Build

Describe the feature. ML.ai writes it and waits for your look.

Hand ML.ai Code a feature description and it reads the repo, writes the change across every file it touches, and stops before anything saves so you can see the real diff first.

Install ML.ai CodeAll use cases

Developers shipping ordinary feature work: a new API route, a settings panel, a data-model field that has to reach the UI. The task is well-defined enough to describe in a sentence or two.

Without it

You write the schema change, then remember the API route, then find the UI three files later, testing each layer separately and hoping nothing was missed in between.

With ML.ai Code

One request describes the feature. ML.ai Code reads the schema, the API, and the UI together, and returns one diff spanning all three, reviewed before anything saves.

Which agent actually does this

general

Carries out the feature end to end

Reads the relevant files, plans the edit across every file the change touches, then writes it as one reviewable diff.

architect

Optional, for a bigger feature first

Read-only. Returns the critical files and an ordered plan before general executes any of it, useful when the shape of the feature is not yet obvious.

How it actually runs

Most of what a developer builds day to day is not a novel algorithm, it is plumbing: a new endpoint, a form, a column threaded through three layers of the stack. ML.ai Code reads the surrounding code first, matches its patterns, and writes the change end to end, then holds it behind a diff instead of assuming the first draft is right.

    1

    Describe the feature in plain language

    One sentence is usually enough: what should exist, and roughly where. general, the default agent, is built for tasks well-defined enough to hand off whole.

    2

    It reads before it writes

    The agent opens the files it needs, follows imports and existing patterns, and plans the change across every file it touches before editing any of them.

    3

    Review the real diff

    Every edit and shell command is held behind an Allow/Always/Deny prompt. Nothing writes to disk until you approve it, so the first thing you see is the actual change, not a description of one.

    4

    Approve, or redirect

    Allow it, always-allow that kind of edit for the session, or say what is wrong and it revises. The loop stays yours to stop at any point.

ML.ai Code

Add a `dueDate` field to tasks: schema, API, and the task card UI.

3 of 3 files read
  • Reading prisma/schema.prisma
  • Reading src/api/tasks/route.ts
  • Reading src/components/TaskCard.tsx
Wants to write 5 lines
Allowed
12 model Task {
13 id String @id @default(cuid())
14+ dueDate DateTime?
15 title String
16 }

The Prisma field, the API route accepting and returning it, and a due-date chip on TaskCard land in one reviewed diff, nothing else in the file touched.

Add a `dueDate` field to tasks: schema, API, and the task card UI.

BuildML.ai Standard
Ready

Where this stops

01

general cannot ask clarifying questions mid-task. Give it a vague request and it guesses, it does not ask.

02

Every edit still waits for your approval. Nothing saves itself, even the ones you would have allowed.

03

For a change big enough that the order of work matters, plan it with architect first instead of handing it straight to general.

Questions worth asking

Try ML.ai Code today, or talk to us about what is next.

Install the editor agent on your own machine, or book a call to talk through your team's workloads.

InstallBook a call