Products built as AI · two vendors
76%
Fin. Of the conversations it takes part in. More than 12,000 customers. $0.99 an outcome.
Salesforce's release calls that numeral a share of all support volume.
Same numeral. Not two measurements.
Fin counts a resolution when the customer confirms the issue is solved or does not follow up. A planned handoff can count. Nothing is charged if the customer asks for a person. Sources: Fin and Salesforce, September 10, 2026. The people who publish this site work at OneReach.ai, which sells an agent platform. OneReach.ai is not in the Agentic Index. Conflict of interest.
| Figure | What it is of | Limit |
|---|---|---|
| 6× | Messages from the 95th percentile ChatGPT Enterprise worker, against the median worker. | Both ends already use ChatGPT Enterprise. OpenAI's logs, published December 2025. |
| 7× | Messages frontier firms send to Custom GPTs, against the median firm. | Same logs. Both ends are firms that already use the product. |
| About 2× | Messages per seat at frontier firms, against the median firm. | Same comparison. Not a share of all firms. |
| More than 80% | Code merged into Anthropic's codebase that Anthropic said was written by Claude, as of May of 2026. The engineer directs and reviews. | A count of merged code. Not Google's count. |
| 75% | New code at Google that Google said is AI-generated and approved by engineers, up from 50% the previous fall. | A count of new code. A different event from Anthropic's merged code. |
| 0% to 20% | The share of their work that more than half of 132 Anthropic engineers and researchers said they could fully hand off. Survey published December 2025. | Claude, which helped compile this research, is Anthropic's. |
| 70% | HubSpot's stated average resolution rate for its customer agent. $0.50 a resolution. Counted when no person took the conversation for 72 hours, or the agent qualified a lead. | The page gives no base and no period. HubSpot sells the agent. |
| About half | Of 1,082 Agentic Index vendors, data of October 29, 2026, fully document a control that lets a person approve an agent's actions. | Documentation, not whether the product works. |
| About one in ten | Fully document memory that lasts beyond a single session. The fewest of the index's 14 capabilities. | A share of vendors, said in words, not drawn as a percent. |
| About one in five | Of its agent platforms, fully document all four basics the index sets for enterprise buyers. | Platforms, not all 1,082 vendors. |
| 328, 200, 145, 144, 141 | Vendors in back-office operations, agent infrastructure, agent builders, customer support, and sales prospecting. A vendor can sit in more than one. | Counts of listings, not use. |
Uber used up its full-year 2026 AI budget within about four months, driven by coding tools such as Claude Code, and now caps each employee at $1,500 a month per AI coding tool. The Information reported it. Claude Code is Anthropic's product, and Claude helped compile this research. The full page.