Documentation

Where Coroid fits

How Coroid differs from autocomplete, IDE assistants and prototyping tools — and when you should reach for something else.

"AI coding tool" covers several genuinely different products. They are not competing for the same job, and picking the wrong category is a more expensive mistake than picking the wrong product within one.

Four categories

CategoryExamplesDesigned for
AutocompleteGitHub CopilotFinishing the line you are typing
IDE assistantsClaude Code, CursorA developer working on a task, attended
Prototyping toolsLovable, v0, Bolt, Replit AgentTurning an idea into something you can look at
Production deliveryCoroidChanging an existing codebase, under review

The first two put a model next to a developer. The third builds something new, fast. Coroid does neither: it takes a described outcome and delivers a reviewed change to a codebase you already run.

Prototyping and production are different problems

This is the distinction worth being clear about, because prototyping tools are very good and it is easy to assume the same approach scales to production work.

Prototyping optimises for speed to something visible. You start from nothing, there is no existing architecture to respect, no test suite to keep passing, no conventions to match, and nobody depends on the result yet. Constraints would only slow you down — and correctly, those tools impose very few.

Production work is almost entirely constraints. The code already exists and has decisions baked into it. Other things depend on it. There are tests that must keep passing, conventions that should be matched, security and compliance requirements, and a review process that has to be satisfied before anything ships.

Coroid is built around those constraints rather than around avoiding them:

PrototypingCoroid
Starting pointA blank slateYour existing repository
DestinationA running demoA pull request on a branch
VerificationDoes it look rightYour test suite, quality gates, independent QA
GovernanceMinimal by designPolicies, review rules, audit trail
ReviewYou look at itYour engineers review a diff, under your branch protection
SuccessThe idea is worth buildingThe change is safe to merge

They compose well

This is not an either/or. The two fit together in an obvious way:

Prototype to decide what to build. Validating an idea, exploring a UI, getting something in front of a stakeholder this week — a prototyping tool will beat Coroid at all of these, and you should use one.

Then build it where you will maintain it. Once the idea is settled, the work moves into the codebase you actually run, with your conventions, your tests and your review process. That is Coroid's job.

The prototype answers should we build this. Coroid answers build it properly in the system we already have.

When not to use Coroid

Being straight about this is more useful than a feature list:

  • You are validating an idea, not shipping it. Use a prototyping tool. Coroid produces a reviewed change to a real codebase, which is unnecessary overhead for something you may throw away tomorrow.
  • You want to see something visual in minutes. Coroid produces pull requests, not live demos.
  • You want a pair programmer. If you want to stay in your editor and work alongside a model line by line, that is an IDE assistant. Coroid operates on whole units of work while you do something else.
  • You have no repository yet. Coroid needs somewhere to work — see Connect your code.

What follows from the production focus

Several things in Coroid look like overhead until you see them as production requirements:

  • Work is specified before it is built, so there is something to verify against.
  • Verification is performed by an agent that did not write the code.
  • Nothing reaches your default branch without a pull request.
  • Policies and gates are enforced in the execution path, not suggested.
  • Every execution is recorded — cost, model, steps and evidence.

None of that is worth paying for on a throwaway prototype. All of it is worth paying for on a system your business runs on.