Will AI automate away expert professional roles such as clinicians, or only parts of their work?

Harry Stebbings · Apr 11, 2025

Law firms and accountants won't be replaced by direct-to-consumer AI, though very low-level legal tasks like a one-page rental agreement plausibly could be

He treats the idea of eliminating law firms or accountants outright as not a serious position

Scope: concedes the low-level, templated end of legal work

52:43 20Product: How Scale AI and Harvey Build Product | Why PMs Are Wrong: They are not the CEOs of the Product | How to do Pre and Post Mortems Effectively and How to Nail PRDs | The Future of Product Management in a World of AI with Aatish Nayak

Harry Stebbings · Apr 20, 2026

Much of professional services demand exists to allocate blame, not to supply capability — you hire lawyers so it's their fault if something goes wrong

Clients could often do the work themselves (e.g. write an NDA); what they're buying is someone to be responsible

32:23 20VC: Everyone is Wrong; We Will Have More Developers in Five Years | Why Frontier Labs Will Be Way More Valuable Than They Are Today | Are SaaS Companies Cooked: Which Thrive & Which Die with Aaron Levie, Founder at Box

Harry Stebbings · Jul 18, 2026

Even for routine legal work like NDAs, companies will keep using law firms because the firm serves as the accountable party if something goes wrong.

If it goes wrong, they're the ones getting fired.

Scope: about accountants and lawyers doing simpler work

57:16 20VC: $5BN in Revenue, 7 to 7,000 Employees in 9 Months, 206,000 Tests in a Single Day: The Craziest Story in Startups: Curative with Fred Turner

David Frankel · Aug 8, 2026

Mid-career workers retain an advantage where deep vertical knowledge and human relationships matter — especially in sales and high-stakes services

In services like audit or litigation, clients want a human with years of experience to look them in the eye and confirm they haven't just left it to AI; the stakes justify the human interface

Scope: applies to bigger-ticket, high-stakes work, not $1,000 contracts; grunt work behind the human interface does go away

62:32 20VC: The AI Boom Will Create Enormous Roadkill: Who Wins & Loses | Why Founders Should Never Take Multi-Stage Money at Seed | Why Triple, Triple, Double, Double is Good Enough

David Luan · Jun 24, 2024

AI will not take all jobs; humans will remain the drivers of agentic systems that give them enormous leverage on their own creativity, and that requires a copilot-style approach

Pricing-by-work is just a corollary of the 'AI takes all jobs' vision, which extrapolates absurdly from invoices to consulting decks to being AI CEO — that's not how it will play out

38:11 20VC: Why Foundation Model Performance is Not Diminishing But Models Are Commoditising, Why Nvidia Will Enter the Model Space and Models Will Enter the Chip Space & The Right Business Model for AI Software with David Luan, Co-Founder @ Adept

David Luan · Jun 24, 2024

The biggest misconception about the next decade of AI is that it will take each human capability and fully automate it; most human work won't be neatly captured by AI, and AI will instead be a tool to level up human intelligence.

The implicit goal of AGI today is replacing human work, but so much of human work doesn't decompose that way.

49:59 20VC: Why Foundation Model Performance is Not Diminishing But Models Are Commoditising, Why Nvidia Will Enter the Model Space and Models Will Enter the Chip Space & The Right Business Model for AI Software with David Luan, Co-Founder @ Adept

Joelle Pineau · Nov 3, 2025

A better barometer of AI productivity is whether most of your employees can do 10x the work with AI, not whether AI can replace a slice of the workforce

Humans and AI have very complementary abilities, so flat-out replacing a portion of the workforce is unrealistic

Scope: acknowledges some firms will try, and some may slow hiring

16:17 20VC: Cohere's Chief AI Officer on Why Scaling Laws Will Continue | Whether You Can Buy Success in AI with Talent Acquisitions | The Future of Synthetic Data & What It Means for Models | Why AI Coding is Akin to Image Generation in 2015 with Joelle Pineau

Arvind Jain · Jul 11, 2026

Telling employees to replace themselves with AI is the wrong goal because AI is not ready to replace any job today.

