What Sourcegraph shipped
Sourcegraph introduced Agentic Batch Changes, a capability that translates natural language prompts into organization-wide code updates. By combining Deep Search and Code Finder, the system allows developers to roll out complex refactors—like standardizing Java build tools—across disparate repositories without manual intervention.
The feature addresses the friction of scaling repository changes across large engineering organizations. By automating the scoping, canary testing, and pull request generation, it shifts the developer's role from manual execution to high-level orchestration.
How the motion works
When demonstrating a multi-step AI workflow—like scoping a change, running a canary, and scaling the rollout—pacing is critical. Editors should rely on the hard-cut to skip CI loading times and wait states, keeping the focus strictly on the agent's decision-making process. Accelerating the idle moments ensures the viewer's attention remains on the core value proposition rather than the reality of compilation times.
Screen recordings of dense developer interfaces often overwhelm the viewer. To counter UI clutter, applying a punch-in isolates specific actions, such as the initial natural language prompt entry or the executor logs when a preview fails. Smooth, eased camera moves guide the eye exactly where the agent is operating, preventing the audience from getting lost in the peripheral noise of the dashboard.
Highlighting the AI's autonomous actions requires clear visual cues. Applying a typewriter effect to the natural language input can emphasize the simplicity of the user's task, while subtle UI highlights can track the agent as it reads logs and fixes the batch spec. This visual contrast between human input and machine execution clarifies the division of labor.
What to steal from it
- Use aggressive punch-ins to focus attention when demonstrating complex, multi-pane developer interfaces.
- Emphasize the simplicity of natural language inputs with a typewriter effect to contrast against the complex backend execution.
- Edit out loading states and CI wait times with hard cuts to maintain momentum in technical walkthroughs.
- Visually separate the user's initial prompt from the agent's autonomous troubleshooting to highlight the product's intelligence.
Want a video like this for your product?
Impractical cuts one from a single prompt — the same motion craft, in about twenty minutes.
