Key takeaways from Sourcegraph's live webinar on driving engineering initiatives across large codebases—covering AI code trends, Deep Search, MCP, Batch Changes, and cross-team visibility.
On April 29, we hosted a live webinar where Graham McBain (Head of DevRel) and Travis Lyons (Field Engineering) walked through how engineering teams are tackling their biggest initiatives across large, complex codebases. From AI-generated code trends to live demos on a 3.5-billion-line codebase, the session was packed with practical workflows.
Here are the key takeaways.
The numbers are hard to ignore: 90% of codebases now include AI-generated code, a milestone reached in 2025, two years ahead of Gartner's predictions. Even more striking, 41-50% of code is now AI-generated, roughly double what it was a year ago.
The implication? The bottleneck in software development has shifted. Writing code is no longer the hard part. Understanding, reviewing, and fixing AI-generated code is where teams are spending their time. That shift changes what tools and workflows matter most.
Travis demoed Sourcegraph's Deep Search on a codebase spanning 12,000 repositories. Using natural language queries, he explored the Spring Framework, navigated cited results across repos, and showed how the AI teaches you search syntax as you go.
The key insight: you don't need to know where something lives or how it's structured. Deep Search surfaces relevant code with citations, so you can trace the reasoning behind every result.
One of the most compelling demos showed a Deep Search thread URL being passed directly to a coding agent via MCP (Model Context Protocol). The agent automatically pulled a GraphQL schema from the search results and generated a new CLI command in minutes, no manual context-gathering required.
For teams already using coding agents, this collapses the "go find the context, then feed it to the agent" loop into a single step. And there's no performance difference between using MCP and the UI; it's purely a workflow preference.
Travis demonstrated replacing deprecated Java primitive wrapper constructors across 534 matches in 44 repositories. The result: roughly 4,000 lines of coordinated changes, managed from a single interface with PRs generated for each repo.
Paired with Code Insights, the team tracked remediation progress over time, turning a one-off fix into a measurable engineering initiative with clear visibility for leadership.
A recurring theme in the Q&A: Sourcegraph isn't just for engineers anymore. Marketing teams use it to understand product capabilities. Engineering managers track migration progress and adoption metrics. The same search and insights tools that power developer workflows are helping cross-functional teams stay informed without needing to read code line by line.
We covered a lot more in the live session, including a preview of what's coming next with agentic migrations and CI/CD feedback loops.
Watch the on-demand recording to see the full demos, Q&A, and roadmap preview.
With Sourcegraph, the code understanding platform for enterprise.
Schedule a demo