Companies Need an Owner of the Problem

Earlier this year, Sequoia partner Julien Bek published Services: The New Software and made a claim that is easy to skim past and hard to forget: the next $1T company will be a software company masquerading as a services firm.

The argument is a single comparison. A company might spend $10K a year on QuickBooks and $120K a year on an accountant to close the books. QuickBooks is the tool. Closing the books is the work. Sell the tool and you race the model: the next release can make your product a feature. Sell the work and every model upgrade makes you faster, cheaper and harder to compete with.

That one move split the commentary into two camps. One says founders are building the wrong thing: tools, instead of companies that do the job. The other defends the tool layer and points at copilots compounding. Both camps argue about what to sell.

Neither asks who owns the problem.

A tool has no owner of the outcome. A service company is an owner, but only on rent. Neither leaves you owning the problem.

Buyers do not buy tools, and they do not really buy work. They buy relief from a problem. And relief only arrives when someone owns the problem end to end: scope, execution, quality, deadline. Ownership means a person who is accountable, or a company accountable by contract. That unit of ownership is the missing variable in the whole $1T debate.

Buy the tool and nothing changes. This is the pattern every AI adoption story eventually hits: the licence is bought, the rollout is announced, and the work still does not get done better. Not because the software is bad. Because no one owns the outcome it was supposed to produce.

A tool does not attach responsibility to anyone. The professional still owns the outcome, so a copilot just makes a busy person slightly faster. If the expert is overbooked, the tool gets used at the edges of the day, or not at all. The productivity gain goes to the individual. The company's problems stay the company's problems.

Enterprise software has always had this failure mode, and AI made it cheaper to repeat: it is now trivial to licence a dozen agents and have every one of them be someone else's job.

This is not a technology failure. It is an ownership failure. Everyone's problem is no one's problem.

Adoption without ownership is a status update.

The fix was never a better tool. It was an owner: a single accountable person who uses whatever it takes to move the outcome. Companies already know this; it is why org charts exist. The constraint was never insight. It was cost. An owner needed headcount underneath, so ownership was rationed.

This is the deeper reason autopilots win, and it is more structural than the tool versus work framing. Crosby is not an NDA product. It is a service company that owns the NDA, drafted, reviewed and delivered. WithCoverage owns the policy, not the broker seat. When you buy the outcome, you outsource the ownership along with it.

It is the right move and the right wedge: the outsourced budget line already buys an owner, so the substitution is a vendor swap, not a reorg. That is why the opportunity map is ordered by outsourced share and intelligence ratio.

But watch what happens to the buyer. The owner now sits outside the company. The margin, the data and the compounding accrue on the vendor's side. You become a client: better served than before, and no closer to owning the capability. Renting an owner is strictly better than owning no one. It is still rent.

Autopilots win because they are the owner. They keep the compounding because the owner is theirs.

The question nobody in the commentary asks: if ownership is the unit that delivers, and an owner is not something you can deploy, why is the only choice between no owner and a rented one?

Ownership was rationed because it cost headcount. An owner was $120K for an accountant, similar for an analyst, a whole bench for some problems. Companies could afford owners only for their most expensive problems. Everything below the line went unowned.

Look at Sequoia's opportunity map with that lens and it stops being a list of startups and becomes a list of rationed ownership:

• Insurance brokerage ($140-200B). Shopping carriers and filing forms, standardised work across thousands of small brokers, with no one owning the long tail of renewals.

• Accounting and audit ($50-80B outsourced in the US alone). 340,000 accountants left the profession in five years while demand grew; 75% of CPAs near retirement. The close is a problem waiting for an owner.

• Healthcare revenue cycle ($50-80B). Clinical notes to medical codes: complex rules, but rules. Mature, outcome-based outsourcing with no in-house owner.

• Claims adjusting ($50-80B). Policy language against damage schedules, a workforce aging out, and every claim triaged by whoever has time.

• Tax advisory ($30-35B). 80-90% intelligence. Multi-jurisdiction complexity is exactly why SMBs rent the owner.

• Legal, transactional ($20-25B). Contract drafting and regulatory filings, verifiable, high-intelligence, routinely outsourced. The NDA has a rented owner; the review queue has none.

• IT managed services ($100B+). Patching, monitoring, provisioning, alert triage across identical environments, with nobody owning uptime as an outcome.

• Supply chain and procurement ($200B+). Contract leakage runs 2-5% of spend: work no human was ever economical enough to own.

• Recruitment and staffing ($200B+). Screening and matching are pure intelligence; the funnel is owned by whichever recruiter is least busy.

• Management consulting ($300-400B). Intelligence components disaggregate from judgement components, and the data-gathering part is already awaiting an owner.

Two of the essay's distinctions explain what actually flipped. Intelligence, translating a spec into code or a chart into codes, is rules, and rules run at AI cost. Judgement, what to build, when to ship, what risk to accept, stays human. An owner used to command a handful of people at headcount prices. Now the same owner commands an AI workforce at model cost.

What did not flip is the owner itself. Accountability is not deployable. An owner is a human, or a company accountable by contract; an agent is neither. What changed is the span of an owner: the intelligence under a single accountable person now scales without the org chart.

The bottleneck flips from budget to assignment. The question is no longer "can we afford an owner here?" It is "have we named one, and have we given it a workforce?"

That is the whole opening. The startup's move is to rent you an owner. Yours is to keep the owner and deploy the workforce.

Ownership is an organizational act, not a purchase. The playbook has four moves.

1. Inventory the problems. List what consumes the most hours and money: the reports, the reviews, the renewals, the tickets, the claims, the closes. Name each in one line. These are your unowned or under-owned problems. Almost every company finds its top five instantly, and they are rarely the ones with the biggest software budget.

2. Name the owner. An owner is never a licence and never an agent. It is a named human accountable for the outcome, or, when you choose to rent, a service company accountable by contract. Assign the accountability, then give the role a job description: scope, access, deliverable, review cadence.

3. Separate intelligence from judgement. The rules-based share goes to the workforce: drafting, coding, matching, triaging, reconciling, filing. The judgement stays with the owner: what to accept, what to escalate, what the risk is worth. Write both into the owner's definition.

4. Hold the owner to the outcome. A deliverable and a date per cycle. Fix the accountability loop and the technology takes care of itself.

The order of attack follows the same logic as the map. Start where the owner slot already exists: outsourced work, NDAs, bookkeeping, claims triage, the procurement long tail, screening, already has a budget line that buys a rented owner. Move that outcome in-house: name your own owner and give it an AI workforce, and the vendor swap reverses in your favour. Insourced headcount comes next, as models absorb more intelligence and judgement transfers to supervision.

The map is not a list of companies. It is a list of problems, each waiting for an owner with a workforce.

The next $1T company will be a services firm that rents out owners. That half of the bet is probably right. The second half, that it has to be them rather than you, has no reason to hold. The same economics that let a service company own your NDA let your own owner absorb it.

An AI employee is not an owner. An owner is a person who is accountable, and accountability cannot be deployed. What an AI employee is, precisely, is the workforce underneath the owner: it executes the intelligence, while the owner keeps the judgement and the accountability. Keep the owner in-house, put an AI workforce beneath it, and the margin, the data and the compounding stay on your side of the contract.

So the practical test for the next year is simple. Take the five most expensive problems in your company and ask one question: who owns this?

If the answer is no one, you have an ownership problem. If the answer is a vendor, you have a rent problem. If the answer is a named human owner with an AI workforce underneath, you own the outcome and everything that compounds with it.

Stop buying tools. Stop renting owners. Power the ones you have.