It gives AI too much credit; you cannot name a single job it can actually replace.

Scope: as of today

26:07 20VC: Why OpenAI and Anthropic Won't Win the App Layer | Why Teams Will Get Bigger Not Smaller in a World of AI | Why AI Removes Incumbents Advantage of Bundling | China vs America: Who Wins the AI War with Arvind Jain, Co-Founder @ Glean

Andrew Ng · Nov 17, 2025

Only a small subset of jobs is genuinely in trouble from AI; for the vast majority of knowledge workers AI is a net benefit

AI still can't do everything, so even where it covers 30-50% of a role, 50-70% of the work still requires a human

Scope: a small subset of jobs is in trouble

15:04 20VC: Andrew NG on The Biggest Bottlenecks in AI | How LLMs Can Be Used as a Geopolitical Weapon | Do Margins Matter in a World of AI? | Is Defensibility Dead in a World of AI? | Will AI Deliver Masa Son's Predictions of 5% GDP Growth?

Anish Acharya · Feb 9, 2026

AI productivity gains are so far showing up as a four-day work week rather than 20% fewer jobs, because jobs are bundles of tasks that don't lend themselves to 100% automation

Getting to 60–80% of a job is fairly easy but the last stretch is extremely hard — you can automate all the customer support you want, but sometimes you have to take the customer out for a steak dinner

Scope: 'so far'

35:25 20VC: Is SaaS Dead in a World of AI | Do Margins Matter Anymore | Is Triple, Triple, Double, Double Dead Today? | Who Wins the Dev Market: Cursor or Claude Code | Why We Are Not in an AI Bubble with Anish Acharya @ a16z

Brendan Foody · Jun 1, 2026

The most important question amid the layoffs is empirically measuring which specific tasks within job categories AI can and cannot do.

Job categories like consulting, banking, law and software engineering are bundles of tasks, and only task-level measurement answers what is actually automatable.

13:53 20VC: Mercor CEO on Why Application Layer Companies Have No Defensibility, The Model is the Product | Token Spend Will Exceed Headcount Spend in 5 Years | The True Cost of Hiring AI Researchers in the Valley Today with Brendan Foody

Arvind Jain · Jul 11, 2026

AI can handle many tasks within any given role but cannot replace the final intangible part of the role.

26:31 20VC: Why OpenAI and Anthropic Won't Win the App Layer | Why Teams Will Get Bigger Not Smaller in a World of AI | Why AI Removes Incumbents Advantage of Bundling | China vs America: Who Wins the AI War with Arvind Jain, Co-Founder @ Glean

Mati Staniszewski · Sep 8, 2025

AI agents will absorb manual tasks that require no domain expertise, and this makes the remaining specialized humans more valuable rather than redundant

Tasks like appointment scheduling or refunds are easily automated with safeguards, while high-stakes last-mile work such as helping a patient navigate post-hospital care requires deep domain expertise AI cannot supply

Scope: assumes safeguards and authentication are in place; automation percentage will keep shifting upward over time

60:12 20VC: ElevenLabs Hits $200M ARR: The Untold Story of Europe's Fastest Growing AI Startup | The Real Cost of AI from Talent to Data Centres | How US VCs are in a Different League to Europeans | The Future of Foundation Models with Mati Staniszewski

Winston Weinberg · Jan 19, 2026

The corporate work Harvey absorbs is mostly lower-end work previously done by alternative legal service providers, not the work its law firm customers do

39:45 20VC: How Model Performance is Plateauing | Two Key Rules for Effective Deal-Making | Company Building Lessons from Keith Rabois, Brian Halligan and Pat Grady | Why Enterprise AI Adoption is Years Off with Harvey CEO Winston Weinberg

Shiv Rao · May 16, 2026

High-frequency, low-stakes care — like medication refills — should and will be progressively automated, converging on the most boring variant of primary care that nobody wants to practice; the hard cases requiring taste and judgment will not be.

