OpenAI Enterprise Signals: Turning AI Workflows into Operating Capability

AI Execution Gap: Frontier Firms vs. Typical Enterprises

Frontier firms—defined as those in the top 10% of AI usage—are now generating 8.3x as many output tokens per active user as typical firms, a significant increase from 2.6x in January. This widening gap indicates a fundamental shift in operating models: leading companies are moving beyond simple AI assistance to full execution by connecting agents to company context and tools, delegating substantive work, and creating repeatable, successful workflows.

Case Studies in Agentic Workflow Integration

Three companies demonstrate how AI agents can be integrated into specific business functions to move from signal to execution.

Basis: Automating Employee Onboarding

Basis uses AI agents to transform the onboarding process for accounting firms, reducing first-day onboarding time from two hours to 30 minutes.

  • Mechanism: The company created a "onboarding skill"—a reusable set of instructions and resources.
  • Execution: An agent (Codex) welcomes new employees, introduces company concepts, and handles integration setup in the background using the employee's computer.
  • Outcome: Onboarding is now consistent and repeatable. HR can update the skill based on recurring questions or exceptions, ensuring the process improves over time without depending on a single person's availability.

Clay: Managing Persistent Deal Context

Clay utilizes dedicated subagents to solve the problem of fragmented deal context scattered across CRMs, email, Slack, and calls.

  • Mechanism: Every account is assigned a persistent workspace and a dedicated subagent.
  • Execution: Subagents review primary sources and update deal folders overnight. A coordinating agent then synthesizes these updates into a prioritized list of daily moves (e.g., answering customer questions or filling gaps in the buying committee) for the GTM engineer.
  • Outcome: This workflow saves approximately one hour of nightly inbox triage and ensures that supporting evidence remains attached to recommendations for human inspection before action.

Exa Labs: Scaling Developer Ecosystem Growth

Exa Labs uses AI agents to automate the path from discovering integration opportunities to implementing them.

  • Mechanism: A defined workflow for Codex that includes clear priorities and access to necessary sources.
  • Execution: Codex monitors for high-priority integration opportunities, gathers context, creates pull requests, runs tests, and prepares weekly updates using data from Slack and Notion. It can also draft initial announcements for human review.
  • Outcome: The workflow reduces handoffs between research, engineering, and communication teams while maintaining human control over which commitments the company makes and how external relationships are managed.

Framework for Scaling AI Workflows

To close the frontier gap, OpenAI suggests a six-step process for enterprise leaders to experiment with and scale consequential workflows:

  1. Choose a Consequential Value Surface: Select an end-to-end workflow that is a strategic priority, repeats often, and has measurable stakes.
  2. Define Outcomes and Metrics: Establish a KPI, baseline, and guardrails. Measure "depth" via completed tasks and connected tools, and "value" via cycle time, quality, cost, revenue, or risk.
  3. Write the Agent's Job Description: Define triggers, required context, tools, permissions, and the point where the agent must stop for human review.
  4. Build the Human System: Include the people closest to the workflow in the design loop and explicitly define decision rights for business outcomes and domain logic.
  5. Make Experimentation Visible: Encourage employees to test new use cases. OpenAI research indicates early-career employees send 13 more messages per week than executives six months after adoption. Successful processes should be packaged as skills, Plugins, or shared workspaces using Chat (for collaboration), Work (for multi-step knowledge work), and Codex (for technical execution).
  6. Carry the Operating Pattern Forward: Apply the successful context, permissions, and evaluation frameworks from one experiment to the next value surface.

Sources