Software that finishes the task

Rule based automation was brittle. It handled the path it was programmed for and broke on everything else, which meant a person still sat behind it handling exceptions. Systems that reason about exceptions rather than failing on them change that ratio.

Historically enterprises spent heavily on robotic process automation and got real but limited returns, because maintaining the rules consumed much of the benefit. That disappointment shapes how buyers approach the current wave.

The structural driver is back office cost in processes nobody wants to redesign: claims intake, order processing, accounts payable, onboarding, reconciliation. The underlying systems will not be replaced, so the work must be automated around them.

The technology layer spans document understanding, workflow orchestration, agent frameworks with tool access, permission and approval systems, exception routing to humans, and the observability that shows what an agent did and why.

Adoption economics are measured in cost per transaction and in headcount avoided rather than removed, since most organisations redeploy rather than cut.

The beneficiaries include automation platform vendors, document understanding specialists, process mining firms that find what to automate, and the services companies implementing it.

The value chain runs from process discovery through orchestration to execution and audit. Orchestration with proper permissioning is the defensible layer, not the model.

The overlooked layer includes process mining vendors, document extraction firms, audit and observability tooling for agents, and the business process outsourcers automating their own delivery.

Competitive dynamics are unstable, with incumbent automation vendors defending against newer agent platforms and the large software suites embedding agents into applications directly.

Risks: buyers are sceptical after the previous automation cycle, agents that act carry liability, integration into legacy systems is the hard part, and suite vendors may absorb the category.

What to watch: transaction volumes automated, exception rates over time, incumbent automation vendor renewal rates, and agent capabilities added to major enterprise suites.