AI Coding Assistants — What to Hand Off, What to Review

The idea, in one line

Let AI do the repetitive typing, but give every line it writes a conscious human read before it merges. AI writes; you approve.

Safe to hand off

• Scaffolding a page object (the file that lists a page's buttons and fields).
• Generating fake test data.
• Drafting a parametrized test — the same test run across many inputs.
• Turning a manual test case into a rough script skeleton.

Always review — this is judgment

GitHub's own guidance on AI-generated code says to watch for four things:

• Hallucinated APIs — a function, import, or fixture the AI invented that doesn't actually exist.
• Ignored constraints — your naming rules or conventions quietly skipped.
• Incorrect logic — code that looks right but does the wrong thing.
• Skipped or deleted tests — the most dangerous one in testing: a check removed instead of fixed.

See it work

# AI was asked to "make the flaky checkout test pass". It did this:
def test_checkout_total():
    cart = add_items([("book", 12.00), ("pen", 3.00)])
    total = checkout(cart)
    # RED FLAG: the real check was  assert total == 15.00
    assert total is not None      # weakened to go green - tests nothing

# Quick review checklist for any AI-written test:
#   1. Does the check still pin the real expected value?
#   2. Any skipped or commented-out checks it slipped in?
#   3. A made-up helper, import, or fixture that doesn't exist?
#   4. Does it match our naming and page-object conventions?

Read it top to bottom: the test still runs and still turns green, but it no longer proves the total is correct. That green tick is now lying to you.

Advanced — build trust in layers

You don't have to trust AI all at once. Let it touch tests and refactors first, where mistakes are cheapest. Run its output through your usual automated checks — linting, static analysis, and security scanning. Then give every generated line a human read before it merges. The rule stays simple: AI writes, you approve.

Grounded in GitHub's official docs, 'Review AI-generated code'