For eighteen months, Marc Benioff had one message. Copilots are dead. Human oversight is a crutch. The future is fully autonomous agents that complete entire workflows without asking permission.
He called Microsoft Copilot “Clippy 2.0.” He told investors that Agentforce would “end the era of the copilot.” He positioned Salesforce as the company brave enough to let agents operate alone while Microsoft still needed a human holding the leash.
And Agentforce sold. It crossed $1.2 billion in annual recurring revenue. Over 8,000 customers bought into the vision. The pitch was clean: your agents work for you, not with you.
Then Salesforce’s own admin blog published a post titled “The Importance of Human in the Loop for Agentforce.”
The qualifying line, written on the official Salesforce Admins site: “the most effective AI agents aren’t autonomous solutions; they’re tools that give the humans who use them superpowers.”
Read that again. Not “some agents need guardrails in certain edge cases.” Not “governance is important alongside autonomy.” The post states plainly that the most effective agents are not autonomous. They are tools. For humans.
This is not a competitor saying it. This is not an analyst hedging. This is Salesforce’s own platform team, writing to the administrators who deploy Agentforce in production, telling them that human oversight is not a limitation to work around. It is the design.
Meanwhile, the Agentforce community has built an entire external resource cataloging the workarounds needed for what autonomous agents cannot do in production Salesforce orgs. The gap documentation is not written by Salesforce. It is written by unpaid practitioners patching the distance between the keynote and the deployment.
ServiceNow’s president carefully used “agentic collaboration” in his 2026 messaging. Not autonomy. Collaboration. He read the same deployment data and chose a different word. Salesforce used the word “autonomous” until its own administrators needed the truth.
The company that built a $1.2 billion business mocking human oversight quietly published a blog post explaining why human oversight is the product.
Clippy came home.
👉 Read the agentic security news. Instantly analyze the threat vector, see if it applies to your setup, and find the gaps with our interactive playbook. All free.
👉 Follow me on LinkedIn | X | Substack for weekly analysis of real agent failures, control gaps, and what the frameworks are and are not catching.
Comments
Nothing yet. Say the first thing.
Sign in to join the conversation.