Describe the feature. Get code that’s ready to ship.

You decide what to build. Managed AI agents handle architecture, UX and code, expert humans review every step, and your dev team merges a clean pull request.

See how it works
  • Agents write all the code and design every screen
  • Donux designers and developers review it
  • You only make product decisions

Improve trial activation

  1. Shaping
  2. Plan
  3. Build
  4. Preview
  5. Ship

  1. Too many trial teams drop off before they invite a teammate. Can we fix activation without rebuilding onboarding?

  2. One product question before I plan it. Which should the first version focus on?

    Your turn: pick one

  3. Invite teammate nudges

  4. Plan reviewed by a Donux designer

  5. Agents built it on a preview branch

  6. Preview checked by a Donux developer

  7. Live preview ready

    Invite teammate nudges

    Implement it Pull request #128 opened in your repo

    €15, billed after delivery

“Vibe coding will break your company.”

Forbes, April 2026

Vibe coding will break your company. Magic Team won’t.

Three warning shots Forbes collected in April 2026.

  1. Replit2025

    Production database wiped in seconds.

    During an explicit code freeze, an AI agent deleted the live database. Records of 1,200+ executives and 1,100+ companies impacted.

  2. Klarna2025

    Replaced humans. Then re-hired them.

    After publicly touting AI doing the work of hundreds of customer-service agents, Klarna started hiring humans again.

  3. Air Canada2024

    In court, the AI speaks for you.

    A tribunal ruled the airline was responsible for inaccurate guidance given by its own chatbot. The company paid.

Magic Team puts judgment back in the loop

Clear decision rights
Every deliverable lands as a PR. You decide the merge.
Veto culture
Live preview before payment. Reject without obligation.
Real context
Experts read your analytics, codebase and brand first.
System learning
Your rejections sharpen future outputs.
Ethical discernment
Humans in the loop before the code, not after.
See who reviews every step

You make the product calls. We make everything else.

Magic Team asks about your product and your users, never about frameworks, layouts or pixels. That line is the whole point: no agent ping-pong, no design reviews to run, no technical calls to make.

Your calls

  • Which problem is worth solving
  • Who it’s for and what success looks like
  • Which recommendations get implemented
  • When your team merges and ships

Magic Team’s calls

  • Architecture and technical approach Done by agents
  • UX flows, interface and visual design Done by agents
  • Code, tests and preview deploys Done by agents
  • Plan review and QA before you see anything Done by a Donux expert

Start from wherever you are

Attach your repo, or don’t. Agents deploy every result as a live preview you can click through before anything gets merged.

New product Start from zero. Agents scaffold, build and deploy an MVP, reviewed by expert humans at every step.
  • Start with an idea, a brief, or just a conversation
  • Agents scaffold the project structure, auth, database and core UI
  • Expert humans review architecture decisions at every step
  • You test a working preview before any code lands in your repo
Improve product Agents read your code, analytics and roadmap to surface high-impact improvements. Then build the ones you pick.
  • Connect GA4, Mixpanel, PostHog or Amplitude so agents can read funnels, cohorts and events
  • Each proposal comes with target metric, expected uplift and confidence
  • Expert designers check that the feature actually solves the problem
  • Implement only what makes sense and skip the rest
Bug review Agents scan your codebase for bugs, security issues and performance bottlenecks. You get fix PRs, not reports.
  • Point agents at a branch, a PR, or your entire codebase
  • Fixes come with clear explanations, not just a list of problems
  • Expert developers review every fix before it reaches you
  • Choose which fixes move to implementation

From scan to your code, fast.

Improvements and bug fixes arrive within days, a new MVP within the week. Two experts sit between the agents and you on every task.

  1. 1 Scan

    We connect to your code

    Your repo, your product vision, your goals. Agents scan the codebase while our experts work out what you actually need.

    • Agents
    • Designer
  2. 2 Plan

    We ask the right questions

    Our experts review the agent’s plan, challenge assumptions and shape the scope. You answer product questions only.

    • Designer
    • You
  3. 3 Build

    Agents build, you preview

    Agents build a live, working preview. A senior developer checks it first, then you click around, test edge cases and show your team.

    • Agents
    • Developer
  4. 4 Ship

    You decide

    Like it? Implement it and we deliver a clean PR to your repo for your team to merge. Not convinced? Skip it, no obligation.

    • You

Built like a sharp product partner

The product sense lives in the software too, which is why you never have to drop down into technical detail.

Asks only the questions that matter
A scoped interview, not a wall of questions. Sensible defaults cover the rest.
Learns your preferences
Past calls and rejections shape what it proposes next, so it asks less over time.
Keeps a decision record
Every product and technical decision is logged, so you can always see why something was built that way.

Not AI guessing. AI reading your analytics.

Connect GA4, Mixpanel, PostHog or Amplitude once. Every proposal then shows where it moves the numbers, so you know where to focus.

Metric, uplift, confidence, source
Every proposal comes with the four numbers you need to decide. No backlog grooming.
Your stack, your source of truth
No new dashboard to learn. Agents read the analytics provider you already use.
Connected in 30 seconds
An API key plus an optional project ID. Read-only, encrypted, masked in the UI.

