Salesforce killed its own UI for AI agents
By Alfred Belvedere — Founder, Omni AI
This week Salesforce did something its competitors will spend the next 18 months trying to undo. It exposed its entire platform — every object, workflow, and permission boundary — through a headless API designed for AI agents to consume directly, no UI required. Mistral shipped Workflows, an orchestration layer that moves agents from sandbox to production processes. And OpenAI's models are now available natively through AWS, ending the Microsoft-only distribution lock that defined the last 18 months. Three moves, one shape: the product surface your customer actually uses is no longer your screen. It is your API for their agent.
Today’s Key Insights
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.
Power Move
Pick the single most-used screen in your product (or a product you sell into) and map it to four artifacts by Friday: (1) the JSON schema an agent needs to operate it, (2) the auth model with scoped tokens, (3) the rate limits, (4) one webhook event that fires on completion. Hand the spec to your engineering lead with a date attached. If your product cannot answer those four questions in a public docs page by end of May, you are pricing yourself like a UI in a market that has stopped buying UIs.
Salesforce killed its own UI for AI agents
That’s the signal — here’s the move. Book a free 30-minute strategy session and we’ll walk through exactly how to apply today’s insight to your revenue, your team, and your next 90 days. No pitch. Just straight advice from operators who run AI systems for a living.
30 minutes · free · no obligation
Powered by Omni AI
More from Interlinked
Why Decagon at $4.5B exposes Sierra's pilot trap
Last week Decagon closed a $250M Series D at $4.5B — three times its valuation six months ago, on roughly $35M of ARR. Sierra, with 4× the r…
NVIDIA just funded Harvey's $5.6B Swedish rival
Three weeks after Harvey hit $11B at 58x ARR, NVIDIA quietly wrote a check to the company hunting it. Legora — Swedish, $100M ARR, four year…
Microsoft just turned agent chaos into a $99 SKU
At 9 AM Pacific today, Microsoft 365 E7 went generally available with Agent 365 inside it — a single control plane for every AI agent your t…
Today's trending · Omni AI
Big Companies Are Firing Teams to Fund AI. Is Yours Ready?
Atlassian lays off 10% of workforce to redirect resources toward AI
See the trending brief →