries are actually going to get disrupted, and how much time do I have before it happens?
So I went to the source.
Last night I sat down with Claude and asked it, directly, which industries it thinks AI is about to disrupt in the next five years. There's no better person (or model) to ask. It's like asking the burglar which houses are easiest to break into. The model knows what it can do, where it breaks, and what it can't touch yet.
Obvious caveat: Claude has an incentive to talk up its own capabilities. So take all of this with the appropriate amount of salt. But it was also clear about where it can't go, which made the answers more credible, not less.
This post is the output of that conversation, lightly edited and organized into three waves: 2025 to 2027, 2027 to 2029, and 2029 to 2031. That's the next five years. At the end, the industries Claude itself thinks are safe, and whether any industry survives 10 years out.
Wave 1: 2025 to 2027
If you're in one of these industries, you've already noticed. The ground is moving.
Customer support. Tier 1 support is largely gone. Tier 2 is going. The "talk to a real person" button is becoming the rare exception, not the default.
Outbound calling and collections. Inbound support gets all the attention. The bigger industry is outbound: collections, appointment reminders, payment follow-up, surveys, telemarketing. Voice AI does this now, at any volume, with no bad days and no turnover problem. The call center floor that ran on headsets and scripts is emptying out.
Back-office processing and data entry. The entire offshore BPO industry (the Philippines, India, the whole thing) was built on digital, rules-based, low-liability work done cheaply by people. That is the exact profile a foundation model eats first. The old arbitrage was labor cost. The new one is no labor.
Copywriting and content marketing. Marketing copy, blog posts, social content, ad creative, email sequences. The bottom 80% of this work has already been absorbed. The agency model that depended on selling words by the hour is in trouble.
Translation. Effectively done. High-stakes legal and literary translation survives. Tourist-level and business-level translation is over.
Travel planning. Travel agents were already dying. AI finished the job. "Plan me a trip to Tokyo" is a query now, not a service.
Basic coding work. Boilerplate, scaffolding, debugging, code review, simple feature builds. Coding is the leading indicator for everything else, and it went fast.
Journalism and news writing. Aggregation, wire copy, earnings recaps, sports box scores. Already gutted. Local news was dying before AI. AI is closing the casket. Opinion, investigative, and original reporting survive.
Sales prospecting (SDR/BDR). Cold outreach, list-building, initial qualification, follow-up sequences. Already automated in practice, even where it's not automated officially. The closer survives. The opener does not.
Wave 2: 2027 to 2029
This is where it gets real for professional services. If you make your living in any of these, you should be paying close attention.
Bookkeeping. Transaction categorization, reconciliation, monthly close. Rule-based document work. Gone.
Tax prep for individuals and small businesses. Deduction.com is the leading edge. An AI named Taylor preps your return year-round, a CPA signs it, for $499 a year. By 2029, paying a regional CPA $1,500 for a 1040 will look like paying for a typewriter repair. Complex multi-entity tax work survives. The bread-and-butter does not.
Contract review, NDA triage, commercial legal. Claude for Legal launched in May 2026. The work that used to bill $400/hour at a midlaw firm is being done in minutes. Litigation and high-stakes M&A survive longer. The grunt work doesn't.
Recruiting (sourcing and screening). Sourcing, the actual finding-of-candidates, goes first. Initial screens follow. The close-the-deal human-touch work lasts longer, but the team sizes shrink dramatically.
Medical scribing and clinical documentation. Doctors are already using AI scribes in production. An entire transcription-and-coding industry is collapsing into the patient visit itself.
Insurance claims processing and basic underwriting. Document-heavy, rules-based, repetitive. Classic AI territory. The carriers that get there first eat the ones that don't.
Market research and survey analysis. Report synthesis, coding open-ended responses, competitive teardowns, segmentation. Information processing dressed up as insight. The strategist who knows which question to ask survives. The analyst who turns raw data into slides does not.
Drafting and CAD work. Not architects and engineers. The drafting layer underneath them. Roughly 30% of design and drafting tasks can already be automated, and most firms have adopted the tools. Same pattern as paralegals and SDRs: the senior judgment holds, the pyramid under it shrinks.
Basic graphic design. Logos, social posts, ad variations, deck design. The Canva-and-Fiverr tier is being eaten from underneath.
Administrative and executive support. Calendar management, scheduling, email triage, expense reports, meeting notes, travel booking. The work that justified a $70K EA is becoming a $20/month subscription.
Mortgage origination and loan processing. Document collection, eligibility checks, income verification, underwriting prep. A massive document-heavy industry, almost perfectly suited for AI absorption.
