▶ Watch ↗AI Engineer World's Fair 202612:23
Conference affiliation: Sourcegraph
Dan Adler leads Sourcegraph as CEO, an appointment announced in December 2025 after years spent helping build its business and data infrastructure. His work centers on codebase ownership: keeping an organization’s software understandable, consistent, and maintainable as coding agents accelerate changes across thousands of repositories.
Adler’s career spans software, investing, and business operations. Before Sourcegraph, he worked at Bain & Company and Vector Capital and founded Cells Software. His startup pursued spreadsheet-based data labeling; he was shutting it down shortly before joining Sourcegraph in fall 2016.
The experience of searching code and following definitions and references directly in a browser drew him toward Sourcegraph. He joined before its first customer, sold early contracts, delivered demos, supported customers, and built the company’s first dashboards. He also wrote its original data infrastructure and contributed to its production codebase. His responsibilities combined hands-on software work with establishing the business around it.
Adler subsequently served as director of business operations and strategy and VP of business, establishing functions spanning finance, analytics, strategy, and IT. He was CFO before becoming CEO during Sourcegraph’s separation from Amp. Sourcegraph co-founders Quinn Slack and Beyang Liu founded the independent Amp company and remained on Sourcegraph’s board. The separation gave the businesses distinct directions: enterprise code understanding at Sourcegraph and frontier coding agents at Amp.
Adler focuses on the people responsible for systems that accumulate decades of business logic, dependencies, and architectural decisions. He argues that AI-generated changes can compound duplicated functionality, inconsistent standards, and brittle dependencies. Faster implementation increases the burden of understanding how the whole system fits together.
His argument for maintaining long-lived software makes that problem concrete. An insurer’s reimbursement rules may depend on a deeply layered COBOL decision tree and thousands of supporting tables. Modernizing the system requires understanding its behavior and dependencies; generating replacement code alone does not resolve that work.
For coding agents, that concern becomes a problem of organization-wide code visibility. Agents use search to understand software, but their conclusions are limited by the code they can access. A tool that can change one repository still needs broader visibility to address a request spanning thousands. Adler treats code understanding as infrastructure for both people and agents, with context becoming a bottleneck at enterprise scale.
Several connected ideas explain how Adler wants that visibility to translate into useful work:
▶ Watch ↗AI Engineer World's Fair 202612:23
Loading more talks and highlights…