Salesforce killed its own UI for AI agents

The user-interface era valued pixels and clicks. The agent era values endpoints, schemas, and rate limits. Salesforce going headless is not a product update — it is an admission that a meaningful share of their next $10B in ARR will be billed against API calls that never render a single button. If your roadmap still leads with a screenshot, you are optimizing for a buyer who is being replaced by their own agent.

For 18 months the joke was that AI strategy meant Microsoft strategy. That joke is dead. OpenAI on AWS adds a second hyperscaler to the same model family, Mistral Workflows on European infra adds a third pole, and procurement teams now have a real multi-cloud lever heading into Q3 negotiations. If you signed a single-vendor commit before Easter, your CFO has six weeks to renegotiate before capacity allocation resets and the conversation reverts to list pricing.

Yes — 40% of enterprise applications will embed task-specific AI agents by year-end. The line nobody quotes from the same report: 40% of those projects will fail by end of 2027, mostly on cost and security. The operators winning this quarter are the ones writing the failure mode into the procurement contract now — outcome-based pricing, model-portability, documented kill switch — instead of discovering it during the post-mortem.

If your product has a UI but no agent-callable API with documented schemas, scoped tokens, and webhook events, you are building for a customer being replaced. Not in three years — in nine months. The companies adding MCP-compatible to their landing page this month are doing it because their enterprise buyers are asking the question on the procurement form, not in the demo. The buy-side already shifted. The sell-side has roughly two quarters to catch up.