Salesforce says agents per org went 5 to 13 and build time fell to 1.9 days. It sells the agents.

Salesforce has published aggregate telemetry from its Agentforce platform showing the average number of production agents per organisation rising from five in February 2025 to 13 in April 2026, and the time to create a new agent falling 53%, from four days to 1.9 ^[1][2][3]. That is the first deployment-velocity figure operators can hold a roadmap against, and it comes from a vendor whose revenue depends on the figure going up.

The measurement base is better than most vendor research. According to Salesforce, the index tracks AI engagements in production across five consecutive quarters at 400 real businesses ^[4]. It then bundles that telemetry with a survey of nearly 5,000 respondents across nine markets ^[5], which is where the rigour thins out: production logs and stated intent are not the same evidence, and the report presents them under one index.

Take the build-time number at face value and it reframes the work. Four days to 1.9 is a 52.5% reduction ^[1], and 13 agents at 1.9 days each is roughly 25 days of build effort to reach the current average ^[2]. If a year of agent rollout costs a month of construction, the constraint was never the builder. It is integration, permissions, ownership of the escalation path, and deciding which process is stable enough to hand over.

The scope figures are the more interesting part. Average unique actions per agent went from two to four, with a retail peak of nine ^[6]. The share of functions that are secondary rather than primary rose from 1% to 6% ^[7]. The average monthly share of agentic action went from zero in early 2025 to 15% by April 2026 ^[8]. Salesforce also reports that consumer-facing industries such as retail and travel lead on deployment speed and overall use, while regulated industries adopt more slowly and account for a lower share of work units, but run more sophisticated agents ^[9]. That split is the useful planning signal: velocity benchmarks from retail do not transfer to a bank.

The ROI section is where the marketing shows. The strongest evidence offered is that weekly employee sessions with agents tripled year over year to nearly eight per week ^[10] while customer escalations stayed steady ^[11]. Steady escalations under tripled volume is a real operational result, but it is the absence of a bad outcome, not a return. There is no cost per unit of work, no deflection rate, no headcount or handle-time figure.

Two numbers in the report should be treated with care. Salesforce says it saw 734 million Agentic Work Units consumed in April 2026, described in the same breath as a 15% month-over-month increase in the action-calls-to-output-token ratio ^[12][13]; a count of work units and a token ratio are not the same quantity, and the AWU total is attributed to Salesforce rather than scoped to the 400-company panel ^[4][12]. Separately, the claim that agents improved their capabilities by 350% ^[14] arrives with no stated baseline. Note too that 5 to 13 is 2.6x ^[3], reported as "nearly threefold" ^[2] and headlined as tripled.

Watch whether Salesforce repeats this index against the same 400-company panel rather than a refreshed one ^[4], whether the escalation rate holds as agentic share climbs past 15% ^[8][11], and whether AWU, a Salesforce-defined unit ^[12], migrates from a research metric into a pricing metric. Until a buyer-side body publishes comparable telemetry, 1.9 days is the benchmark, with the seller's thumb on it.