Photo and video editing (commercial). Color correction, basic cuts, ad assembly, product photography retouching, social reel editing. Wedding photographers and serious filmmakers survive. The product-shot-and-Instagram-reel tier doesn't.
Wave 3: 2029 to 2031
These have moats. But the moats are getting thinner every quarter.
Financial planning for the mass affluent. Anyone with $100K to $2M to invest currently pays a financial advisor roughly 1% per year, most of which buys them a quarterly check-in and a rebalance. AI handles 90% of that work, more consistently, for $50/month. The advisors who survive will serve UHNW clients or specialize in life-event complexity. The middle gets gutted.
Equity research and retail investment research. Sell-side research, Morningstar reports, stock screeners. The information layer collapses into a conversation.
K-12 tutoring and test prep. Personalized AI tutoring is already wildly effective and roughly free. The $80/hour SAT prep industry is going to look very different.
Language instruction. You already see it in test prep and K-12 tutoring. Language learning is further along. An AI that talks back, corrects you in real time, and never gets bored beats most human tutors for most learners, at a fraction of the price. The $40/hour conversational tutor is in trouble.
Paralegal work. Discovery, document review, legal research, citation checking. Whole categories of paralegal jobs disappear. Partner-level legal judgment doesn't, but the pyramid underneath shrinks dramatically.
Real estate agents. The most contested pick on this list. The 6% commission for "I showed you Zillow listings and unlocked the door" model is dead. Buyer's agents take the biggest hit first. Listing agents with real local expertise and seller relationships survive, but commission compression is coming for everyone.
PR and corporate communications. Press releases, media lists, statement drafting, talking points, executive ghostwriting. The strategic relationship work survives. The output factory doesn't.
Radiology, partially. This one's nuanced. AI is already better than human radiologists at reading many imaging types. There are now more than 1,100 FDA-cleared radiology AI tools and two reimbursement codes. The technology is not the bottleneck. The system around it is. Licensure, liability, and trust requirements keep this at Stage 3 (AI-first, human-reviewed), not Stage 4. The radiologist's role becomes reviewer, not reader. The number of radiologists needed drops dramatically.
What survives in the 5-year window
Notice what's not on the list. Each of these is protected by something AI can't easily get past: physical presence, regulated liability, or a relationship that can't be automated.
Physical trades. Plumbers, electricians, HVAC, construction, auto repair. The robots aren't here yet.
Surgery and hands-on medicine. Same reason.
Nursing and most hands-on healthcare. Physical presence plus relationship plus judgment in the moment. Documentation work shrinks. The role doesn't.
Courtroom litigation and big M&A advisory. High-stakes, relationship-driven, performative.
Big 4 audit of public companies. Licensure plus liability plus regulator relationships.
UHNW wealth management. Complexity plus relationship moats.
Therapy and counseling. Contested. Some of it gets absorbed. The deepest work probably doesn't, or shouldn't.
High-end real estate transactions. The browsing layer dies. The transaction-execution layer survives.
Skilled artisan and trade work. Specialty chefs, instrument makers, custom fabrication, master electricians. Anything where the value is "this specific human's hands."
The pattern
If you map every industry on the list, the pattern is brutally simple.
The industries that get eaten first all share one deeper trait: their value to the customer is information processing, not trust, relationship, or physical presence. The moment a foundation model can do the information processing, the industry collapses to whoever ships the best AI-native experience.
That's it. That's the whole game.
Where does this leave you?
Two questions to ask yourself this week.
If you own a business in Wave 1 or Wave 2: Are you going to become the AI-first version of your industry, the Stage 3 disruptor, or are you the Stage 2 incumbent waiting to be disrupted? There is no third option. Pretending the wave isn't coming is just choosing to be on the wrong side of it.
If you work in any of these industries: Are you going to be the one using AI to eat your boss's business, or the one being eaten? In every wave of disruption I've studied, a few employees inside the incumbent figured it out early and rode it. The rest got laid off. Which group are you going to be in?
What about 10 years?
We just walked through five years. But I also asked Claude the harder version: what happens in 10 years? Here's what it said.
"Give me 10 years instead of 5 and nothing is safe. By 2036, the only work left is the kind people pay for because they want a real human, like a therapist or a live performer. The kind where a licensed person signs their name and takes the blame if it goes wrong, even though the AI did the actual work. And the messy hands-on jobs that are different every time, like fixing an old house, where no one will bother building a robot for it."
Maybe in 10 years AI is so deflationary that humans don't need to work 40 hours like they currently do, just to survive, but my concern is how will they work at all?
Nobody knows what 10 years looks like. But the next five are not a mystery. Claude just told you, directly. Don't ignore the warning.