Selling the work: Sequoia's services thesis, applied to your business

Sequoia published an essay arguing that the next great software company will look like a services firm. It will not sell you the accounting tool; it will close the books. I sell judgement while machines eat execution, so I read it twice. Once nodding, once looking for what it costs you. Both readings are below, with one takeaway for whichever business you run.

What the essay argues

The setup is the anxiety every AI founder carries: whatever tool you build, the next model release may ship it as a feature. Bek’s answer is to stop selling the tool and sell the work, because if you sell the work, every model improvement makes your service faster and cheaper instead of making you redundant. His example gives the size of the prize. A company pays five figures a year for accounting software and six figures for the accountant. The software budget was never the market. The labour budget is.

Two distinctions carry the essay. The first splits all work into intelligence and judgement. Intelligence work is complex but rule-governed: translating a spec into code, filling a claim against a damage schedule. Judgement is knowing what to build next, when to ship, which risk to eat. AI has crossed the line where it does most of the intelligence work autonomously. Software engineering crossed first, and Bek’s claim is that every profession follows, in order of how rule-governed its work is.

The second is the business-model consequence. “A copilot sells the tool. An autopilot sells the work.” Copilots make a professional faster and leave them responsible. Autopilots sell the finished outcome to the person who needed it, and cut the professional out of the transaction. His go-to-market observation is the best operator thinking in the piece: start where the work is already outsourced. There, the buyer has already accepted external delivery, already has a budget line, and already buys outcomes. Replacing a vendor is a purchase order. Replacing an employee is a reorg.

Where I push back

Three places, all of them load-bearing.

The judgement line moves, and the essay says so quietly. The comfort in the intelligence/judgement split is that your judgement is safe. But the convergence section concedes that today’s judgement becomes tomorrow’s intelligence, as systems accumulate data about what good decisions look like. So the split is not a wall. It is a waterline, and it is rising. The honest version of the thesis is not that machines do intelligence and humans keep judgement. It is that humans keep whatever judgement has not yet been recorded often enough to become rules. Plan on the waterline, not the wall.

Accountability is doing more work than intelligence. What a company actually buys from an accountant is not book-closing. It is someone whose name is on the filing when the tax office calls. The essay treats trust as a product-quality problem, but it is a liability problem: insurance, recourse, a throat to choke. That is why licensed professions will bend rather than break. The autopilots that win will answer “who pays when you are wrong?” on their pricing page, not in their FAQ. Cialdini’s authority chapter is about exactly this: the credential is a shortcut for accountability, and machines do not yet have one.

The savings get competed away. An investor reads “the work budget dwarfs the tool budget” and sees market size. An operator should see a cost line about to collapse for them and for every competitor at once. When everyone’s books close for a tenth of the price, nobody wins on cheaper books. The durable advantage moves to whatever the machine is not doing, which is, for now, the judgement layer. Execution getting cheap makes deciding-what-to-do the expensive part. That is the entire premise of what I sell, so discount my agreement accordingly.

What it means for you, by the business you run

One takeaway each. They compound if several describe you.

1. You sell software. Your buyer never wanted your tool. They wanted the work it shortens. List your product’s top three jobs and ask, for each, what it would mean to sell the completed job at a price anchored to the labour it replaces. Then notice that your best user today may be the person an autopilot cuts out of the deal tomorrow. That conflict is the essay’s closing point, and pretending you can serve both sides indefinitely is how incumbents lose to pure-plays.
2. You run an agency or a services firm. You are holding two assets: an execution layer about to be repriced, and years of recorded judgement about what good looks like in your niche. Productise the execution layer yourself before a startup does it to you. And start charging for judgement as its own line, strategy, review, sign-off, so that when clients unbundle the execution, the part they keep buying has your margin in it.
3. You are a licensed professional: accountant, lawyer, broker. The shortage in your profession is a window, not a moat. The licence protects the signature, not the work behind it. Move your practice up to the accountability layer: own the client relationship, the judgement calls and the liability, and become the buyer of the machine layer rather than its competitor. The practices that fail will be the ones still billing hours for what the machine does in minutes.
4. You run a SaaS with a services line, or a marketplace of professionals. You are the copilot facing the innovator’s dilemma by name. The essay is right that your product knowledge and distribution are an advantage, and right that you will hesitate to cut your customers out. Decide now which side of the transaction you serve, the professional or the outcome buyer, because straddling prices you into the middle of a collapsing spread.
5. You run an ecommerce shop, a local business, anything offline. You are not the disruptee here. You are the buyer. Your bookkeeping, your IT contract, your insurance brokerage and your long-tail supplier negotiations are all on the essay’s target list. Re-tender the outsourced intelligence work over the next two years, insist on outcome pricing with liability attached, and redeploy what you save into the work nobody sells you: positioning, pricing, deciding what to build. The savings are temporary. What you do with them is not.
6. You are building an AI product right now. The wedge logic is the most useful page for you: pick the task that is already outsourced, already budgeted, already bought as an outcome, and land as a vendor swap instead of a reorg. But run the canvas on it first. Whose budget line, which segment, what the revenue stream anchors to. The essay’s map is drawn by market size, and you will live or die by segment.
7. You advise, consult, or coach. The essay ends by admitting consulting is mostly judgement and leaving the automation candidates open. Do not relax. The disaggregation it sketches, machines taking the research and benchmarking while humans keep the recommendation, describes your deliverable exactly. If half your fee is for gathering and half for concluding, the gathering half is already repriced. Make sure your invoice knows which half is which.

The pricing angle nobody will enjoy

The essay’s quiet superpower is a pricing observation: autopilots price against the services line, not the software line, and the services line is six times bigger. That is a once-a-generation pricing umbrella. Charge a third of the human cost, keep software margins. It will not last, because umbrellas invite rain. As autopilots multiply in a category, the anchor decays from “fraction of a salary” to “compare three quotes”, and sellers who priced on the labour anchor without building a moat under it will reprice all the way down to cost.

If you are buying, the same clock runs in your favour. Early autopilot pricing is anchored to what you used to pay a human, which means the first quote you get is the worst one you will ever get. Negotiate like the anchor is fake, because it is. This is the habit trap seen from the buyer’s side, and the contrast principle seen from the seller’s. It is also the argument for repricing before your market does it for you.

Should you read it

Yes. It is free, short, and written to be quoted. Read it as a map of where cost lines collapse next, not as a promise about whose job is safe. And remember who is talking: a venture firm describing the companies it wants to fund. The analysis is genuinely good. The frame, in which the interesting position is always the disruptor’s, is theirs rather than yours. Most readers of this page are not building an autopilot. They are about to buy from one, compete with one, or sell judgement above one. Those three positions get no airtime in the essay. They got this page instead.

Common questions