After using Continue for the past six months to assist with our marketing automation scripts and analytics tracking, I recently decided to trial Sourcegraph Cody for a two-week sprint. The shift was prompted by some hiccups in Continue's handling of our CRM integration context.
I set up both tools side-by-side in VS Code to compare them on a few key workflows. Here’s what stood out.
**Primary Use Cases & Context Handling**
* **Continue:** Its context-gathering from open tabs and the codebase was reliable for in-file work. However, when I asked it to adjust a segment of a HubSpot API script that referenced our internal data model, it often failed to pull in the relevant schemas from adjacent files unless I explicitly opened them.
* **Cody:** The link to our code graph via the Sourcegraph index was a game-changer for cross-repository tasks. I could ask a question about our lead scoring calculation that spanned three different services, and it consistently assembled the right context. This felt more holistic for understanding data flow.
**Accuracy & "Hallucination" Rate**
For marketing automation tasks—like writing Python scripts for lead enrichment or modifying Google Analytics 4 event tracking snippets—both were competent. I noticed a subtle difference:
* Continue occasionally suggested outdated methods for our CRM's API.
* Cody seemed more cautious, often pointing out when it needed more information from our specific code, which I preferred. It felt less like guessing.
**Integration & Daily Workflow**
* Continue’s sidebar chat is clean and the slash commands are intuitive. It felt lightweight.
* Cody’s interface is busier, but I grew to appreciate the "explain" and "smell" commands for debugging legacy attribution code. The ability to manually feed it specific files via `@` was crucial for complex content personalization logic blocks.
**Where Each Struggled**
* **Continue:** Struggled with deep, multi-layered questions about our customer data platform (CDP) event structure. Its context window felt limiting.
* **Cody:** Sometimes felt slower in providing initial answers. Its free plan had noticeable rate limiting during intense sessions, which was frustrating.
For now, I'm sticking with Cody for its superior codebase-wide awareness, which is critical for my work in attribution modeling and conversion optimization where logic is spread thin. Continue felt snappier for smaller, tactical edits. I'm curious if others have compared them on similar integration-heavy projects.