OpenAI showcases agent workflows at three companies
Basis, Clay, and Exa Labs deploy agents in onboarding, account management, and developer integrations. The full post was not retrievable for us.
Illustration: thin ribbons of light run from desk to desk across an empty open-plan office — a depiction of chained workflows, not a photograph of any real company.
OpenAI's own post presents Basis, Clay, and Exa Labs as companies running AI agents inside onboarding, account management, and developer integration work — with no impact figures we were able to verify.
At a glance
- The post is published by OpenAI itself — vendor communication, not an independent case study.
- Companies named: Basis, Clay, and Exa Labs.
- Use cases named: onboarding, account management, developer integrations.
- Retrieval on September 2, 2026 failed: the server returned HTTP 403.
- Metrics, quotes, and dates from the post could not be checked by us.
OpenAI has published a post of its own describing how three companies — Basis, Clay, and Exa Labs — fold AI agents into work that already exists: bringing new customers on board, running day-to-day account management, and wiring things up for developer teams. Beyond that mapping of company to use case, nothing can currently be confirmed.
What holds up
The frame is solid. OpenAI is the publisher, the intended readers are enterprise decision-makers, and the named examples are Basis, Clay, and Exa Labs. The three areas of use appear explicitly in the summary available to us.
Everything beneath that is open. Which tasks the agents actually handle, which models sit underneath them, how long the systems have been running, and what effect was measured — none of it reached us in a form we could check.
Why the careful wording
The post comes from the vendor whose technology is being described. That does not make it worthless, but it changes the evidentiary weight: vendor case studies select the presentable cases, and the numbers inside them are usually customer-reported rather than independently audited.
There is also a practical obstacle. Our newsroom could not reach the page on September 2, 2026; the server answered with status 403. What follows rests only on the short summary we had.
Tool or workflow — where the difference sits
The phrasing in the headline points at a question many companies have not settled: whether a language model stays a tool that individual employees open when they feel like it, or becomes a fixed step in a process that starts without anyone triggering it by hand. The second case demands clear ownership, logging, and a rule for what happens when the agent is wrong.
That reading is ours, not the post's. What OpenAI argues on the point, we were unable to read.
What leaders can do with this
Anyone planning to cite the post in an internal debate should open it first and hold its claims against their own numbers. It is useful as a list of places agents are being pointed at — onboarding, account management, developer integrations — not as proof that they worked.
We will extend this article once the full text becomes reachable, or once a second, independent newsroom reports on the same examples.
FAQ
Which companies does OpenAI name in the post?
Basis, Clay, and Exa Labs. Whether the full text names others is unknown to us.
What numbers does the post give?
None that we have. The page was not reachable when we tried, so we report no metrics from it.
Is this an independent case study?
No. OpenAI publishes it, and OpenAI supplies the technology described; the examples are vendor-selected.