Low-stakes repetitive workflows are the natural automation frontier, while judgment-heavy care gets very real; and a tsunami of healthcare need is coming that the system isn't prepared to deliver.

Scope: hard cases excluded; cites current Utah controversy over automated prescription refills

38:58 20VC: Lessons from Jensen Huang on "Founder Mode" | How to Know if OpenAI or Anthropic Will Kill your Company | How USV Liking Music Made Them $1BN on an Investment | The Five Year Desert to Product Market Fit & a $5.3BN Valuation with Shiv Rao @ Abridge

Harry Stebbings · Nov 3, 2025

A good barometer of enterprise utility for AI is whether it can replace the work of the bottom 5% of employees in any category, and we generally overestimate AI's capability

15:48 20VC: Cohere's Chief AI Officer on Why Scaling Laws Will Continue | Whether You Can Buy Success in AI with Talent Acquisitions | The Future of Synthetic Data & What It Means for Models | Why AI Coding is Akin to Image Generation in 2015 with Joelle Pineau

Harry Stebbings · Mar 30, 2026

Most social media managers are bad at their jobs, so a company like AT&T would rationally replace an average social media manager with an AI-driven tool

If an AI tool can match the output of a mediocre incumbent, the substitution is obviously worth making

Scope: applies to mediocre/average performers, not top ones

56:54 20VC: Marc Andreessen on The Future of Venture Capital: Will a16z Go Public | Why Labour Displacement with AI is Wrong | Why Introspection is Dangerous | Why "Diamonds in the Rough" is BS in VC | Why a16z Invested $300M into Adam Neumann

Jonathan Siddharth · Dec 1, 2025

All knowledge work will eventually be automated; it is only a matter of time.

0:00 20VC: Scale, Surge, Turing, Mercor: Who Wins & Who Loses in Data Labelling | Is Revenue in Data Labelling Real or GMV? | Why 99% of Knowledge Work Will Go and What Happens Then? | Why SaaS is Dead in a World of AI with Jonathan Siddharth @ Turing

Jonathan Siddharth · Dec 1, 2025

All knowledge work will eventually be automated — any job that involves looking at a screen, using tools, and working a keyboard and mouse

From generating data for all the frontier labs Turing sees a glimpse of the future, and computer-use agents will keep improving over the next decade; it's only a matter of time

Scope: framed as a matter of time rather than an immediate event

17:33 20VC: Scale, Surge, Turing, Mercor: Who Wins & Who Loses in Data Labelling | Is Revenue in Data Labelling Real or GMV? | Why 99% of Knowledge Work Will Go and What Happens Then? | Why SaaS is Dead in a World of AI with Jonathan Siddharth @ Turing

Harry Stebbings · Aug 25, 2023

Media companies and gaming houses are on a path to cutting roughly 90% of their asset, content and graphic creation teams and replacing them with AI tools.

He speaks to CEOs of portfolio companies doing exactly this, and the teams know the transition is underway.

Scope: gradual transition over time; based on his own portfolio companies

10:46 20Product: Enterprises are not Adopting AI Yet, When Will AI Break Into Enterprise, What are the Blockers, What Do Enterprises Need from AI & Why Services Companies Will Win in the Next 10 Years of AI Implementation with Howie Liu, Founder & CEO @ Airtabl

Harry Stebbings · Jul 11, 2026

Once AI does 90% of a role, that role gets cannibalized even if the last 10% still needs a human occasionally.

The residual human tasks, like buying a personal gift once a year, are too infrequent to sustain the role.

26:40 20VC: Why OpenAI and Anthropic Won't Win the App Layer | Why Teams Will Get Bigger Not Smaller in a World of AI | Why AI Removes Incumbents Advantage of Bundling | China vs America: Who Wins the AI War with Arvind Jain, Co-Founder @ Glean

Joelle Pineau · Nov 3, 2025

Efficiency gains from AI land hardest where tasks can be precisely specified; ambiguity in task specification is what current machines struggle with

Where you can be very precise about what a great result looks like, the task can be automated far more easily than nuanced, complex tasks

18:45 Task specifiability not role decides what automates

Harry Stebbings · Aug 18, 2025 · hedged

AI won't displace brain surgeons for years, because patients won't choose the robot version even though surgical robots are already good

Patient willingness, not robot capability, is the binding constraint

62:37 Patient trust not capability blocks surgical automation

Harry Stebbings · May 16, 2026

Fears that AI will make doctors redundant are ridiculous.