Improvement proposal

Draft 42

Simplify checkout step 2

Cut the form from 8 fields to 3 and bring address autocomplete above the fold. Removes friction on the step where most buyers drop.

Expected uplift
+13%
Confidence
94%
Target metric
Checkout conversion

Funnel analysis from GA4, last 30 days

  • GA4
  • Mixpanel
  • PostHog
  • Amplitude

Every model. Zero lock-in.

Magic Team routes each task to the model that does it best. You pay for results, not API tokens or seats. We absorb the infrastructure cost.

Routed to the best model for each task

  • OpenAI
  • Claude
  • Gemini
  • Grok
  • Qwen
  • Llama
  • DeepSeek

And more. Model access is included during the beta.

Pay for value, not infrastructure

Seats, overages and infrastructure add up fast. Magic Team absorbs all of it.

Monthly cost of common AI subscriptions compared with Magic Team
Claude Max $200/user/mo
ChatGPT Pro $200/user/mo
Google AI Ultra $249.99/mo
API extra usage $$$
Magic Team All included

Works with
your tech stack

Click it before you commit.

Magic Team is free during the beta. Every result is a working preview first. Choose what to implement, and pay only once the implementation is delivered.

Analysis, plan and preview

Free during beta

  • Product and codebase analysis
  • Findings and recommendations
  • Shaping and planning
  • Expert human review
  • A live preview with real UI and seeded data

Implementation

€15 each, billed after delivery

  • A clean PR in your repository
  • Code that follows your conventions
  • Full IP ownership, forever

Beta price, charged only after an implementation is delivered. No subscription, no minimum commitment. Post-beta pricing isn’t final yet.

Raw AI tools or a managed team?

Claude Code, Codex, Lovable and AI design tools are great for fast prototypes and direction. Magic Team can take those outputs as input, then build the production-ready app without asking you to make technical, UX or UI calls.

See all comparisons

Questions

Something we didn’t cover? Write to hello@magic-team.com and a person will answer.

Who's behind Magic Team?

Donux, a product studio specialized in B2B SaaS. The expert humans who supervise every plan and review every PR are the same senior designers and developers who have shipped 80+ products in more than six years for clients like Fluida, Up2You and Balanced. Magic Team is also available as a full strategy, design and software development engagement at donux.com/services/software-development.

How does beta pricing work?

Magic Team access is free during beta. Analysis, findings, suggestions, shaping and planning are included. Each implementation costs €15, a beta-only price, and you pay only after the implementation is delivered.

How much will implementations cost after the beta?

Post-beta pricing is not final yet. As an indication, expect each implementation in the €500–1,000 range depending on scope. During the beta, every implementation is €15.

Do I have to implement every recommendation?

No. Review the findings and suggestions, then choose which improvements you want Magic Team to implement.

What's an iteration?

Magic Team uses your feedback to refine a recommendation and its plan before you choose whether to implement it.

How fast do I get results?

Improvements and bug fixes arrive within days, a new MVP within the week. Once a result is ready, you click around the live preview before deciding to confirm the build.

What does my dev team need to do?

They handle the last 5%: wiring in auth, permissions, feature flags and monitoring. Like reviewing a PR from a new team member.

Do I own the code?

Yes. Once a result lands as a PR in your repo, full IP transfers to you. A clean PR that matches your conventions.

What if I don't like a result?

Skip it. No obligation. Rejections improve future results through Magic Team's learning model.

Is the code AI-generated?

Yes. MagicFabbro, our AI agent, writes 100% of the code. Every task also passes through two expert humans before it reaches your repo: a senior designer reviews the plan, and a senior developer reviews the working preview. By the time a PR lands in your codebase, a human has verified it builds, behaves correctly and is safe to merge. Your team's review is the final sign-off on code that has already been checked end to end.

What AI models does Magic Team use?

All the leading models, including Claude, GPT, Gemini, Llama, Mistral and more. Each task is routed to the best model. Model access is included during the free beta.

How does Magic Team use my analytics data?

Read-only. Magic Team connects to GA4, Mixpanel, PostHog or Amplitude to read funnels, cohorts and events. It never writes back to your analytics, and the insights feed the rationale shown on every proposal.

Is my analytics connection secure?

Yes. The API key is encrypted and masked in the UI (for example mp_live_••••••1234). Disconnect the provider and all insights are removed from existing proposals, with no orphaned data left behind.

Can I switch analytics providers?

Yes. Moving from, say, Mixpanel to GA4 automatically re-aligns all existing proposals to the new source. No manual cleanup needed.

Can I build a product from scratch?

Yes. Connect an empty repo (or none at all) and Magic Team's agents scaffold, build and deploy an MVP. Expert humans review every step.

What kind of payments do you accept?

All major cards and SEPA payments.

Is there a minimum commitment?

No. You choose which improvements move to implementation.

Start your first scan.

Free during beta. No contracts, no lock-in, results in days.

hello@magic-team.com