TLDR: If your job is a horizontal task (a step in someone else's chain), you are, to use the technical term, fucked. If your job is a vertical (you own a whole process from start to finish), you are about to have the best decade of your career. Same goes for companies. The ones who win the next 5 years are the ones who stop thinking about work like a factory, and most of them won't, because they can't, because their org chart is the thing being eaten.
I keep having basically the same conversation with three different types of people: friends who are panicking that AI is going to take their jobs; execs at big companies who are panicking that their AI strategy isn't working; and founders who are quietly running rings around both groups. All three groups are dealing with the same issue but most of them are not thinking about it correctly....
First, what do I actually mean by vertical and horizontal
Pick any piece of useful work, getting a customer onboarded, running a research study, closing a deal, building a product, writing a report, whatever it is. It's a chain. You start with some inputs (a question, a brief, a lead, a problem), you go through a bunch of steps, and at the end of it something of value comes out the other side (a decision, a customer, a product, a contract). Cost gets added at every step, and hopefully (sometimes) so does value.
That whole chain, from raw input at one end to finished thing at the other, is the vertical.
A horizontal job is when your work is one of the steps (or more likely a few steps as most jobs are a bit fuzzy, but you get the idea). You build the chart, you format the deck, you qualify the lead, you write the SQL, you triage the ticket, etc. The key thing is that what you produce isn't the finished output, it's an input that gets handed to the next person, who hands it to the next person, who hands it to the next person, and somewhere 12 handoffs later something of actual value comes out the other end (you hope).
A vertical job is when you own the whole chain (or at least a large enough chunk of the chain that its inputs and outputs are distinct, and the utility of the "thing" created is significantly modified). Question goes in one end, value comes out the other, and you (with or without help) carried it the whole way. The researcher who takes a business question and hands back a recommendation, vertical. The founder who takes a customer pain and ships a product, vertical. The salesperson who finds the lead, qualifies it and closes it, vertical. The doctor who diagnoses, treats and discharges a patient, vertical.
Simplest test, does the value chain start and end inside your work, or are you a step in someone else's? If you're a step or a too small group of steps, you're horizontal, and life could get tricky. This though is also causing trouble at large companies:
Why companies are getting bugger all back from AI (for now)
So big companies are spending fortunes on AI and getting basically nothing back, and the reason is pretty simple, value is created vertically, but companies are organised horizontally, and the two don't really meet.
Pick any process inside a big company, getting a customer live, running a market study, closing the books, shipping a feature, whatever, and it's going to go through 20 to 50 actual steps, owned by different teams, with handoffs and approvals and emails and tickets and Jiras and Slacks and "let me loop in Dave" in between. And most of the time, and most of the cost, lives in the seams between the steps, not the steps themselves. The step is almost never the slow bit. The queue before the step is the slow bit.
So when these companies "do AI", they automate one step (or a small group). They drop a copilot on the SQL writers, they put GPT on top of the customer service team, they throw an LLM at the contract redlines, and everyone pats themselves on the back. Step 14 of 30 now takes 4 minutes instead of 4 hours, brilliant. And then it sits in step 15's queue for 2 days like it always did, because step 15 is owned by a different team in a different building (probably a different country) who haven't even seen the email yet. End-to-end throughput barely moves and you've just spent a million quid on a tool that makes one person look fast on their LinkedIn.
This could be called the factory model trap (Claude suggested this phrase for me) and I think it's the biggest problem in enterprise AI right now. We're still thinking of work like an assembly line where the move is to optimise each station, when really the assembly line itself is the problem. The seams between the stations are where the time goes, and the seams are exactly the bit AI is best at eating, if you let it.
Startups are getting huge value out of AI for one reason and one reason only, one founder or a tight team owns the whole vertical, so they let AI eat the seams, not just the steps. Big companies struggle to do this because the seams are the org chart, the seams are people's job descriptions, the seams are who reports to who and whose budget pays for what. To actually let AI eat them you'd have to redraw the org, kill departments, re-bundle roles, give individuals scope they don't currently have, and have a few very honest conversations with the executive layer about why some of them, frankly, don't need to exist anymore. Most big companies would rather buy another copilot license than have any of those conversations, which I get (no one wants to fire their mates), but it's why their AI ROI is basically zero. Also the companies that are firing people due to AI are probably firing the wrong people (but that's a different blog post).