Anyone who has seen an emergency room on a Friday night knows the demand and pain vastly exceed supply.

40:23 Demand far exceeds supply so no redundancy

Jonathan Siddharth · Dec 1, 2025

Some roles can skip the partial-autonomy intermediate step and go straight to full autonomy, while others cannot

Roles well represented in internet pre-training tokens (e.g. customer support) give models enough intelligence out of the box; roles like AI researcher or a VC-financing lawyer are underrepresented, and firm-specific practices differ, so they require fine-tuning on proprietary data

43:06 Pretraining representation of the role decides the automation path

KR Sridhar · Jun 29, 2026

Once intelligence is abundant, wisdom becomes the scarce and valuable asset, because AI cannot supply wisdom, happiness, or human empathy

A bot giving counseling is not the same as two humans reading each other's faces and relating; humans are social animals

15:54 Wisdom and human empathy cannot be supplied by a bot

Harry Stebbings · Jul 11, 2026

For most people, AI can already do the majority of an executive assistant's job.

26:23 Assistant type roles are already mostly automatable

Arvind Jain · Jul 11, 2026

Firms will not settle for a 90% AI-only solution, because competitors have the same AI tools plus a human on top and will out-compete them.

You have to be competitive with others who have identical AI access; the human is the differentiator.

26:58 Equal ai access makes the human on top the differentiator

Cliff Weitzman · May 9, 2026

We can solve essentially everything — anyone can now educate themselves to a doctor's level of sophistication or learn to fix anything with AI tools, and kids will grow up with that agency ingrained

His experience with his father's cancer and his brother's illness plus the ease of learning anything on demand support a Promethean view

101:50 Ai lets laypeople self educate to expert level

Max Junestrand · Jan 26, 2026 · hedged

Within 12 to 24 months AI will be able to complete end-to-end tasks deterministically in the legal vertical, and in aggregate that probably does mean labor displacement

Unless firms find other work or grow the pie, the work being automated has nowhere to go; individual firms can win share from others, but at the total level displacement follows

52:49 End to end task automation implies aggregate displacement unless the pie grows

Harry Stebbings · Aug 8, 2026

AI in professional services is primarily a TAM expansion play rather than a displacement play

Many things you previously couldn't afford a lawyer for you will now use and benefit from

63:45 Affordability unlocks new demand so the market expands

Winston Weinberg · Jan 19, 2026

AI will not cannibalize junior lawyers; law firms will get more work and the professional services market will keep growing roughly in line with GDP

Clients want to stop paying for commodity tasks like marking up NDAs but are buying new categories of work (AI risk, cross-border regulatory advice); professional services is cyclical, and if AI makes the economy and clients' product output explode, they will need more legal advice, not less

40:32 New work categories and gdp growth offset the automated commodity tasks

Avishai Abrahami · Jul 13, 2026

LLMs will not replace white collar jobs

They aren't that smart, they keep making silly mistakes, and they're weak at reasoning — good mainly at collecting data and reframing it

37:56 Weak reasoning means no white collar replacement

Harry Stebbings · Apr 14, 2025

Someone entering university today should become a brain surgeon or dentist to be safe from AI disruption for at least twenty years

Those professions will remain safe for about twenty years minimum

44:38 Specific high touch professions like surgery dentistry stay safe for decades

Alex Lebrun · Jun 19, 2023

AI will not replace doctors, but doctors who use AI will replace doctors who don't

AI can hugely change how doctors work, so those who embrace the change will thrive

26:48 Ai wont replace doctors but ai using doctors will replace non adopters