How to assess AI skills beyond buzzwords
Almost every engineering CV now mentions AI. ChatGPT, Copilot, Claude, Gemini, “AI-assisted development”, “productivity improvement through AI”. The claims have spread far faster than the demand behind them. When we look across our own demand, clients ask for AI-enabled engineers inconsistently. For some roles, it has become a standing requirement, while for others it appears rarely, or not at all.
That gap creates a problem for anyone hiring without a Tech Recruitment Specialist. Candidates have learned that AI belongs on a CV whether or not the role calls for it. In our screening work, we regularly find that impressive-sounding AI claims describe personal use of these tools in a candidate’s own time, presented in language that suggests established professional practice. The terms sound technical, the claims sound credible, and the cost of getting it wrong is a full salary spent on someone whose actual working habits look nothing like those in their CV.
This guide sets out how we, at INSCALE, separate engineers who genuinely work well with AI tools from those who have simply learned the vocabulary. None of it requires you to be technical, but it does require knowing what to ask, what a real answer sounds like.
What is an AI-enabled engineer?
An AI-enabled engineer uses AI tools to build software faster and better. They treat Copilot, ChatGPT, or Claude as an assistant: code generation, debugging, documentation, test scaffolding. The delivery model stays familiar. Sprints, pipelines, and code reviews continue as before, with AI layered on top to remove low-judgment work and increase throughput. What you are hiring is still their engineering judgment, not their tool use.
That’s the profile most companies actually need, and the one most CVs are now claiming. It’s also rapidly becoming a baseline expectation, much like using version control. The question is no longer whether a candidate has touched these tools, because almost everyone has, but whether their use is disciplined professional practice or casual experimentation dressed up for a CV. The labels candidates attach to themselves vary, and the terminology keeps shifting, so treat any label on a CV as a claim to be tested, not a credential.
Warning signs in a candidate's AI claims
The clearest warning sign isn’t any particular claim, but the absence of specifics behind it. These are the patterns we see most often dissolve under questioning:
- “Used AI in development” - On its own, this usually means occasional prompts to ChatGPT for routine coding help. That is normal, but it tells you nothing about discipline, judgement or daily practice.
- A list of tool names with no context - ChatGPT, Copilot, Gemini, Claude as keywords, with no project or task where any of them appears in a sentence.
- Productivity claims with no anatomy - "Improved productivity with AI" or "led AI adoption" mean little unless the candidate can explain what they did, what changed, and how they know.
- Experience that turns out to be personal, not professional - A pattern we see constantly: AI claims that describe use in the candidate's own time. Personal experimentation isn’t disqualifying, but it shouldn’t pass as workplace practice.
None of these rules a candidate out on its own. They are prompts to dig, and the signal is in what happens when you do.
The questions that separate real AI experience from buzzwords
Our method is simple: ignore the labels and ask for one concrete story, then follow it down.
- "Tell me about a specific task at work where an AI tool genuinely helped you. Walk me through it." - One real case, not a general philosophy.
- "Which tool did you use, and why that one?" - Genuine daily users have preferences and reasons. They reach for Copilot for one kind of task and a chat model for another, and they can say why.
- "What was the outcome?" - Real practice produces results the candidate can describe, even roughly: time saved, a bug found faster, documentation that finally got written.
- "Where did the tool get it wrong, and what did you do?" - This question separates rehearsed answers from lived ones. Anyone who uses these tools daily has been confidently misled by them and can describe a specific instance without hesitation.
- "Does your current company use AI tools formally, or is this something you do on your own?" - Followed by whether they received training or are self-taught. Neither answer disqualifies, but together they tell you whether you are hearing about professional practice or personal habit.
You don’t need to follow every technical detail to run this sequence. Candidates with real working habits explain fluently and concretely. Candidates without them become vague within two follow-up questions, and vagueness is easily spotted even by a non-technical interviewer.
How do you know an engineer uses AI responsibly?
This matters more than it first appears. An engineer who ships AI-generated code without reviewing it’s a liability, however productive they look in their first weeks.
The signal we look for is attitude, not tooling. Strong candidates describe AI as a productivity aid whose output requires verification and human judgement. Specifically, they can explain:
- How they review AI-generated code before committing it
- What they check for, typically security issues, performance, and edge cases
- How they test it
- That they understand the code they ship, rather than accepting it on trust
- What "AI-enabled" means for this role in practice, for example, disciplined daily use of coding assistants with proper review habits
- The must-have engineering skills, separated honestly from the nice-to-haves
- The seniority level, defined by responsibility rather than years
- The team context the person will join
- What success looks like in the first six months
A simple revealing question: "When AI generates code for you, what happens between the suggestion and the commit?" The plainest version of the test: do they paste the whole thing in and take what comes back, or do they amend and shape the output? A candidate who reads, edits, and tests will answer instantly, because you are describing their daily routine. A candidate who accepts output wholesale will struggle to describe a step that doesn’t exist.
What credible AI tool use looks like at each level
Expectations should shift with seniority. The same claim means different things from a graduate and a staff engineer.
Junior engineers typically use AI for learning and debugging. That is appropriate and healthy. What you want to hear is awareness: they can describe a time the tool gave them a convincing but wrong answer, and what they did about it. A junior who treats AI output as authoritative is a bigger concern than one who barely uses it.
Mid-level engineers should be able to explain how AI tools fit into their workflow and point to specific projects where that made a practical difference. The use is integrated rather than occasional, and they make informed decisions about when the tools help and when they get in the way.
Senior engineers should be thinking beyond their own keyboard. At this level, credible experience moves from personal use to influence over how the wider team works. That means driving sensible adoption, setting standards and governance for how AI tools are used and where they are not, and connecting that use to outcomes the business cares about, such as delivery speed, code quality and risk. A senior candidate whose entire AI story is personal Copilot use has a junior-shaped skill in a senior-shaped role.
What to specify before hiring an AI-enabled engineer
The fastest way to undermine this entire process is to start with a vague brief. We are direct with clients about which requirements slow searches down: "strong AI experience", "rockstar developer" and "AI ninja" without definitions all produce long shortlists of the wrong people.
Before you brief a recruiter or write the advert, settle five things:
If the AI requirement appeared in your job description because it appears in everyone else's, pause on that. Hiring for a fashionable phrase you can’t define is more expensive than leaving it out.
Assessing AI skills is genuinely difficult right now. The terminology is unstable, the market is noisy, and methods across the industry are still maturing, including ours. The approach in this article won’t make you infallible, but it will make you considerably harder to bluff, and in this market, that is most of the battle.
Related Insights

Why the right engineering environment matters in the AI era

The real cost of waiting for “better talent” in mature markets

INSCALE redefines IT recruitment: Building global tech teams
Let’s talk about the team your business need
Tell us about your roadmap and the skills you need. We’ll explore how a Client Tech Hub could work for you, where it could be based and what it takes to get started.