What I Actually Use Claude For (and What I Don't) as an SDET

After integrating Claude into my Python + Pytest workflow, here's the breakdown of where it genuinely saves me hours and where I still trust my own instincts.

A working SDET's take on using Claude for PR reviews, test generation, and repetitive pytest tasks.

What Claude Does (in Plain English)

Claude is a conversational AI assistant made by Anthropic. Unlike a code completion tool that finishes your line, Claude reasons about what you give it, you can paste a function, a PR diff, or a failing test and have an actual back-and-forth about it. For an SDET, that distinction matters.

  • PR code review I paste the changed code and ask Claude to flag anything that looks hard to test, likely to break, or missing error handling. It catches things I skim past after the fifth PR of the day, null returns with no guard, untested branches, side effects buried in helpers.
  • Generating initial test coverage I give Claude a function or class and ask for a pytest test file covering happy paths, edge cases, and expected failures. The output needs cleanup, but it scaffolds 70–80% of the boilerplate instantly, parametrize decorators, fixture stubs, mock patches, the lot.
  • Repetitive tasks Generating test data, writing conftest fixtures, converting manual test cases into pytest format, drafting docstrings for test functions. The stuff that takes 20 minutes and feels like zero-value work. Claude handles it in seconds.

My Honest Pros & Cons

✅ What I Love

  • It asks clarifying questions when the prompt is vague. Instead of confidently generating wrong tests, Claude often pushes back — "should this mock the database or hit a real connection?" That alone saves a round-trip of corrections.
  • Context retention within a session. I can paste a base class, then ask about a subclass, and Claude remembers the full picture. Iterating on test design feels like pairing with someone who actually read the code.
  • It explains its reasoning. When it suggests a test approach I wouldn't have taken, I can ask why, and often it teaches me something about the code I'm testing, not just the test itself.

❌ What Could Be Better

  • It doesn't know your codebase. Every session starts cold. I have to re-paste context each time, which adds friction for large or deeply coupled modules. A persistent project-level context would be a game changer.
  • Generated tests can be overconfident. Sometimes Claude produces tests that pass but don't actually assert the right thing — they test the mock, not the behaviour. You still have to read every line before committing.

Pricing: Is It Worth It?

Claude is free to use at claude.ai with a generous daily limit — enough to try it seriously. Claude Pro runs around $20/month and removes the rate limits, gives you access to the most capable models, and adds features like Projects for organising context. For daily professional use across a sprint, Pro pays for itself if it saves you even one hour a week.

My take: If you're using Claude more than a few times a day for real work, Pro is worth it — the free tier will frustrate you at exactly the wrong moment.

Final Verdict

Claude has become a genuine part of my SDET workflow, not a replacement for thinking, but a tool that handles the scaffolding so I can focus on what actually requires expertise: understanding the system under test, designing meaningful assertions, and knowing when a green test is lying to you.

Use it if you spend significant time on boilerplate test setup, PR reviews, or converting manual cases to automation. The ROI is real and fast.

Skip it (for now) if you're working in a highly regulated environment with strict data policies around sharing code externally, or if your test suite is so domain-specific that generic scaffolding creates more noise than signal.

For most SDETs working in Python + Pytest, it's the productivity tool I wish I'd picked up earlier.