THE NUMBERS
● Agents per organisation: 5 in early 2025, 13 by April 2026.
● Time to build an agent: down 53 percent.
● Customer service: 7 in 10 sessions handled autonomously.
● The number that validates it: escalations stayed steady. Volume went up without quality going down.
What was measured
Salesforce's Agentic Enterprise Index analyses production activity from 400 businesses, alongside a survey of nearly 5,000 people. That combination matters — it is telemetry from systems actually running, not a survey of what executives believe is happening.
| Metric |
Change |
| Agents deployed per organisation |
5 to 13 |
| Agent creation time |
Down 53% |
| Employee sessions with agents |
Tripled |
| Customer-service sessions handled autonomously |
7 in 10 |
| Escalations to humans |
Steady |
The escalation number is the whole story
WHY FLAT ESCALATIONS MATTER MORE THAN 7 IN 10
Any support system can raise its autonomous-resolution rate by making it harder to reach a person. That looks like success in a dashboard and feels like failure to a customer.
Escalations staying steady while autonomous handling rose is the evidence the deflection was real rather than a queue being hidden. It is the one figure worth asking for when a vendor quotes you a resolution rate.
Where it is happening
Retail, travel, financial services and the public sector all showed sharp increases in agent activity. Regulated sectors move more slowly, which is the expected pattern rather than a surprise.
The 53 percent drop in build time is the underrated figure. When creating an agent costs half what it did, organisations stop reserving them for high-value processes and start pointing them at ordinary ones. That is how you get from five to thirteen — not thirteen strategic deployments, but five strategic ones and eight that were simply cheap enough to try.
Read it against what we know about coding agents
Linear published telemetry this month from its own paid workspaces: teams using coding agents went from 21 weekly pull requests to 65, while teams without them went from 8 to 10. Total product development time rose anyway, with engineering time on create and triage up roughly 17 percent.
Both datasets are real and they are not in conflict. Customer service has a natural verification loop — the customer either got what they needed or escalated. Software does not; a pull request needs a human to decide whether it is right, and that review scales with volume.
Which lines up with the broader pattern this month: agents perform where a cheap checker exists and struggle where judgement decides.
What to take from it
| If you are... |
The useful read |
| Evaluating a support agent vendor |
Ask for the escalation rate alongside the resolution rate. One without the other is meaningless |
| Deploying agents internally |
Start where a verification loop already exists. Those are the deployments that hold up |
| Running one agent and wondering about more |
The median is now 13. If you have one, you are behind the curve rather than ahead of it |
| Reading the 7 in 10 figure as universal |
It is from organisations already in Salesforce's dataset, which skews toward mature deployments |
FAQ
How many AI agents does a typical company run?
Thirteen, up from five in early 2025, per Salesforce's Agentic Enterprise Index measuring production activity across 400 businesses.
Are AI agents actually resolving customer issues?
Seven in ten customer-service sessions are handled autonomously among organisations in the dataset, with escalation rates staying steady. Flat escalations are the evidence the deflection is genuine rather than a hidden queue.
Why did agent numbers grow so fast?
Build time fell 53 percent. When an agent costs half as much to create, organisations stop reserving them for high-value processes and start trying them on ordinary ones.
Does this contradict the finding that coding agents made teams slower?
No. Customer service has a natural verification loop — the customer either got what they needed or escalated. Software does not, so review scales with volume and lands on humans.
Is the data representative?
It measures organisations already using Salesforce's agent platform, which skews toward more mature deployments. The direction is credible; treat the absolute figures as a ceiling rather than an average.