Market segmentation + GTM
For SMBs, AI agents remove the adoption barrier. For enterprises, they expose the coordination tax.
The dividing line is whether your customer’s workflows are efficiency artifacts (built to get work done faster) or coordination artifacts (built to manage accountability across teams). Agents reduce the cost of the first. They surface the tax embedded in the second.
For SMBs, the adoption barriers were training cost and attention. Small businesses don’t have implementation specialists or dedicated ops teams. The person running the business is context-switching constantly and doesn’t have capacity for software onboarding. General-purpose agents like Claude close that gap.
An SMB operator can now call their CRM, invoicing tool, and scheduling software in sequence through natural language. No specialist required. The operator becomes the integrator. The software that wins is whichever one owns the golden record, because that’s what the agent queries.
For enterprises, general-purpose agents surface coordination complexity that was already there.
In multi-BU organizations, workflows encode agreements about who owns which decision and how accountability crosses organizational seams. A medication order in Epic isn’t a task. It’s a standing agreement between nursing, pharmacy, billing, and compliance. A general-purpose agent disturbs that agreement.
That generates second-order effects: approval chains, compliance logging, and downstream processes the agent wasn’t designed for. The initial fix is human-in-loop subroutines at each checkpoint so stakeholders can audit before the agent proceeds.
Enterprise workflows are discontinuous and asynchronous. A change request submitted Monday goes to a change advisory board Wednesday, a compliance sign-off Thursday, and executes in Friday’s maintenance window. The agent holds state across four days and multiple handoffs. The world doesn’t pause with it. Priorities shift. New constraints emerge. The information the agent acted on may no longer be valid when the approval arrives. Human-in-loop catches problems at checkpoints. Staleness accumulates in the gaps between them.
The complexity keeps building until the organization stops patching and replaces the workflow wholesale with custom automation designed for agents from the ground up. That’s the enterprise endgame.
In SMB, the workflow layer is a general-purpose agent. The software that wins is the one it queries. In enterprise, the workflow layer is custom-built from the ground up. SaaS survives as the system of record that custom automation depends on.
Those are different products, different roadmaps, and different bets. The PM question for 2026 is which customer you’re building for.