Service as a software

TL;DR

SaaS sold you tools. Service as a software sells you the finished work, priced by the outcome. The shift is real, the trillions are real, and for small firms it is a fine trade. But every outcome bought from a vendor's engine teaches that engine, not your organisation. Labour arbitrage is being replaced by learning arbitrage. The firms that get this right will buy outcomes for the work that does not matter and build the engine for the work that does.

For twenty years the software industry performed one great trick. It turned products into subscriptions. You stopped buying the disc and started renting the seat, and we called it software as a service, and it built the modern economy of tools. Now the same industry is running the trick in reverse. It is turning services into products. The people who name things are calling it service as a software. The name is clumsy. The idea is not.

Here is the idea with the deck stripped away. Software used to help your people do the work. Now software does the work and hands you the result. A resolved ticket. A closed set of books. A filed return. A shipped feature. You are not buying the tool anymore. You are buying the finished Tuesday.

The trillions arrive

The venture firms have done the arithmetic, because sizing things is their love language. Foundation Capital calls it a 4.6 trillion dollar opportunity, software that stops organising the work and becomes the worker. Sequoia goes further and predicts the next trillion dollar company will be a services firm powered by AI. Their reasoning fits in one line.

For every dollar spent on software, six are spent on services.

Sequoia Capital, Services: The New Software ↗

That line explains the stampede. Software budgets are counted in hundreds of billions. Work budgets are counted in trillions. So every vendor is climbing out of the tool budget and into the labour budget. Zendesk now charges per resolved ticket. Salesforce charges per agent conversation. Sierra wrote a manifesto for pricing by the outcome. And in June, Deloitte published accounting guidance for outcome-priced AI agents. When the accountants start writing rules for a business model, the argument about whether it is real is over.

Much of this is plainly good. A five person company could never afford an agency, a consultancy, or an adviser on retainer. Now it can buy the outcome off a shelf for the price of a phone plan. Cheap service means more service gets bought, the way cheap compute meant more computing. The market will not shrink. It will explode downward, into the long tail of businesses the service industry never bothered to visit.

The chair nobody writes from

Those essays all share one piece of furniture. Every one of them is written from the vendor's chair. How to sell work as a product. How big the prize is. Start with work that is already outsourced, where the budget line exists and nobody's pride is at stake, then expand inward. It is good advice, for vendors. Almost nobody writes from the buyer's chair. So let me.

When a vendor's engine resolves your ticket, the ticket gets resolved. Something else happens too. The engine learns. It learns your customers, your edge cases, your escalation paths, the strange exception in your returns policy that takes a human a year to absorb. That learning does not accrue to you. It accrues to the engine, and the engine belongs to the vendor. Sequoia says this out loud, as a feature. The winning autopilots will build compounding data advantages out of their customers' work.

Labour arbitrage built the last services industry. Learning arbitrage is building this one.

For forty years, outsourcing ran on labour arbitrage. Someone else's people, cheaper, doing your work. The new model swaps it for something stranger. Someone else's machine, faster, doing your work, and keeping the lesson. Nothing in the model requires the veteran it is training on your work to serve only you. You are paying, per outcome, to make someone else's asset smarter.

Last essay we borrowed Satya Nadella's test. You can offload a task, or even a job. You can never offload your learning. Buying your work wholesale, as finished outcomes, from an engine that remembers everything and belongs to someone else, is offloading your learning. It does not fail the test narrowly. It fails it completely, and by design.

The trap, round two

We have seen this movie. Round one was the SaaS trap. You rented a tool that almost fit, hired an army to bridge the gap, and discovered years later that the subscription owned you. Round two is not about the tool. You are renting the work itself, and folded inside the work, invisibly, the learning loop. Round one cost you customisation. Round two costs you compounding.

Two honest complications, because this is an opinion series, not a brochure. First, accountability refuses to become a product. A services firm never really sold hours. It sold a signature. Someone carried the liability, someone could be summoned, someone was insured. Software has spent fifty years disclaiming exactly this, in every licence ever shipped. Until someone prices the risk of the work being wrong, the last moat of the services firm is not judgment. It is blame.

Second, the trap has a floor. A five person shop should take the trade and not look back. Its learning loop is not yet worth owning. The argument bites at the size where the lessons inside your work become the most valuable thing you produce. Past that line, handing the loop to a vendor stops being efficiency and starts being a slow transfer of the firm itself.

Whose software

So I believe the thesis and distrust its direction. The work will become software. That part is settled, and no amount of professional nostalgia will unsettle it. The only question still open is whose software the work becomes.

There are two endings. In one, the engine lives in the vendor's cloud, and service as a software becomes the deepest outsourcing ever invented, the first that ships the learning out with the labour. In the other, the engine is built inside your walls, on your stack, reading your memory, and the work becomes your software. Same technology. Same economics. Opposite endings.

The firms that understand this will buy outcomes for the work that does not matter and build engines for the work that does. The firms that do not will wake up in five years lean, efficient, and hollow, nothing left inside that knows anything. The work becomes software either way. Ours is becoming ours.