Salesforce published its 2026 Agentic Enterprise Index on August 7, 2026, and the headline number is striking: the average enterprise now runs 13 AI agents in production, up from 5 in February 2025. That is a 3x increase over 14 months, compounding at 7% per month. Measured in total work output, Agentforce agents had completed 734 million Agentic Work Units as of April 2026, a metric growing at 15% compounding monthly.
The report is the clearest quantitative snapshot to date of what enterprise AI agent deployment actually looks like inside a single large platform. For GTM, operations, and AI strategy teams wondering whether the agent narrative is maturing into measurable business impact, this index is worth examining carefully, both for what it shows and what it does not yet prove.
What the Data Actually Shows
Salesforce drew on aggregated usage telemetry from companies that consistently ran agents on Agentforce from February 2025 through April 2026. The qualifying threshold mattered: only organizations with agents active in production every month across the full period were included, filtering out one-off experiments.
Key metrics from the report:
| Metric | Feb 2025 | Apr 2026 | Growth |
|---|---|---|---|
| Avg agents per org | 5 | 13 | ~3x (7% CMGR) |
| Time to first agent (days) | 4.0 | 1.9 | 53% faster |
| Avg agent skills | 2 | 6 | 3x |
| AWU volume (cumulative) | Baseline | 734M | 15% monthly CMGR |
| Actions per account | Baseline | +31% CMGR | Sustained acceleration |
| Employee agent engagement | Baseline | +300% per week | 15-month trend |
Sources: Salesforce Agentic Enterprise Index 2026, CIO.com coverage, August 7, 2026.
The “skills” figure deserves attention. At the start of 2025, the average Agentforce agent could perform two distinct types of actions, typically within a single system such as answering an FAQ or checking an order. By April 2026, that had grown to six skills, including retrieving and summarizing data across systems, drafting communications, updating records directly, and extracting structured data from user inputs. This shift from single-system chatbots to cross-functional process agents is the architectural change enterprise buyers have been waiting for.
The Industry Breakdown
Not all enterprise sectors are deploying agents the same way, and the index surfaces a meaningful split that frames where agents are delivering value today.
Consumer-facing industries deploy at high volume for narrow, task-specific work. Retail agents averaged one to two actions per interaction for most of the year, handling order status, returns, and product recommendations. During peak holiday shopping periods, that jumped to nine skills per agent, a 350% increase, as businesses pushed agents into more complex multi-step customer-service workflows. Retail companies that deployed agents also recorded 4x higher online sales growth than non-deploying peers, according to the index, though Salesforce’s report does not isolate whether agent deployment caused the growth or whether growth-oriented companies are also more likely to invest in automation.
Operationally complex industries such as manufacturing, financial services, healthcare, and the public sector are deploying fewer agents in raw count but configuring them for cross-system orchestration. Public-sector agents saw a 227-fold increase in AWU volume year over year. Healthcare agents saw a 19-fold increase. Financial services agents accounted for roughly 10% of total monthly AWU output across the platform, with spikes driven by tax season workloads.
The manufacturing case study that Salesforce highlighted is the most concrete evidence of multi-agent orchestration at scale. Siemens, which runs 18,000 sellers across seven business units and processes roughly 2,800 unqualified inbound leads every week, deployed a coordinated two-agent workflow: one agent engages and nurtures incoming leads in real time while a second agent concurrently gathers missing qualification data, applies scoring rules, and routes qualified leads with cross-division context attached. The system operates around the clock without human handoffs for initial qualification. This is the “headless architecture” pattern that analysts at CIO.com described as becoming a practical necessity as agents move across multiple cloud systems.
The Customer Service Signal
The customer service numbers are the most directly comparable across vendors and the hardest to dismiss as vendor-defined metrics. Agentforce agents handled 170 times more customer service chats than in previous years while resolving 7 in 10 conversations without human assistance. Separately, Salesforce research found that 77% of shoppers who interacted with a branded AI shopping agent on a website felt more confident in their purchase than those who did not.
For enterprise teams assessing whether AI agent investment is worth the governance overhead, the resolution rate metric is meaningful. A 70% no-touch resolution rate on inbound customer contacts, if sustained at scale, changes headcount math in contact centers.
Internally, Salesforce reports that its own Slackbot (the AI agent embedded in Slack that can summarize conversations, draft content, and retrieve information from across channels) is now used regularly by 83% of employees and saves the average employee five hours per week. Agent sessions within the company grew three times between February 2025 and April 2026.
What the Index Does Not Prove
The Diginomica analysis of the index is worth reading alongside Salesforce’s own framing. Adoption growth rates are impressive, but the overwhelming majority of Salesforce’s customer base has not yet activated persistent agents. The AWU metric, while useful for tracking activity, has drawn analyst criticism for not tying directly to business outcomes such as revenue uplift or cost reduction. And the index publishes shortly before Dreamforce, Salesforce’s largest annual conference, timing that shapes the context in which readers encounter the data.
The MIT research published in mid-2025 that found many enterprise AI investments had not delivered expected value is still the counterpoint. The gap between agents deployed and agents delivering measured ROI remains real for most enterprise buyers. The July 2026 AI pacing letter from frontier lab leaders underscored that enterprise AI deployment is accelerating faster than governance frameworks are being built.
The Salesforce index does not contradict either of those points. It shows that within a committed cohort of production Agentforce users, usage is growing, agents are becoming more capable, and some measurable outcomes (retail sales, customer service resolution, employee time savings) are correlating positively with deployment. That is meaningful. It is also not the same as an independent audit of enterprise AI agent ROI across the market.
What Enterprise Teams Should Take From This
Three practical signals stand out for GTM and AI strategy teams:
Deployment speed has become a non-issue. 1.9 days from provisioning to production agent means the experiment-to-production cycle is now measured in days, not quarters. Teams that have been hesitating because of implementation timelines should revisit that calculus.
Cross-system agent skills are the inflection point. The jump from 2 to 6 average skills per agent, and especially the Siemens pattern of coordinating two specialized agents across data gathering and routing, is where operational leverage actually lives. Single-system agents answering FAQs deliver incremental value. Agents orchestrating multi-step workflows across CRM, data, and routing logic deliver structural change.
Industry-specific deployment patterns matter for benchmarking. Consumer-facing businesses should benchmark against the 170x customer service chat figure and the 4x retail sales correlation. Regulated industries should look at the healthcare and public sector AWU growth trajectories as signals that compliance and complexity concerns have not blocked deployment where organizations committed to it.
Book a call with Enera to map where AI agent orchestration creates the most leverage for your GTM and operations workflows.