What horizontal vs vertical actually looks like
Example 1: the junior analyst vs the researcher
The junior analyst at a big consultancy, builds the chart, formats the slide, runs the regression the senior associate asked for, sends it up the chain. Horizontal. AI can do most of this already and will do all of it inside 12-24 months (it's already happening at the firms that have got their act together). The bit that wasn't horizontal, the judgement of what the chart should even be, was always sitting with the senior, not the analyst. So when AI eats the analyst, the senior doesn't really notice, except the slides arrive faster and there's no one to blame for the typos.
The end-to-end researcher (a one-person research team in a startup, a freelancer, or what we let people become at Brox), takes a business question from a stakeholder, designs the study, runs it, interprets it, hands back a recommendation. Vertical. AI doesn't replace this person, AI lets them do 10 studies a month instead of 1, because the boring middle bit (transcribing, coding, summarising, drafting) collapses to basically nothing. The judgement at the start and end is still theirs, and demand for that judgement doesn't go down, it goes up, because cheap good research means a lot more questions get asked.
Example 2: the automated step that still waits for everyone else
This is the dumbest version of the trap and I see it everywhere (there are whole startups dedicated to helping enterprises with it). A team automates a few steps beautifully, the lead enrichment, say, or the contract drafting, or the QBR generation.
The step that used to take 3 days now takes 3 minutes, amazing, Slack channel of high fives, internal newsletter, the lot. Except the step before it still takes 2 weeks (because legal hasn't looked at it), and the step after it still takes 2 weeks (because finance hasn't approved it), so the actual end-to-end process is still 4 weeks long, just with a really efficient step 14 in the middle. The ROI on the AI investment is basically zero but the deck announcing it gets passed around the exec team like it's the second coming.
The vertical version is the same business doing the same process, but one person owns the lot from lead-in to contract-out, with AI doing the heavy lifting across all 30 steps at the same time.
End-to-end goes from 4 weeks to 2 days. THAT is the actual transformation, and almost no big company is set up to even attempt it, because attempting it would mean firing the person who currently owns step 12, and the person who owns step 12 is probably the one writing the AI strategy. You'd also have to give individuals a lot more autonomy and responsibility but the great thing about AI is that this is easier than ever to do (if you trust your people).
Example 3: the dev shop vs the founder
A dev shop that takes specs and writes code is mostly horizontal. The world will keep needing humans to build software for a while, but the easy "write me this CRUD app" tier is already getting hoovered up by Claude Code and Cursor and the lot of them (I've launched two of my own websites with Claude Code in the last month, I am not making this up).
The founder of that same dev shop, who finds the client, scopes the problem, decides the architecture, ships the thing, and owns the relationship, vertical. AI lets that founder do the work of 5 people, which doesn't shrink the business, it grows it, because the marginal cost of taking on another client drops to basically nothing. Same skills, a different posture, and a completely different fate.
The horizontal worker is competing with the model, the vertical worker is using it as a weapon. (AI came up with that)
Why going vertical creates more work, not less (the Jevons bit)
This is the bit that should make you a lot more optimistic if you can get yourself on the right side of this.
Jevons paradox is the economics thing where when something gets cheaper, people consume more of it, not less. When steam engines got more efficient we burned more coal, not less. When LED bulbs got cheap to run we lit up places no one had bothered to light before. When cloud compute got cheap we used a comically larger amount of it. Same thing is about to happen to verticals.
If you own an end-to-end research process and AI makes it 10x cheaper to run a study, you don't run the same number of studies and pocket the savings, you run 10x more studies, because there were always 10x more questions worth asking that no one could justify paying for at the old price. If you own an end-to-end sales process and AI makes it 10x cheaper to qualify and close a lead, you don't talk to the same number of leads, you talk to 10x more. The demand was always sitting there, the cost was just the blocker.
When you go vertical, you create demand on both sides of yourself, more demand for your inputs (data, leads, sources, briefs) and more demand for your outputs (decisions, customers, products, deals). Your bottleneck stops being your own throughput and starts being how much pull you can generate on the system around you, which is a much better bottleneck to have because it's the one thing humans still have a massive edge on (we are extraordinarily good at wanting things and asking for things, possibly our top skill).
