We've solved assessments
for the age of AI.

Use AI. Show your judgment.

Real coding tasks. Decisions, tests, and token use—saved for review.

Contribute a real problem See the assessment approach
00 / THE PREMISE

AI is a multiplier.

Judgment turns
that 0 into 1.

SystemQuestionVerify
Where does judgment come from?
01 / THE EXPERIENCE GAP

Who becomes
the senior
when nobody
gets to be junior?

Experience comes from understanding what broke—and why.

AI can skip that lesson.

NoticeTraceLearn

Finishing isn't the same as learning.

Anthropic · 52 developers
With AI
50%
Without AI
67%
Post-task quiz average. One Python-library learning study.
Blind iteration has a price.
02 / THE COST OF GUESSING

You pay for
every “try again.”

More tokens. More review. Another lesson missed.

PromptPatchRepeat
THE HIDDEN INVOICE+
01

Tokens

Context. Output. Retries.

+
02

Your time

Review. Untangle. Repair.

+
03

Your understanding

The experience you skipped.

+
+19%

Time spent with AI.

One early-2025 study: 16 developers, 246 tasks. 2026 context.

METR study
Compute has a physical cost, too.
03 / THE OTHER BILL

The cloud has
a power bill.

Electricity. Hardware. Cooling.

Spend compute deliberately.

PowerComputeHeat
GLOBAL DATA-CENTRE ELECTRICITY
415TWh
2024 Estimated
945TWh
2030 Projected

All data centres, not AI alone. Global figures—not a session footprint.

IEA · 2025 base case
Make deliberate decisions the assessment.
04 / THE FACTORYZE APPROACH

Let them use AI.
Make judgment
the test.

EditsChecksUsage
01

A real problem.

A bug worth investigating.

02

Visible decisions.

Edits, commands, and approvals.

03

Evidence.

Correctness checks and saved usage.

Explore 2 starter tasks Good problems need your experience.
05 / AN OPEN CALL TO EXPERIENCED DEVELOPERS

You earned
your judgment.
Pass it on.

Turn the bug you'll never forget into a lesson someone else can learn.

Help build the next generation of experienced engineers.

What to send.

  1. 01
    The situation.

    A symptom to investigate.

  2. 02
    Runnable code.

    A small Python starter.

  3. 03
    Useful tests.

    Public examples. Hidden correctness checks.

Task-pack format

Prefer a pull request?

Add a task under tasks/practice/.

Contribute on GitHub
NO PULL REQUEST REQUIRED

Upload a task pack

Upload a ZIP for maintainer review.

Download the editable template
Include task.yaml, workspace/, and tests/hidden/.

Uploads stay private for review.

Email is used for review. Remove secrets and personal data.

Before you
pass it on.

Do I need an API key to contribute?

No API key is needed to contribute. The desktop AI agent needs a model connection.

What happens to my upload?

Private maintainer review. Your code isn't executed or published automatically.

Can I contribute an idea instead?

Open a GitHub issue with the situation and the lesson.

Does Factoryze measure environmental impact?

Token usage only. No electricity, water, or carbon estimates.

Evidence & context
  1. 01

    Anthropic, January 2026. 52 developers learning a Python library. Quiz averages: 50% with AI, 67% without.

  2. 02

    METR, July 2025. Early-2025 tools, 16 developers, 246 tasks. Later-estimate limitations.

  3. 03

    IEA, Energy and AI, 2025. All data centres. 2030 is a base-case projection.