The Trust Dividend Report
Trust one agent. Speed up the next.
Every AI agent deployment leaves behind evidence, controls, and judgment. See how leaders reuse them to move faster.
What 300 leaders running enterprise AI agents told us about trust
The first AI agent deployment tests whether you can trust a system to act. The next tests whether you can carry that trust forward. Build the system around the agent once, with limits, human handoffs, and clear accountability, and it becomes reusable infrastructure. That’s the Trust Dividend.
We asked 300 leaders already running AI agents in production what carries over from one deployment to the next. Produced by Atlantic Insights, The Atlantic’s marketing research division, in partnership with Salesforce, their answers show how much trust work can travel with them.
65
%
Of trust work is reusable
71
%
Say trust speeds rollout
+
21
%*
Better outputs from trust
Survey of 300 U.S. business and IT decision-makers, director level and above, at organizations with 300+ employees, all with an AI agent in production. Figures are respondent estimates, not measured results.
*+21% is the average improvement respondents attribute to their trust capabilities in the accuracy, acceptance, or quality of agent outputs, among those who could quantify it based on their organization’s measurement or best available estimate.
“Once you get a process to agent scale, every subsequent process has to be agent scale as well.”
Joe InzerilloPresident, Enterprise & AI Technology, Salesforce
Ready to learn more?
Become an Agentic Enterprise: A Refreshed Step-By-Step Guide
The Grown-Up's Guide to Operating an Agentic Enterprise
See Salesforce in action
Frequently asked questions
The return organizations earn when trust becomes reusable infrastructure rather than a recurring cost. Each AI agent deployment leaves behind evidence, controls, and operating judgment that make the next one faster and better grounded.
You can’t count on an agent always being right, so trust is the system around it: what the agent may do, the evidence supporting that authority, the conditions for human intervention, and who is accountable for the outcome.
Respondents estimate an average of 65% could be reused for a new deployment. Of those who gave an estimate, 81% say at least half could carry forward. Each new use case still needs its own review where things change.
It can, when requirements surface late. 77% of respondents say a deployment or autonomy expansion has at least sometimes been paused, reworked, or re-reviewed for that reason. But 71% say trust requirements speed deployment of a similar agent.
The report argues “human in the loop” is too blunt. Human judgment matters most before money moves, before a customer is contacted, before sensitive data is accessed, and when an exception falls outside the agent’s proven competence.
96% of respondents use more than one platform or vendor environment for agents, and 58% say that makes consistent trust, governance, or security standards very or extremely difficult.
Produced by Atlantic Insights, The Atlantic’s marketing research division, in partnership with Salesforce. It surveyed 300 U.S. business and IT decision-makers at organizations with 300+ employees, all with an AI agent in production, plus a diary study and interviews with three Salesforce executives.