The Agentic Enterprise: Moving Toward the Autonomous Future
What Dreamforce 2026 Signals About the Architecture, Economics, and Governance of Autonomous Work
TL;DR
The stack, not the feature set, is what changed. A year ago, agents were features bolted onto applications. At Dreamforce 2026, Salesforce redefined four things at once: the interface, the workforce, the model, and the control layer. That is a re-architecture, and it is the clearest signal yet that an incumbent is reorganizing around agents rather than decorating its existing products with them.
AIforce operationalizes the death of the interface as the center of gravity. Under the banner "AI replaces the UI," Salesforce is pushing its platform out to where people already work, including Slack, Claude, and custom surfaces. The interface layer is being demoted from destination to delivery, which is the shift we described in our April Future Enterprise report.
Koa marks an incumbent building its own model. A CRM vendor shipping a reasoning model built on NVIDIA Nemotron is a moat-and-efficiency play against the horizontal labs. It is also benchmarked largely on Salesforce's own CRM Bench, which is a caveat worth stating plainly.
Pricing moved decisively toward consumption. The shift from per-conversation charges to per-action flex credits is the consumption model we forecast when we argued that per-seat pricing breaks in an agentic world. It also surfaces the metering problem most enterprises have not solved.
The control plane is the real strategic layer. Agent Fabric and the "Enterprise AI Harness" providing the infrastructure for discovering, governing, and orchestrating agents across vendors. It is maturing, not finished, and the hardest part, cross-organizational trust, remains unsolved by any vendor.
Autonomy is arriving in bounded form. "Job-ready" agents, most generally available, operate under a "human at the helm" model. The distance between agentic marketing and autonomous reality is exactly where enterprise judgment earns its keep.