OpenAI Guide to Managing AI Investments in the Agentic Era

OpenAI Guide to Managing AI Investments in the Agentic Era

OpenAI recommends that enterprise leaders shift their AI investment strategy from tracking token costs to measuring "useful work per dollar," focusing on tasks completed, time saved, and scalable workflows. This transition is critical as organizations move from simple chat interfaces to longer-running, agentic workflows that require deeper visibility and governance.

Prioritizing Outcome-Based ROI Over Token Pricing

The lowest token price does not always result in the lowest total cost. While OpenAI notes that prices per million tokens fell 97% from GPT-4 to GPT-5.4, and GPT-5.6 further improves efficiency (delivering better performance in the Artificial Analysis Coding Agent Index with 57% less time per task and 54% fewer output tokens), token cost is a poor proxy for value.

To accurately evaluate model efficiency, leaders should:

  • Track cost per accepted outcome: Measure the full cost to reach a "good enough" standard, including model usage, tool calls, retries, latency, and human review.
  • Define task-specific benchmarks: Use evaluations that reflect real-world tasks and edge cases.
  • Optimize the workflow: Use clear instructions, focused tools, and explicit stopping conditions to reduce wasted spend and loops.
  • Match intelligence to complexity: Reserve frontier models for high-stakes or ambiguous work, using smaller, faster models for tasks that meet the quality bar.

Enhancing Visibility into Usage and Spend

Enterprise leaders require a transparent view of AI usage to distinguish between wasteful spending and business-critical workflows. Because agentic workflows in ChatGPT Work can vary widely in resource consumption, admins need to see the specific work driving the spend rather than just the total credits consumed.

OpenAI provides updated usage analytics and spend controls in the Admin Console to track:

  • Workspace levels: Whether adoption and spending are scaling proportionally.
  • Team and user levels: Identifying where demand is growing and who requires additional support.
  • Product and model levels: Determining if expensive intelligence is being used for sustained, high-value demand.

Governing Advanced Workflows for Scalability

Governance should serve as the operating layer that determines which AI workflows are permitted to scale. As teams adopt Computer Use, connectors, and plugins that interact with enterprise systems, centralized control becomes essential.

Key governance strategies include:

  • Centralized Access Control: Using ChatGPT Work to manage approved context, connected tools, permitted actions, and spend limits (including workspace defaults and individual overrides).
  • Deployment Engineering: Utilizing OpenAI’s Deployment Engineers to optimize architecture, reliability, and latency for priority deployments.
  • Privacy and Compliance: Implementing access controls, retention postures, and Zero Data Retention options for high-trust environments.

Portfolio Management and Funding Models

AI investments should be managed as a portfolio, with funding aligned to the maturity of the workflow. OpenAI suggests categorizing investments into three tiers:

  1. Broad Access: For everyday productivity.
  2. Function-Specific Workflows: To improve repeatable work.
  3. Strategic Bets: Built around proprietary company context.

Funding should progress through three stages: Exploration (testing model capability), Validation (testing against a quality bar), and Production (funding integrations, reliability, and change management). To accelerate this process, shared capabilities—such as model routing, reusable agent patterns, and curated knowledge—should be funded centrally.

Matching Capacity to Proven Demand

Once a workflow's value is proven, leaders must match the product and support model to the actual demand. Organizations can start with the foundation provided by ChatGPT Work—which includes capabilities for coding, agentic workflows, and Computer Use—and extend it with proprietary data and logic.

For large-scale strategic deployments, OpenAI Frontier and Deployment Company provide specialized support to help enterprises build and manage "AI coworkers" across their systems, preventing the need for every individual workflow to rebuild its own underlying infrastructure.

Sources