The horizontal worker thinks "if my task can be automated I'm out of a job" and they're, sadly, correct (if they stay horizontal). The vertical worker thinks "if my whole process gets 10x faster I can take on 10x the work" and they're also, happily, correct.
And there's a more subtle thing going on, which I think is actually the biggest unlock and the one no one is really talking about, which is that the biggest 10x in productivity isn't doing tasks faster, it's no longer waiting for other people (hell is other people). The hidden tax on all knowledge work is waiting, waiting for the analyst, waiting for legal, waiting for design, waiting for the data team to pull the numbers, waiting for IT to approve the tool, waiting for the deck to come back from formatting, waiting waiting waiting waiting. When you own the vertical, you stop being a downstream victim of upstream slowness, you just do the upstream bit yourself with AI and keep moving. That alone is worth way more than any individual task speedup, and it's the bit the productivity gurus never talk about (because productivity gurus mostly spout nonsense and have never worked a proper job).
This is going to be brutal on big companies
A bit depressing if you work at a big company, very exciting if you compete with one, but the org charts of most large companies are basically machines for manufacturing horizontal jobs.
Departments, functions, specialisms, layers, the whole bloody shebang is built around slicing work into horizontal stations and giving each person one to stand at for 40 years. To go vertical inside a big company you'd have to flatten the org, re-bundle roles, kill committees, give individuals real scope, fire (or move) a lot of middle managers, and let AI eat the seams that the political economy of the company is literally built on. Good luck (I mean that, some of them will figure it out and they'll do unbelievably well). Most won't. The ones that do will look unrecognisable in 5 years and will quietly hoover up the ones that don't.
This is also why so many big company "AI strategies" are basically polite ways of automating individual roles instead of redesigning processes. It's the easier political move, even though it's the worse economic one, and the consultants are very happy to sell it because a multi-year transformation programme is a way fatter engagement than "redesign your org from the ground up, sack half your VPs, give the survivors a proper job". Funnily enough no one's pitching that one. (Could you imagine the keynote, "Hi, I'm McKinsey, please pay us £40m to recommend that you fire most of us." It's never going to happen.)
What to actually do
If you're a founder or an exec, stop asking "which jobs can we automate?" and start asking "which end-to-end processes can one person now own?" Then redesign around those people and let AI fill in the seams. That's where the actual 10x lives. Automating individual steps will get you a 1.1x, a fat consulting bill, and a slightly demotivated workforce.
If you're an employee or a freelancer, find or build your vertical, even a tiny one. The smallest viable vertical beats the biggest horizontal speciality, every single time. If your job is currently horizontal, the move is to expand upstream and downstream until you own a slice end-to-end, even if the slice is small to start with. A horizontal SQL writer can become a vertical "I take business questions and return decisions" person pretty quickly, mostly the same skills, completely different posture, completely different career trajectory, completely different bank balance.
If you're a student or early career, please for the love of god do not optimise for being the best at one task in someone else's chain. That's the worst trade in the modern economy and they're not going to tell you that at the careers fair. Optimise for owning outcomes, even tiny ones. Build a small thing end-to-end, sell a small thing end-to-end, write something end-to-end, ship something end-to-end. The leverage on owning outcomes is about to be insane, and the people who get a few reps in early are going to have an absurd advantage by the time everyone else catches on. (Also, learn to use Claude Code, or Cursor, or whatever the current best is when you read this, I genuinely cannot stress this enough.)
The Brox bit (the shilling section)
I'd be a hypocrite not to mention that this is one of the bets we're making at Brox.AI. The whole point of what we build is that one researcher, one strategist, one product person can own a research vertical end-to-end, from question to insight to decision, instead of being one step in a 6-week, 8-person, 4-vendor, 12-meeting chain that the old market research industry was built on (and is, slowly, dying on). The work that used to take a department now takes a person plus our twins, and the people who use us aren't doing less research, they're doing 10x more, because suddenly they can, and because the demand for "actually understanding what humans want" has always been bottomless, the cost was just the blocker.
That's the whole game, in our space and in pretty much every other knowledge industry I can think of. The horizontal version of the work gets compressed into the model. The vertical version gets supercharged. The people who clock this early are the ones doing the supercharging, the people who don't are, well, the ones being supercharged at.
Get vertical, because if you stay horizontal you'll be screwed (which outside of a very narrow range of professions I wouldn't recommend while at work).
Onwards and upwards,
Cheers,
Hamish