The short answer
Two weeks of real work, paid at the normal rate, on a task that matters but is not on the critical path. You learn how they handle your codebase, your review process and your ambiguity, none of which an interview reveals.
Write the criteria down before it starts. Deciding afterwards what you were looking for produces a judgement about whether you liked them.
Why interviews predict poorly here
An interview tests how someone performs in an interview. Augmentation asks something different: can they become useful inside an unfamiliar system, with incomplete documentation, while asking a reasonable number of questions.
- Reading unfamiliar code is the core skill and is rarely tested
- Knowing when to ask rather than guess is a judgement call
- Working within existing conventions matters more than personal preference
- Handling review feedback well is visible only in review
Designing the trial
- Pick genuine work with a clear definition of done, ideally two or three days of effort for someone who knows the system.
- Prepare it properly: context, acceptance criteria, where to look.
- Give the same onboarding you would give for a full engagement.
- Let them go through the whole process including review and deployment.
- Hold a short conversation at the end about what was unclear.
That last step is genuinely useful regardless of the decision. Someone seeing your process for the first time will point out friction your team stopped noticing.
What to judge
| Look at | Good sign | Concern |
|---|---|---|
| Questions asked | Specific, after looking | None at all, or constant |
| First pull request | Follows existing conventions | Personal style imposed |
| Response to feedback | Absorbed and applied | Repeated, or argued at length |
| Handling ambiguity | Raises it, proposes an answer | Guesses silently, or stalls |
| Communication | Proactive about blockers | Silence until asked |
The ambiguity row is the most predictive. Every real engagement contains unclear requirements, and how someone handles the first one tells you how the next six months will go.
Pay for it
Unpaid trials get you people with no better options, and they are not a fair test because nobody does their best work for free. Pay the normal rate for real work you keep.
It also changes the relationship. A paid trial is the start of an engagement that you might not extend. An unpaid one is an audition, and people behave differently in auditions.
Be honest about what happens next
Say at the outset that it is a trial, what you are assessing, and when you will decide. Suppliers and developers both plan around availability, and being vague costs them real money.
If you decide not to continue, say why. It is a short conversation and it is the difference between a supplier who will work with you again and one who will not.