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

TL;DR: OpenAI Solutions Engineer Conor Spicer introduced Codeex, an AI coding agent that automates the entire software development lifecycle, dramatically improving code delivery speed and security review capabilities. A demonstration showcased an end-to-end workflow from requirements to deployment. ## Codeex: Beyond Autocomplete – An AI Coding Agent Conor Spicer began by introducing Codeex—OpenAI’s AI coding agent. Its capabilities go far beyond code autocompletion; it can fully automate the software development cycle and many of the different tasks developers perform. Users can assign Codeex a task, have it examine the entire codebase, and then let it run for hours or even a full day until the work is complete. In February, OpenAI launched the Codeex desktop application, allowing anyone to use natural language to start generating code and completing various tasks. Spicer believes this Codex application (note: the original text says "Codeex," but the video likely refers to the same product) has already become the ideal interface for technical and semi-technical teams in the financial services sector. Codeex’s adoption rate has broken nearly all of OpenAI’s chart records. The app had over a million downloads in its first week, and growth has continued, with weekly active users now exceeding 4 million. ### Internal Impact at OpenAI Within OpenAI, engineers use Codeex by default. Spicer stated: “We now ship more code in a week than we used to in a month the year before.” The number of PRs (code blocks completed and submitted) per engineer has increased by 50%. Teams have dramatically boosted code output and product delivery without adding proportional headcount. He emphasized that Codeex has not replaced engineers at all—it has transformed their workflows, making them far more efficient. ## Use Cases in Finance Spicer then discussed Codeex’s potential in finance: refactoring and migrating legacy Cobol systems, automating regulatory reporting, creating auditable documentation, or rapidly prototyping loan, trading, or payment products and services. For the demo, he adopted the “Blossom Bank” scenario. Blossom is a successful bank with strong consumer banking products, but faces threats from competitors and pressure to quickly deliver new features customers want. Customers have repeatedly called for “predictive budgeting”—not just being told how much they spent, but having tools that help predict future expenses and plan ahead. Such tasks previously required a long time and extensive coordination across multiple teams, so they turned to Codeex. ## Demo: End-to-End Workflow from Requirements to Delivery Spicer entered the Codeex application (powered by the GPT5 model) and demonstrated how to transform a historical review feature into a predictive forecasting tool. ### 1. Context Retrieval and Toolchain Integration Although Spicer was eager to start coding, real work often involves urgent meetings, incidents, or retrieving other data. Codeex can examine existing context about the codebase, documentation, and other tools to help understand and quickly extract what’s needed. It searches across different observability tools and the codebase, pulling together a summary for Spicer. Previously, such tasks required coordinating across multiple teams; now they can be run instantly, even to handle topics that come up during meetings in real time. ### 2. Automated Templates and Custom Workflows Beyond ad-hoc requests, teams can extend these best practices into automated flows. Codeex provides some template solutions to help teams get started quickly. Users can also create their own automations, such as a weekly engineering summary—understanding what the team built, delivered, and any blockers—so they are prepared at the start of the week. ### 3. Building a New Feature Spicer returned to the main code-building thread. Suppose the feature’s context is scattered across many places: management-approved specs in SharePoint, design and product documents in Jira, Notion, or even email. Without switching between multiple apps, he can simply ask Codeex to find the relevant information using its connectors. Once started, Codeex understands the request and devises a plan for implementing the feature. It searches SharePoint for referenced files, examines the codebase, and immediately creates a plan. Spicer can open the plan within the application to verify it aligns with expectations, then let Codeex begin implementation. During implementation, Codeex also runs tests to ensure the code meets standards. ### 4. Observing the Agent Building Code Spicer emphasized that this is where engineering has fundamentally changed: “I’m no longer typing code myself; I’m driving it with prompts.” He watches the agent build the code and can intervene mid-process to guide it if needed. Once complete, he can inspect the actual code written, build understanding, and see a summary of how everything works. Switching back to the application, the new feature is already live, ready for immediate feedback and sharing with others. Finally, Spicer commits and pushes all new code, entering the review stage. ### 5. Handling Regulation and Legacy Systems Spicer acknowledged real-world challenges like dealing with regulators and friction from legacy systems. But Codeex can help there too. He demonstrated a very old-fashioned portal (likely without an API) that requires submitting summaries of all consumer-facing application changes to regulators. Codeex used its skills and browser automation tools to handle the task completely automatically: it understood the form requirements, searched the codebase for relevant information, documentation, and evidence, then fully filled out that portal. It compiled a change summary and shared it with Spicer for review, but did not submit directly, ensuring the user remained informed. Tasks that used to take hours now take minutes or less. ### 6. Code Review and Security Spicer also showed how Codeex can be embedded in the review process on GitHub. Automated test suites had already run, and human reviewers had signed off. But Codeex identified a cybersecurity issue—a potential mishandling of sensitive fields that the human reviewers missed. After catching the problem, Codeex could even be asked to start implementing a fix. ## Key Takeaways Spicer summarized: - Codeex dramatically accelerates development workflows and other tasks developers perform. - Codeex not only delivers more code but also ensures code safety through production-context code reviews. - The combination of speed and security generates enormous excitement—this is what we’re seeing with Codeex today. ## Team Support and Customer Success Spicer noted that organizations can feel pressured when faced with all this new code and new tools to learn. He and his team focus on supporting and advising customer engineering teams, helping them set up these new processes so that when they scale their code output, they can truly seize the opportunity and deliver a comprehensive win for their customers. --- Source: Win through AI powered products: Conor Spicer, Solutions Engineer, OpenAI (https://www.youtube.com/watch?v=re-18gil_ec)