I've seen a dozen demos where vendors claim their AI can accelerate onboarding. Most are glorified chatbots. I tested Amazon Q Developer's "codebase tour" feature against a real scenario: getting a new backend engineer up to speed on our legacy service repository.
The premise is solid. Instead of a senior engineer spending days writing documentation, Q can generate a guided walkthrough. But does it actually work, or does it just hallucinate file paths?
Here's my blunt assessment after running it on a ~200k LOC monorepo.
**What it got right:**
* It correctly identified the three main entry points for our service (API routes, message queue consumers, scheduled jobs).
* It generated a coherent narrative, grouping related files by functional area (authentication, data models, core business logic).
* The tour included specific file paths and could explain the purpose of key functions when prompted inline.
**Where it fell short (and this is critical):**
* It missed crucial configuration files. The new hire would have been blocked because the tour didn't mention the environment schema or the dependency injection setup.
* Explanations were shallow. It described *what* a function was named, but not *why* the business logic was structured that way. Historical context was absent.
* You can't just run a tour and be done. A senior dev still needs to vet and edit it.
**My workflow for making this usable:**
1. Generate the initial tour from the root of the repo:
```bash
# This creates a .q/tours/ directory with a markdown file
q developer generate-tour --name "backend-onboarding" --paths "src/"
```
2. Open the generated markdown file in your editor. The structure is there, but you must:
* **Add critical missing files** (e.g., `config/schema.ts`, `infra/`).
* **Insert commentary** on historical decisions or pitfalls.
* **Prune irrelevant sections** Q might have included.
3. Share the final, curated `.q/tours/backend-onboarding.md` file with the new hire. They can use the Q Chat pane in their IDE to ask questions about any line or file referenced in the tour.
**Bottom line:** It cuts the initial documentation drafting time from 4 hours to about 30 minutes of editing, but it's not a zero-effort solution. The value is in the scaffold it provides. Without a human expert to fill in the gaps and correct omissions, the new hire will hit dead ends. Treat it as a junior tech writer that needs heavy supervision.
Show me the query.