OpenAI's latest customer writeup shows Basis, Clay, and Exa Labs handing entire onboarding, account management, and developer-outreach workflows to AI agents, not just assisting with them.
Picture the account manager juggling forty enterprise deals, each one scattered across CRM notes, Slack threads, call transcripts, and a dozen half remembered promises from last quarter's QBR. She used to burn close to an hour every night just rereading her own inbox to remember what has to happen the next morning. According to OpenAI's newest customer writeup, that hour is gone. At Clay, a dedicated AI subagent now reviews every account overnight and hands her a short list of priority moves each morning, built while she was asleep.
The example comes from OpenAI's post How AI-native companies turn workflows into operating capability, published today alongside new numbers from the company's ongoing Enterprise Signals research. Frontier companies, the ones in the top ten percent of AI usage, now generate 8.3 times as many output tokens per active user as typical companies. In January that gap was 2.6 times. The companies pulling ahead aren't just using AI more, they're handing entire workflows to it and refining the process every time it runs.
Take Basis, which builds AI agents for accounting firms. First day onboarding used to take two hours of someone's time walking a new hire through systems and paperwork. OpenAI says it now takes thirty minutes. New employees get immediate access to Codex and a company specific onboarding skill that welcomes them, explains how the company works, and quietly finishes account setup in the background, while HR steps in only for exceptions.
Clay, the go-to-market data and orchestration platform that more than 500,000 GTM teams already use for prospecting and enrichment, is the account management example. One of its GTM engineers built a persistent workspace with a dedicated subagent assigned to every account. Each subagent checks primary sources overnight and updates a deal folder. A coordinating agent turns all of that into a morning priority list, with the underlying evidence attached so the rep can double check before acting on anything.
The third example is Exa Labs, which builds the search infrastructure a lot of other AI agents already run on. Exa wanted its API available everywhere developers might reach for it, a goal it calls Exa everywhere. That used to mean developer relations and account teams manually watching repositories for integration opportunities. Now Codex monitors for high priority opportunities, gathers context, opens pull requests, runs tests, and drafts weekly updates and announcements for a human to review before anything ships.
None of this is about a single tool replacing a single job. It's about what happens when a company stops treating "we need someone to do the onboarding" or "we need someone to keep account context current" as a headcount problem and starts treating it as a workflow you can hand to an agent, watch, and improve. That's a different question for a business leader to bring into a budget meeting than "should we buy this AI tool." The real question becomes which of your recurring, well understood workflows are worth turning into a skill an agent can run, and who signs off when it makes a mistake.
Here's the part OpenAI's post doesn't dwell on. Basis, Clay, and Exa Labs are all AI native companies that build agent products for a living, and all three already had engineers on staff who could design, test, and maintain these workflows. Clay's own site still runs a job board for something it calls a GTM engineer, because building an account subagent workspace isn't a settings toggle, it's a technical role. A 200-person SaaS company without anyone in that seat won't get Clay's overnight subagent by flipping a feature on. It gets there by hiring or growing someone into that exact new job first.
The account manager in Clay's example didn't lose her job to an agent. She lost the worst hour of it, the one where she reread her own week just to remember what mattered. The companies figuring out how to hand over that hour, and only that hour, are the ones about to look a lot bigger than their headcount suggests.