From Selling Software to Selling Work: What AI Agents Change About Enterprise Budgets

For thirty years, enterprise software has been sold the same way: a tool, a licence, a seat. The vendor gives your team something to operate, and your team supplies the labour. Whatever value comes out is a product of both, which is why software budgets and payroll budgets have always been separate lines on the P&L.

AI agents quietly break that model. An agent doesn’t hand your team a tool. It performs the work itself. When a system can nurture a lead, chase an invoice or resolve a support ticket end-to-end, what you are buying is no longer software, it is completed work. Commentators on the enterprise AI shift, including Box chief executive Aaron Levie, have made a version of this point repeatedly: the opportunity for AI vendors is not the software budget at all, but the far larger pool of money organisations spend getting work done.

That reframing matters more for mid-market firms than anyone else. A 100-person company was never going to buy another six-figure platform licence. But it absolutely spends money, in wages, contractors and overtime, on lead follow-up, order processing, reporting and customer service. When an agent can take a slice of that work, the relevant comparison stops being “this software versus that software&rdquo. And becomes “this agent versus the fully loaded cost of doing the task manually”.

Three practical consequences

First, pricing logic changes. Seats made sense when value scaled with the number of humans using the tool. Work-based buying points instead toward outcome-based pricing: pay when the task is completed, verified and counted. If a vendor still quotes you per seat for an autonomous agent, the model and the product are misaligned.

Second, the evaluation checklist changes. A tool is evaluated on features. Work is evaluated the way you would evaluate a hire: what is the task completion rate, what is the error rate, how does it escalate what it cannot do, and what does a unit of completed work cost? These are operational questions, not IT questions, which is why agent purchases increasingly sit with COOs rather than CIOs.

Third, the market gets bigger, not smaller. A common fear is that AI shrinks what companies spend. The more likely outcome, as Levie and others have argued, is expansion: huge categories of work that were never economical to do, following up every single quote, auditing every invoice, personalising every customer touch, suddenly become viable when the marginal cost of doing them approaches zero. Businesses will not do the same work with less. They will do far more work with the same team.

What to do with this

Audit your operations for work that is high-volume, rules-plus-judgement, and currently rationed because people are expensive. That backlog, not your software stack, is where agents pay for themselves first. It is exactly the territory our digital staff are built for, and why we price them on proven results rather than seats: if the work is not done, you should not pay.