OpenAI Shows How AI-Native Workflows Turn Agents Into Operating Capability

OpenAI’s Enterprise Signals report says leading AI users connect agents to company workflows. Its September 1, 2026 case study shows how Basis, Clay, and Exa Labs turn repeatable processes into operating capability—and where human judgment still belongs.

The Usage Gap Is Becoming an Operating Gap

According to OpenAI’s Enterprise Signals data, frontier firms—the top 10% of enterprise users—now generate 8.3 times as many output tokens per active user as typical firms, up from a 2.6-times gap in January. The number is not a productivity score, but it signals a difference in depth of use: leading companies are giving AI more context, connecting it to tools, and repeating workflows that prove useful.

The full OpenAI analysis argues that successful work should be measurable and improvable. The shift is from requesting an answer to assigning a bounded process with a clear result.

Basis Turns Onboarding Into a Reusable Skill

Basis, which builds AI agents for accounting firms, uses Codex to make first-day onboarding more repeatable. The company says the process now takes 30 minutes instead of two hours. New employees receive access to Codex and a company-specific onboarding skill—a reusable set of instructions and resources for a defined workflow.

Codex welcomes the employee, explains company concepts, and completes integration setup in the background. HR can update the skill when recurring questions or exceptions appear. Basis first demonstrates the process, then gives it a trigger, known steps, the right tools, and a definition of “done.” It is a teachable workflow, not an autonomous HR department. The Basis site and OpenAI’s Codex guidance provide product context.

Clay Gives Agents Persistent Account Context

Clay’s problem is different. Its go-to-market teams have important deal information scattered across CRM records, email, Slack, calls, presentations, text messages, and conversations with customers. A single prompt cannot keep that context current.

Clay’s approach, described in its company overview and the OpenAI case study, assigns a persistent workspace and subagent to each account. The subagent reviews primary sources overnight. A coordinating agent turns changes into priority moves, such as answering a customer question or filling a gap in the buying committee. Clay says the workflow saves roughly an hour of nightly inbox triage. Recommendations retain evidence so sellers can inspect the source before acting.

Exa Carries Signals Into Tested Action

Exa Labs, which builds web-search infrastructure for AI agents, calls its distribution goal “Exa everywhere.” The team wanted to identify integrations across repositories and the wider developer ecosystem, then move from discovery to implementation without losing the thread between research, engineering, and communication.

Using a defined workflow for Codex, Exa monitors high-priority opportunities, gathers context from sources such as Slack and Notion, creates pull requests, runs tests, and prepares weekly updates. It can draft an announcement when appropriate, but people still decide which opportunities matter and what commitments the company should make. Tests and review points keep the agent’s work visible before anything ships.

What Enterprise Leaders Can Borrow

OpenAI’s examples suggest four practical design rules. Start with a stable process rather than a vague ambition. Give the agent persistent context and only the permissions it needs. Define what completion means, then add tests or evidence that make the result inspectable. Finally, place human review at the point where the work becomes consequential. OpenAI’s enterprise AI strategy and investment guidance make the same distinction between broad experimentation and workflows that can be governed.

This model connects with Business 2.0 coverage of research agents, enterprise AI controls, AI governance, AI literacy, and frontier-model access controls. Advantage will come from workflows where context, tools, permissions, evidence, and review are designed as one system—not from adding an agent to every department.

OpenAI’s admin-plugin announcement and AI adoption coverage provide additional context. The lesson is narrower than “AI will run the company”: repeatable work becomes easier to teach, monitor, and improve when agents operate inside a clearly bounded workflow.