In February, both Anthropic and OpenAI launched enterprise agent platforms within three weeks of each other — a coincidence of timing that probably reflects shared market intelligence more than coordination. Anthropic's push introduced plug-in agents pre-built for finance, engineering, and design workflows, designed to slot into existing enterprise software stacks without requiring custom development. OpenAI launched Frontier, a platform framed explicitly around treating AI agents like employees: giving them defined roles, managing their access to systems, and evaluating their output against performance criteria. Both launches signal the same underlying shift: the enterprise AI agent market has moved from custom bespoke deployments to standardised product offerings, which is what happens when a market matures from early adoption to mainstream.
The results data that has started to emerge from early production deployments is striking enough to explain the investment momentum. A major manufacturer using agents for production optimisation cut a process that previously took six weeks to a single day. A global investment firm deployed agents across its sales workflow, freeing up over 90% of salesperson time previously spent on administrative tasks for direct client work. An energy company reported a production output increase of up to 5% from agent-assisted operations — a figure that, at the scale of a large energy producer, translates to over a billion dollars in additional annual revenue. These are not productivity improvements in the range of a few percent; they are structural changes in how work gets done.
Anthropic's research team published a paper this year attempting to formally measure AI agent autonomy in practice — a harder problem than it sounds. Autonomy in this context is not binary; it exists on a spectrum from a tool that responds to individual queries, through a system that executes multi-step workflows with occasional human checkpoints, to a fully autonomous agent that pursues goals over extended timeframes without supervision. The research found that most current enterprise deployments cluster in the middle of that spectrum: agents handle well-defined task sequences reliably, but companies are reluctant to extend autonomy to decisions with significant irreversible consequences. That reluctance is rational, but it also defines the ceiling of value capture in the near term — the biggest gains come from the most autonomous deployments, and most enterprises are not ready to go there.
The sector breakdown of adoption is telling. Telecommunications leads enterprise AI agent adoption at 48%, followed by retail and consumer packaged goods at 47%. These are sectors characterised by high transaction volumes, repetitive process steps, and tolerance for marginal error rates — exactly the operating environment where agents perform most reliably. Healthcare and financial services, where error rates carry regulatory and liability exposure, are moving more slowly. The adoption curve is not uniform across industries; it follows the profile of which industries have the operational characteristics that match what current agents can actually do well.
The honest caveat is that agents will still be in early adoption by the end of 2026. The technical barriers are real — integrating agents into legacy enterprise software stacks is expensive and slow, agent reliability degrades on tasks outside their training distribution, and the failure modes of autonomous systems are harder to audit than the failure modes of tools that require human input at each step. The compliance and liability questions are even less resolved than the technical ones. Enterprises operating in regulated industries need answers to questions about agent decision audit trails, liability for agent errors, and employee rights implications of workflow automation that regulators have not yet provided.
What has changed in 2026 is not that all of these problems are solved. It is that enough companies have seen enough evidence of transformative productivity gains that the risk calculus of waiting has flipped. The companies that move fastest to solve the integration and compliance problems are capturing real competitive advantage. The companies that wait for the problems to be solved before adopting are watching that advantage accumulate against them. This is the familiar dynamic of technology adoption curves — the same one that played out with cloud infrastructure, mobile, and enterprise software before it — and it tends to resolve in favour of early movers faster than most incumbent organisations expect.