Continue (Mission Control) - Quality control for your software factory
by•
AI agents multiplied code output. Review didn't scale with it. Tests still pass, but conventions erode, security patterns slip, and your codebase starts feeling like it was written by ten different people.
Continue is quality control for your software factory: source-controlled AI checks on every pull request. Describe a standard in plain English, commit it as a markdown file, and it runs as an AI agent on every PR. Catches what you told it to. Passes silently when everything's fine.
Replies
Continue
This is exactly the kind of tooling that's been missing in the AI-assisted development workflow. We use 4 different AI providers at TubeSpark (OpenAI, Anthropic, Groq, Gemini) for content generation, and the quality variance between models is real — what passes review from one provider often needs manual fixes from another.
The idea of encoding quality standards as source-controlled markdown files that run on every PR is brilliant. Right now we rely on manual code review to catch AI-generated inconsistencies, which doesn't scale.
Curious about the feedback loop — when Mission Control flags an issue, does the developer fix it manually or can it suggest/apply fixes automatically?
Continue
@aitubespark Super cool! The multi-provider quality variance is exactly the problem checks solve. You encode "good" once as markdown files in .continue/checks/, and it runs on every Pull Request (PR) regardless of which model wrote the content / code.
On the feedback loop: checks show up as GitHub status checks on the PR. If one fails, you click through to Mission Control where you get more detail and can quickly accept or reject the suggested fix. Once you build trust in a check, you just flip it to auto-fix. At that point the AI catches the issue, fixes it, and pushes, all before you look at the PR.
How do you know when to flip that switch? We wrote about this in Intervention Rates Are the New Build Times. Measure how often you have to correct the AI per check, and as that drops toward zero, that's your signal to let it run autonomously. This is easy to do on the Metrics page in Mission Control.
For your four-provider setup, I'd start by writing checks for the specific patterns where you see quality variance between models. The stuff reviewers keep catching. That becomes your quality floor every PR has to clear.
Continue
It's been very surprising how quickly we're able to ship code once checks are in place. For the last few years there has been (rightfully so) a dominating conversation on scaling the writing of code, but with that problem feeling largely solved, it's become, for us, a question of how to make sure that all code meets our standards. Checks in Mission Control have made this pretty easy to do, especially given we can just ask Claude to write checks for us. With checks doing the heavy lifting, most PRs probably don't take more than a few seconds of human review
Weave
let's go, amazing product in the wild
Continue
@brennan_lupyrypa Thanks Brennan, much appreciated!
The markdown-file-as-check approach is smart because it keeps standards reviewable and diffs visible, same as any other code change. One thing I'd want to know: how do you handle checks that are too broad and start flagging everything? That noise problem killed a couple internal lint rules on our team before we got the scope right.
Continue
@abayb Yeah this is definitely a key observation. We built into the system an easy way to toggle the "Sensitivity" of suggestions, and whenever you reject you can provide feedback. Between the two of these things, and also just updating the check files on occasion, it's not hard to keep down noise. And even if there is a high volume of checks, you can just ask your coding agent to address them, it will do a good job!
Very much needed, as I'm always weary of pushing code written by AI.
Continue
@abhinavramesh As one should be! If you try it out would love to hear how it goes : )
Continue
My favorite check on the Continue team is our "Next.js best practices" based on this skill. It runs on every PR and catches something subtle almost every time!
Even though we have this same skill for the agent to use locally, agents still make mistakes as context windows grow, so running it in CI gives us assurance that we aren't letting slop make it into production.
Love that this stops our code from becoming a mess before it even hits the repo. Can we "refactor'' the price a little bit for a new team? A cheeky discount would make this a total no-brainer for us today. What do you say? ;-))
Continue
@john_oliver11 We are happy to give free trials for small startups. Want to ping support@continue.dev and we'll figure something out?