Where AI Is Going: From Copilots to Agents That Do the Work

For the past few years, most businesses have experienced AI as a helper: a chatbot that drafts an email, a copilot that suggests code, an assistant that summarizes a meeting. Useful, but incremental. The person still did the job, and AI made them somewhat faster. That is changing. The systems now arriving can take a goal, break it into steps, use software tools, check their own work, and hand back a finished result. The industry calls them agents. For investors, the move from assisting work to completing it is the most important development in AI since the technology went mainstream.

From software that helps to software that does

The economic difference is large. Traditional software sells tools to people, usually priced per user. An agent sells an outcome: a claim submitted, an invoice reconciled, a prior authorization filed, a support ticket resolved. That moves AI out of the software budget and into the much larger amount companies spend on labor and outsourced services. Whatever label it carries, it means the market for AI is no longer measured in software seats but in the cost of the work itself.

It also explains why public markets have been so volatile around software companies this year. If a business charges per seat and its customers need fewer people to do the same work, its pricing model is under pressure even when its product is good. Investors now ask every software company the same question: does AI make your product more valuable, or does it make your customers need less of it?

Where agents land first

Early adoption is concentrated where work is high-volume, rules-based, and well documented, and where a review step catches errors. That describes much of the back office in healthcare and business services: revenue cycle management, claims processing, eligibility checks, scheduling, accounts payable and receivable, customer support, compliance documentation, and IT service management. These functions employ large numbers of people, follow defined procedures, and produce measurable results, which makes them natural places for agents to work alongside staff.

Regulated work moves more carefully, and it should. In healthcare, an agent that drafts an appeal or assembles an authorization request still needs a person to review it, an audit trail, and controls on patient data. The companies that win will not be the ones that remove people fastest. They will be the ones that redesign the workflow so people review and decide while software does the assembly.

What becomes more valuable

When the cost of routine work falls, value shifts to what remains scarce. We see four things gaining value. The first is ownership of the workflow and the system of record, because agents need somewhere to act and someone has to own the data they act on. The second is proprietary data and context, since an agent is only as good as the information it can draw on. The third is trust and compliance in regulated markets, where customers will choose a provider they can audit over a cheaper one they cannot. The fourth is distribution: established relationships with customers who would rather buy outcomes from a partner they know than assemble tools themselves.

What becomes less valuable is just as clear: undifferentiated labor arbitrage, per-seat software with little proprietary data, and services businesses whose only advantage is the size of their workforce.

What this means for investors

For a services business, AI is either the largest margin opportunity in a generation or a serious threat, and which one depends on management. A revenue cycle company or an IT services firm that uses agents to handle more volume with the same team can grow faster and earn more per client. One that waits may find a competitor offering the same outcome at a lower price. In diligence, we now ask how a company uses AI today, what it has measured, how it plans to price as its costs fall, and whether its leaders understand the technology well enough to direct it.

Founder-owned businesses in essential industries often have an advantage here that is easy to overlook. They own customer relationships, operating know-how, and data that AI-native startups lack. Paired with capital and technical expertise, they can put agents to work faster than a startup can earn a customer's trust.

The Kiron take

We think the move from copilots to agents is real and will reshape labor-intensive industries over the next several years, unevenly, and more slowly in regulated settings than the headlines suggest. Our approach is to back companies that own the workflow, hold proprietary data, and have earned trust in regulated markets, and to help them apply AI to the economics of their own operations. We would rather help an established business capture that value than watch a new entrant take it.