How to run a two-week trial with an offshore AI team
By Shaziya Shaikh, AI Engineer at Webify.AI|Updated
Short answer
Pick one real task from your backlog with a clear pass mark, give the engineer access on day one, and judge the result against your own test cases, not a demo. Review the code, tests and written updates as well as the result. At the end, decide: continue, change the engineer, or stop.
Key takeaways
- Use a real task from your backlog, small enough to finish in two weeks, with a pass mark agreed up front.
- Have access, the NDA and your test cases ready before day one, so no trial time is lost waiting.
- Judge the work against your test cases, and read the code, tests and daily updates too.
- Make a check-in at the end of week one, while stopping is still cheap.
- Decide on day ten: continue, change the engineer, or stop. Everything built so far should be in your systems.
What should a two-week trial prove?
That the engineer can ship working AI features in your stack, measure their quality, communicate clearly across the time difference and work safely inside your systems. A trial shows how a team works, not only what it ships.
- Delivery: working software merged into your repository, not a demo on their machine.
- Quality: results measured against your test cases, including what fails.
- Communication: clear daily updates, good questions and no surprises.
- Safety: the least access needed, no code or data in public AI tools, and sensible handling of AI risks.
How do you choose the trial task?
Choose a real task from your backlog that matters, can be finished in about two weeks, and can be checked against a pass mark. It should touch your real systems lightly, and use test or anonymised data rather than live customer data.
| Good trial task | Poor trial task |
|---|---|
| Add retrieval over one set of your documents, scored on 30 real questions | "Explore what AI could do for us" |
| Build one agent tool, such as order lookup, with tests | A full product rebuild |
| Fix a known weakness in an existing AI feature and show the score improve | A task that needs months of data clean-up first |
What does a two-week trial look like, day by day?
Prepare before day one, ship something small in the first days, check in at the end of week one, finish and measure in week two, and decide on day ten. The week-one check-in matters most: it is your cheapest point to stop.
- 1
Before day one
NDA signed, accounts and access ready, the task written down, test cases and pass mark agreed.
- 2
Days 1 and 2
Set-up in your environment, a plan in writing, and a first small change merged.
- 3
Days 3 to 5
A first working version of the task. Check in at the end of the week: continue or stop.
- 4
Days 6 to 9
Finish the task, add tests, measure it against your test cases and fix what fails.
- 5
Day 10
A demo against the pass mark, a written summary, and your decision.
What should you measure in a trial?
The result against the pass mark, and how it was reached. Score the work on your test cases, then read the code and tests, check how quickly questions were answered, and look at how the engineer handled access, data and AI tools.
| Measure | How to check it |
|---|---|
| Result | Run your test cases. Did it reach the pass mark? |
| Code quality | Have your lead review the pull requests and tests. |
| Honesty | Did the updates report what failed, or only what worked? |
| Communication | Were questions clear, and answered within the shared hours? |
| Security | Did they use only the access they needed, and think about risks such as prompt injection (OWASP)? |
What are the warning signs in a trial?
Silence, surprises and demos. Watch for days with no update, problems revealed only at the end, work shown on the engineer's own machine instead of merged into yours, and results that can't be checked against your test cases.
- No written update for a day or more.
- A demo instead of merged code you can run yourself.
- Results that can't be reproduced on your test cases.
- Asking for more access or data than the task needs.
- A different person doing the work from the one you interviewed.
How do you decide at the end of the trial?
Compare the result with the pass mark you agreed, then choose: continue, change the engineer, or stop. If the task fell short but the way of working was good, a clearer brief may fix it. If the way of working was poor, a good result is luck.
- Continue: agree the next piece of work and the hours.
- Change the engineer: ask for a replacement and a handover.
- Stop: remove access, and check everything built is in your repositories.
Where does Webify.AI fit?
Every engagement with us starts with a trial. Two-week trial with the developer you choose. No charge if you stop within the first five working days. Free replacement within five working days, with a handover overlap, for as long as the contract runs.
- You interview the named engineer before the trial starts.
- The work is committed to your repositories, and the IP is assigned to you.
- We sign your NDA before the first call.
Frequently asked questions
Is two weeks long enough to judge an AI engineer?
For one well-chosen task, usually yes. It shows how they plan, ship, measure and communicate. It won't show everything, so keep reviewing in the first months.
Should a trial be paid?
A paid trial on real work gives the most honest picture, and it is fair to the engineer. With us, there is no charge if you stop in the first five working days.
What if the trial goes badly?
Stop, or ask for a different engineer. Remove access and make sure the work done so far is in your repositories.
Can we use real customer data in a trial?
Avoid it if you can. Use test, anonymised or synthetic data, and agree data-protection terms first if personal data is unavoidable.
Who owns the work done in a trial?
Put the trial under the same IP terms as the main contract, so you own the work from day one.
Sources
Checked on 30 October 2026. Webify.AI figures are our published prices.
Start your two-week trial
Tell us the role and when you need someone. We'll email two named engineers with their experience and availability within 48 hours, and you choose who to interview.
- No cost to receive profiles
- Two-week trial with the developer you choose. No charge if you stop within the first five working days.
- Invoiced in USD, GBP, EUR, CAD or AUD by local transfer; IP assigned to you.