Almost every software agency now says it’s “AI-powered.” It’s on the websites, in the sales decks and in every proposal. But using AI in software development can mean anything from one developer who sometimes asks ChatGPT for a regex to a team whose whole workflow is built around AI-assisted coding, testing and review.
That difference matters. Done well, AI speeds up delivery and frees senior engineers to work on architecture and hard problems. Done badly, it produces code that looks fine on the surface but has security gaps, inconsistent patterns and technical debt that shows up months later.
If you’re hiring an outsourced or nearshore team, these eight questions will show you whether a partner actually has an AI workflow or just talks about one.
1. Which AI tools does your team use every day, and for what?
A vague answer like “we use AI to be more efficient” is a red flag. A mature team can name specific tools, such as GitHub Copilot, Cursor, Codex or Claude Code, and explain where each one fits: scaffolding, refactoring, writing tests, documentation or debugging. The answer should sound like a description of their normal workflow, not a list of features.
2. Who reviews AI-generated code, and how?
AI can write code fast, and it can be wrong just as fast. Ask how AI-generated code gets reviewed before it’s merged. The right answer puts senior engineers in charge of every pull request and treats AI output like a first draft from a junior developer: useful, but never trusted without checking.
3. How do you protect our IP and sensitive data?
This one matters most for fintech, healthtech and any company with proprietary code. Ask which tools are approved, whether enterprise or zero-retention settings are turned on, and whether there’s a policy about what may and may not be pasted into an AI model. If the partner hasn’t thought about this, they aren’t ready to work on your codebase.
4. Has AI changed your estimates and pricing?
If AI really makes a team faster, you should see it somewhere: shorter timelines, more output per sprint, or the same budget stretching further. Ask for a concrete example. A partner that claims big productivity gains but quotes the same timelines as five years ago is probably not using AI in any serious way.
5. How do you use AI for testing and QA?
The biggest gains often come from testing, not feature code. AI tools are good at generating unit tests, finding edge cases and increasing coverage on legacy code that nobody wanted to touch. A team that uses AI only to write new features is leaving a lot of value unused.
6. How do you stop AI from adding technical debt?
Left unchecked, AI tends to produce plausible, repetitive code that doesn’t follow your project’s conventions. Ask how the team keeps things consistent. Good answers include shared coding standards, linting in CI, project-specific context or rules files for their AI tools, and regular refactoring.
7. What’s the seniority mix on the team?
This may be the most important question. AI multiplies the skill of whoever is using it. A senior engineer with AI tools can deliver far more. A junior engineer with the same tools can produce far more bugs, and faster. Partners that take AI seriously usually staff senior engineers, because judgment is what makes AI output safe to ship.
8. Can you show us a real example?
Ask for a walkthrough of a recent feature: how it was planned, what the AI generated, what the engineers changed, and how it was tested. Some teams, like Kambda, make AI-augmented development a standard part of how every engineer works rather than an add-on. A partner working that way should be happy to show you the process.
