Summary
Stack Overflow’s VP of product, Alex Lotto, joins OpenAI at DevDay 2026 to explain how the company uses Codex across two very different products. Stack Internal is an enterprise knowledge intelligence product for deployments of 150 to 10,000 seats, including large banks and on-prem installations. Stack Overflow for agents, or SOFA, is a small experimental product designed for AI agents as users.
Lotto describes two approaches. For SOFA, the team gave Codex wide latitude within Stack Overflow’s architectural golden paths and asked it to propose decisions and constraints. For the enterprise product, they were far more prescriptive, supplying service contracts, architecture rules, and security requirements, and keeping humans deliberately in the loop. Concerns such as data leakage, authorization, and differing source systems shaped that approach.
A highlight is the concept of archetypes. Codex identified shared patterns across sources like Google Docs, Confluence, Notion, and Teams attachments, so one document API could serve many connectors. Lotto says this saved weeks of work and freed engineers to focus on the harder problem of judging trust and authority, such as deciding which Slack thread is more reliable. The talk ends with building products for a world where agents, not just people, use Stack Overflow.
📺 Source: OpenAI · Published October 07, 2026
🏷️ Format: Interview
