Skip to content
Notifications
Clear all

Anyone using Copilot for production Go services in a Fortune 500?

4 Posts
4 Users
0 Reactions
88 Views
(@cloud_ops_learner)
Honorable Member
Joined: 6 months ago
Posts: 419
Topic starter   [#16516]

Hey everyone. I'm new to managing cloud infrastructure for our team's Go microservices. We're a large enterprise, and there's talk about adopting GitHub Copilot for our production Go codebase.

I'm curious if any other big companies are using it for Go in production. I have some basic concerns:

How does it handle our internal libraries and AWS SDK patterns? Does it suggest secure code by default, especially for IAM or handling secrets? Any issues with it generating inefficient loops or memory leaks in Go?

Just trying to see if it's worth pushing for a pilot program or if we should look at other assistants. Thanks!


Still learning


   
Quote
(@annab)
Reputable Member
Joined: 4 months ago
Posts: 349
 

> Does it suggest secure code by default, especially for IAM or handling secrets?

That's the part that worries me most too. I've seen Copilot generate some pretty convincing looking AWS SDK calls in Go, but when I checked the IAM policy it was suggesting, it basically gave an admin-level wildcard. That was just a personal project though, so I'd be really curious if anyone in a large org has had success training it with their own internal libraries or if they rely on strict linting rules to catch those issues.

I work more on the marketing automation side (HubSpot, email tools) so I'm definitely not a Go expert, but we're looking at CDP integrations that might touch Go microservices. Have you already tried the Copilot pilot in a sandbox with your internal SDK wrappers? I'm wondering if the context window is big enough to learn from a few examples of properly scoped AWS patterns in your codebase.



   
ReplyQuote
(@data_pipeline_ops)
Reputable Member
Joined: 8 months ago
Posts: 176
 

Yeah, the internal library thing is a big question. We've been trying it for our ETL pipelines, which use a ton of custom wrappers around BigQuery and Snowflake SDKs. Copilot usually doesn't get the patterns right unless we've written something very similar in the same file already. It's more useful for boilerplate like struct definitions or simple loops.

The security angle is real. I'd never trust its IAM policy suggestions without review. Our team rule is it's a fancy autocomplete, not a security consultant. We still rely on our static analysis tools and peer review for anything touching secrets or permissions.

Have you considered rolling it out just for unit test generation first? That's a lower-risk way to gauge its fit with your codebase patterns.


PipelinePadawan


   
ReplyQuote
(@helenr)
Honorable Member
Joined: 4 months ago
Posts: 534
 

From my experience in the content space, that's a very practical set of concerns to start with. The internal library challenge is real. Many teams I've spoken to find Copilot becomes more useful after you've established a consistent pattern within a single file or package, but expecting it to infer patterns from across a sprawling private monorepo on its own is optimistic.

On security, I think the key is reframing it as an assistant, not an author. It doesn't "know" your company's security context. The value of a pilot program could be to test exactly that: how often does it suggest something that passes your existing static analysis and review gates, versus how often does it create a new class of problematic suggestions you need to watch for? That data is useful for deciding on a wider rollout.

Have you looked into whether your internal libraries have consistent, well-documented patterns? That can sometimes help the model pick up on things.


—HR


   
ReplyQuote