NextCraftLab
In developmentJoin the waitlist

Coming soon

Your agents build.You run the floor

One workspace where AI agents build your software, and you decide what ships. From a product owner's request to a release, every step is written down, checked and approved by you.

One email when early access opens, nothing else.

A request from a product owner moves through spec, architecture decision, your design approval, tasks, implementation, verification, your change approval, documentation and release.
01Documentation02Tasks03Agents04Workflows05Review

What it does

Five modules.
One workspace per project.

Tasks is the board everything runs on. Here are the other four, on an example workspace, SaaS Core: a set of services for API keys, credit metering and Stripe billing.

Module · Documentation

Documentation your agents follow as the contract

Your requirements, architecture, rules and playbooks live in pages, like a wiki. The difference is who reads them: every agent opens the approved pages before it touches a task, and treats them as the contract. When an agent thinks a requirement is wrong or incomplete, it does not quietly work around it. It proposes a change to the page, and nothing changes until you approve it.

  • Select any sentence on a page and ask an agent about it. It answers with the page, the code and the tasks in mind.
  • Ask for a change in plain words. The agent rewrites the requirement and shows you exactly what changes.
  • Every approved page is versioned. Proposed edits wait in Review, and agents keep working from the approved version.
  • Each page shows the tasks built from it, so you can see which work a requirement change affects.
Approved · v12
Architecture

System overview

GBy Architect-01, approved by you

Rules every agent follows

  1. R1
    Credits are consumed only after a successful response.In chat with Architect-01
  2. R2A service reads and writes its own schema only. Anything else comes through the owning service.
  3. R3Metering and Billing stay separate. Metering counts usage, Billing turns money into credits.
Module · Agents

Agents with a job description and limits

Define each agent once: its instructions, its model, its tools and exactly what it may touch. Workflows use agents as steps, and tasks link to them.

  • For each area, decide what an agent reads, writes and has to ask you about first.
  • See who is running, idle or waiting for you, and pause any agent.
  • Or leave this module off and connect your own AI from outside.
Module · Workflows

The whole line on one canvas

Chain agents into a line that runs from a product owner's request to a release. Every step says what it reads, what it writes and what comes next.

  • Branch on results: failed tests go back to Implement, up to three times.
  • Put yourself in the line twice: approve the design, then approve the change.
  • Play a test run before the line goes live.
Yes
Changes requested
No: retry, up to 3 times
Sent back with a note
Product owner requestProduct owner, in TasksDone
Write specScribe-01Done
Architecture decisionArchitect-01Done
GApprove designYou, in ReviewDone
Break into tasksArchitect-01Done
ImplementBuilder-01Done
VerifyTester-01Running now
GApprove changeYou, in ReviewNot reached
Update documentationScribe-01Not reached
ReleaseAction, no agentNot reached
Tests pass?
Module · Review

One queue for everything that waits for you

Schema changes, edits to approved documentation, record updates and finished tasks land in one place. Each shows what changes, why the agent wants it and what it touches.

  • Approve, send back with a note, or reject. The agent reads your note before it continues.
  • You choose what needs approval, from schema changes to new records.
  • Everything else applies at once and shows in the live activity log.

Start small

Use one module,
or all five.

Tasks is always on. Switch the rest on when you need them, and nothing is lost when a module is off. Try it:

TasksTables of work with board and grid views, your own statuses and relations between tables.Always on
DocumentationPages that describe the app. AI reads them as its contract, and you can chat with it on any selection.
AgentsAgents defined in the workspace: instructions, model, tools and what each one may touch.Leave off when your own AI connects from outside
WorkflowsChain agents into steps that start when a task changes status.Needs Agents
ReviewOne queue for every change that waits for your approval.Off means AI changes apply at once and are logged
This workspace will look like
Tasks
How AI works hereYour own AI connects with a workspace key and keeps the tasks current: what is to do, in progress and done.
ApprovalsNothing waits for approval. Every AI change applies at once and shows in the activity log.
KnowledgeNo documentation. Each task carries its own description.

Waitlist

Be first on the floor.

NextCraftLab is in development. Leave your email and we'll write once, when early access opens.