OpenAI's enterprise AI usage gap hit 8.3x. Here's what frontier firms do differently (2026)

1/10

OpenAI's measure of the enterprise AI usage gap widened from 2.6x in January to 8.3x in June.

OpenAI's Aug. 12 article, From assistance to execution: How enterprises put AI to work, explains what the companies at the frontier are doing differently.

It is less about access to a better model and more about turning individual workflows into shared systems. 1

2/10

The article's main argument:

Enterprise AI is moving from answering questions to completing work.

That means agents need company context, connected tools, permissions, review, and repeatable workflows.

OpenAI supports the argument with two studies: one on agentic AI in its enterprise customer base, and one on adoption across companies, roles, and seniority. 1

3/10 — The gap is measured in depth of use

OpenAI ranks enterprise customers each month by output tokens per active user.

Frontier firms: the top 10%
Typical firms: the 45th-55th percentiles
June gap: 8.3x
January gap: 2.6x

OpenAI calls this a proxy for depth of use. It is a usage measure, not a direct measure of business impact. 1

4/10 — Agents are doing longer jobs

As of June, Codex generated 64% of combined Codex and ChatGPT output tokens among OpenAI's enterprise customers.

The article's explanation: agentic workflows usually produce more output because they handle longer, multi-step tasks.

The shift is practical: an agent can gather information, use tools, create files, and prepare work for review instead of stopping at an answer. 1

5/10 — The frontier firms wire in reusable context

Among weekly active users:

21% at frontier firms use Plugins, vs. 9% at typical firms
19% use skills, vs. 3% at typical firms
95% of OpenAI employees use Plugins weekly

OpenAI's example is a sales workflow that combines a team playbook with current CRM data and past proposals.

The lesson: reusable context plus connected actions beats a blank chat box. 1

6/10 — Adoption is spreading beyond engineering

Since February, weekly active enterprise Codex users grew:

108x in legal
41x in sales
41x in recruiting
26x in marketing
5x in engineering

OpenAI says its Enterprise Signals analysis draws on more than 10 million messages.

The point is not that every function needs the same agent. It is that agentic work is moving into functions where research, drafting, and review repeat. 1

7/10 — Early-career workers are using AI more

Six months after adoption, early-career employees sent 13 more messages per week than executives, according to the article's analysis of millions of conversations.

OpenAI's suggested response for leaders: find the employees with strong AI habits, make their workflows visible, and help those practices spread.

That turns adoption from a training slogan into an internal pattern library. 1

8/10 — What the article says leaders should build

OpenAI's prescription has three parts:

Connect agents to the context and tools needed for valuable work.
Add clear permissions, governance, review, and human control.
Turn effective individual workflows into shared ways of working.

The model is only one layer. The deployment system around it determines whether usage stays shallow or becomes operational. 1

9/10 — The line worth saving

OpenAI's article puts the agenda plainly:

"connect agents to the context and tools needed to complete valuable work; establish clear permissions, review, and governance; and help employees turn effective individual workflows into shared ways of working."

That is the difference between giving everyone a chatbot and building an AI operating layer. 1

10/10

If you are building inside a company, use this as a quick audit:

Are agents connected to live company context?
Can they take bounded actions, or only draft text?
Are permissions and review points explicit?
Can a useful workflow be reused by the next person?

OpenAI's full article: From assistance to execution: How enterprises put AI to work

Which layer is still missing in your team's stack?

References

1
OpenAI — From assistance to execution: How enterprises put AI to workopenai.com

References

Related content