Win through AI powered products: Conor Spicer, Solutions Engineer, OpenAI

Key Concepts

• Codeex: An AI-powered coding agent designed to automate software development lifecycles, moving beyond simple autocomplete to autonomous task execution.
• GPT-5: The underlying large language model powering the Codeex application.
• Agentic Workflow: A methodology where AI acts as an autonomous agent that can plan, search across documentation/codebases, execute tasks, and interact with external tools.
• Browser Automation: A feature allowing the AI to interact with legacy web portals that lack APIs to perform data entry and regulatory reporting.
• Contextual Code Review: An automated security layer that scans code for vulnerabilities (e.g., mishandling sensitive fields) that human reviewers might overlook.

---

1. Main Topics and Key Points

• Productivity Gains: Codeex has achieved over 4 million weekly active users. Internal usage at OpenAI has resulted in a 50% increase in Pull Requests (PRs) per engineer, allowing for higher output without increasing headcount.
• Workflow Transformation: The tool shifts the developer's role from manual coding to "prompting and steering," where the AI handles the implementation while the human validates the plan and output.
• Cross-Platform Integration: Codeex connects to disparate enterprise tools—including SharePoint, Jira, Notion, and email—to synthesize project requirements and documentation into actionable development plans.

2. Real-World Application: Blossom Bank

• Scenario: Blossom Bank needed to implement a "predictive budgeting" feature to remain competitive.
• Challenge: Historically, this would require significant cross-team coordination and manual effort.
• Solution: Using Codeex, the developer pulled requirements from SharePoint, generated a development plan, implemented the feature across front-end and back-end services, and automated the regulatory reporting process.

3. Step-by-Step Methodology

1. Context Gathering: The agent scans internal documentation (SharePoint, Jira, etc.) to understand the scope.
2. Planning: The agent proposes a technical implementation plan for the user to review and validate.
3. Execution: The agent writes the code across the codebase. The user can steer the agent mid-process if requirements change.
4. Validation & Testing: The agent runs automated tests to ensure the code meets organizational standards.
5. Regulatory Compliance: Using browser automation, the agent populates legacy regulatory portals with the necessary evidence and documentation.
6. Security Review: The agent performs an automated security scan on the final commit to identify vulnerabilities before deployment.

4. Key Arguments and Perspectives

• Augmentation vs. Replacement: Connor Spicer argues that AI does not replace engineers; rather, it makes them "dramatically more effective" by removing the friction of manual, repetitive tasks.
• Speed and Safety: The core value proposition is the combination of rapid development (shipping more code) and enhanced security (automated, contextual reviews).
• Scalability: By creating "template solutions" and custom automations, organizations can standardize best practices across teams, reducing reliance on individual developer decisions.

5. Notable Quotes

• "We now ship more in a week than we were doing in a month previously." — Connor Spicer, regarding the impact of Codeex on internal OpenAI workflows.
• "Instead of typing the code out myself, I'm prompting this. I'm inspecting how the agent is building it." — Describing the shift in the modern developer's workflow.

6. Synthesis and Conclusion

The presentation highlights a paradigm shift in software engineering where AI agents serve as force multipliers. By integrating deep context from enterprise tools, automating legacy system interactions, and providing an automated security layer, Codeex enables organizations to accelerate their development cycles significantly. The ultimate takeaway is that the future of development lies in "scaffolding" these AI-driven processes to handle the increased volume of code while maintaining high standards of safety and regulatory compliance.