What the work actually looks like

You take on discrete engineering tasks — implement a feature, track down a bug in an unfamiliar repository, refactor a module for performance, review someone else's patch and say precisely why it's wrong. The stacks vary from project to project, so you'll spend real time orienting inside codebases you didn't write. Because the output feeds model training and evaluation, the reasoning matters as much as the diff: commit messages, test coverage, and a written account of what you tried and rejected are part of the deliverable, not paperwork bolted on afterward.

Expect a mix of task types within a single engagement. Some days are straightforward implementation against a spec. Others are adversarial in flavour — you're asked to find where a model-generated solution quietly breaks, or to produce a fix that a naive approach would miss. Code review tasks ask you to be concrete: not "this could be cleaner" but "this mutates the input array, and the caller at line 40 depends on it."

What the screen looks for

AfterQuery's screening is AI-led and pushes on specifics. It wants to hear about real repositories, real bugs, and real trade-offs — the language you use daily, the last non-trivial thing you debugged, how you decided a fix was correct. Generic descriptions of "agile development" get follow-up questions until they either resolve into detail or run out. Expect at least one practical component involving reading or writing code, and expect to explain your reasoning in prose.

  • Depth in one language and its ecosystem beats shallow familiarity with six
  • Ability to work without a spec author to ping — asynchronous means genuinely asynchronous
  • Clear technical writing, since documentation is a graded part of the output

Logistics

Fully remote and project-based. Engagements are short — 2–3 weeks — with 10–20 hours weekly on accepted projects, and you choose which projects to accept. Pay in this category has been observed in the $50–100/hr range depending on task complexity and demonstrated depth; it is not guaranteed and varies by assignment. There is no set schedule, but tasks carry deadlines and unresponsive contributors lose queue access.