On 26 August Salesforce reported its quarterly results. The Agentforce figure:
- more than $1.5 billion in annual recurring revenue;
- up 240% in a year.
Just below sits the note: from this quarter the number also includes Slackbot and Headless 360. The growth is real. How much of it is agent adoption and how much is a wider scope, the release does not say. And the number that would say the most is missing: how many customers have an agent in production, and how many are still in pilot.
Revenue says someone bought it. It does not say anyone uses it.
In 2024 I said Agentforce would generate a lot of early excitement and burn out quickly for many, because customers and consultancies were not ready. Today I would not say it has burned out. The risk I see is different: that it stops at the pilot.
Why pilots stay pilots
Almost never because of the technology: an agent gets configured in days. The projects that leave the pilot had three things before they started.
- A number decided up front: hours given back, fewer requests for help. Not "we have an agent".
- A written criterion for stopping, if the number does not move.
- Three names on a sheet of paper: who answers for security, who for compliance, who has the final word.
Without the first, the pilot cannot prove anything. Without the second, it cannot even fail. And a pilot that can neither succeed nor fail stays a pilot forever.
A pilot without a number decided up front is not an experiment. It is a demo with a bill.
What I would do on Monday
I would take every open Agentforce pilot and ask one question: which number, if it moves, takes us to production? If nobody knows, the pilot is not stuck because of Agentforce.
How I decide what is worth doing.
In your company, how many AI pilots have a written criterion for stopping?
Sources
- Salesforce, second quarter fiscal 2027 results, 26 August 2026