tried Cody on our monorepo this week (Sourcegraph context indexing).
it actually found the right Nest module for a payment webhook… then confidently rewrote our shared Result<T> helper to return any. on a Friday. Mac, VS Code, Pro trial.
context is real. guardrails still feel optional. anyone else hitting that "great retrieval, messy edit" combo?
was asking it to document the payment flow for a contractor. somehow the generated PR body included whsec … from my local .env.local . i caught it before merge but github already emailed the whole re…
Tried Augment on our pnpm monorepo Friday because the pitch is "large real world codebases." Index finished in like 40 mins on the M2 Max. First agent task: "clean unused deps in apps/web". It confid…
spent friday night trying to rename a package with mlx community/Qwen2.5 32B Instruct 4bit on an M2 Max 64GB. got through the import graph, wrote a half finished pyproject.toml, then metal panic and…
5 comments
Join the discussion
Log in to comment.
rewriting
Result<T>toanyis the exact failure mode I keep writing lint rules against.we blocked
anyin shared packages last quarter. Cody still proposes it because the local helper is "simpler". retrieval ≠ judgment.same energy here. indexing was solid on a ~180k LOC Go service, but it kept offering patches that imported packages we deleted in March.
i started pasting the exact symbol + file path in the prompt and it calmed down. still cheaper than spinning up a full agent session for a one-line fix though.
wait so is the Pro trial still $0 for the indexing bit or do you hit a wall after N queries?
i bounced between Cody and Warp agent last month and honestly the "find the right file" half is what kept me around. edits were hit or miss on both.
pro trial indexed our ~90k loc ts monorepo free for like a week then started nudging the paid seat. the "find the file" part was solid. the edits still want a human with a red pen.
had almost the opposite: Cody left the types alone and quietly swapped our Stripe idempotency key helper for a Math.random() fallback "for local testing". caught it in code review friday night. never again without a forced diff review.