AI is making it easier to produce the signals of professional competence. The harder question is what those signals actually prove.
A candidate presents an application they built with AI.
The interface is polished. The application works. The README explains the architecture, the portfolio describes the problem it solves, and the candidate can talk through the features.
Then the interviewer asks a question:
Why did you choose that approach?
The candidate explains what the application does. But when the conversation turns to alternatives, trade-offs, and what might break as the system grows, the answers become less certain.
Nothing about this necessarily means the candidate lacks ability. They may have used AI effectively and learned a great deal along the way. But the application, impressive as it is, has left an important question unanswered.
How much does the finished product tell us about the person who produced it?
This question matters beyond software development. AI can help people write CVs, prepare for interviews, build portfolios, produce presentations, and describe their professional achievements. The things we use to recognize talent are becoming easier to produce.
And that creates a problem for anyone trying to hire well.
When everyone can look qualified, what makes someone stand out?
The shortcuts we learned to trust
Hiring has always involved a degree of uncertainty. An employer cannot know exactly how someone will perform six months after joining a team. Even a candidate with an impressive track record must adapt to a new environment, different colleagues, unfamiliar problems, and expectations that may not resemble their previous role.
Employers therefore look for indicators that help them make a decision.
A degree might suggest foundational knowledge, discipline, or the ability to complete a demanding course of study. Years of experience might suggest familiarity with the work. A recognized previous employer might suggest exposure to strong engineering practices. Certifications can demonstrate knowledge of a particular subject. A CV filled with action words such as led, designed, implemented, and optimized might suggest impact.
Even location can function as a shortcut. An employer may assume that a local candidate will be easier to communicate with, manage, or integrate into a team.
None of these indicators is inherently useless. The problem is what happens when we begin treating them as reliable substitutes for the qualities we actually want to identify.
5 years of experience tells us how long someone has been exposed to a field. It does not tell us how much they have learned during that time.
Working at a prestigious technology company tells us something about the environment in which someone worked. It does not, by itself, tell us which problems they solved or how much responsibility they carried.
A certification can demonstrate knowledge of a subject without necessarily demonstrating the ability to apply that knowledge in a difficult situation.
And a CV can describe an impressive achievement without telling us whether the candidate made the critical decisions, contributed a small part of the work, or simply helped communicate what the team accomplished.
These distinctions are not new. What is changing is how easily the signals themselves can be produced.
A proxy can be useful without being conclusive.
For years, hiring systems have relied on these proxy signals because directly observing someone’s ability is difficult, expensive, or impractical.
The question is not whether employers should stop using them. It is whether the evidence they provide is still strong enough to support the conclusions being drawn from it.
AI makes that question harder to ignore.
When the signal becomes cheap
Imagine a recruiter reviewing a CV that says:
Developed and implemented a scalable machine learning pipeline to improve operational efficiency.
The sentence sounds credible. It uses familiar professional language and describes a meaningful accomplishment.
But what does it actually tell the recruiter?
What did the pipeline do? What made the original system inadequate? Which decisions did the candidate make? What does scalable mean in this context? How was the improvement measured?
The sentence might describe substantial engineering work. It might also describe a relatively small contribution dressed in impressive language.
Before generative AI, producing polished professional writing required a certain level of writing ability, knowledge of recruitment conventions, or help from someone who possessed those skills. Now, a candidate can describe their experience in ordinary language and ask an AI assistant to turn it into a compelling CV bullet.
That can be a good thing.
Someone who has done excellent work but struggles to articulate its value can communicate their contribution more effectively. A career changer can better explain how previous experience relates to a new role. A graduate can learn how to describe a project without relying on vague statements about being passionate or hardworking.
AI can make genuine capability easier to see.
But it can also make the appearance of capability easier to manufacture.
A candidate can use the same technology to inflate a minor contribution, describe a project they barely understand in sophisticated language, or prepare answers that sound more experienced than they are.
AI can improve the signal without necessarily improving what the signal represents.
This is the problem of signal inflation. When more people can produce a convincing signal at a lower cost, that signal becomes less useful when considered on its own.
It does not become worthless. A well-written CV still helps an employer understand someone’s background. A polished portfolio can still communicate care and attention. But the employer can no longer safely assume that the quality of the presentation tells them much about the depth of the underlying capability.
The same problem extends beyond writing. AI can help generate code, presentations, business plans, case studies, and even convincing explanations of technical decisions.
The question is no longer simply whether a candidate can produce an impressive artifact. It is what the artifact allows us to conclude about the person behind it.
From described output to demonstrated capability
If a CV is an imperfect account of someone’s abilities, the answer might seem straightforward: ask candidates to demonstrate what they can do.
There is good reason to take that approach seriously.
The US Office of Personnel Management describes work samples and simulations as assessments that mirror activities required by a job. Its guidance emphasizes using them when candidates should already possess the relevant competencies on entry, rather than expecting them to learn those competencies after being hired. Its guidance on work samples and simulations also addresses practical considerations such as cost, time, and the relationship between an assessment and the work itself.
Several technology companies illustrate different ways to put this principle into practice.
Zapier’s hiring guide describes a process involving multiple stages, including interviews and a short take-home assignment. Automattic describes role-specific assessments and, for some positions, practical tests or trial work designed to resemble real responsibilities. GitLab’s technical interview guidance includes an approach in which some candidates review a merge request in advance and then discuss their review and make code improvements during a technical session.
These are not identical methods. One combines a take-home assignment with interviews; another uses role-specific practical work; another examines how a candidate reviews and improves existing code.
What connects them is the attempt to observe something closer to the work the candidate would actually perform.
A candidate might claim experience solving business problems with data. A relevant exercise could reveal how they clarify an ambiguous question, assess data quality, choose an approach, and communicate limitations.
The point is not to catch candidates out. It is to gather evidence that a CV alone cannot provide.
When the usual signals fall short
The question of what counts as evidence extends beyond how candidates are tested. It also affects which backgrounds employers trust, which qualifications they value, and whose ability they recognize in the first place.
Gerard Holland, Founder and Chief Commercial Officer at Talent Match Africa, has seen this tension in international hiring. His work connecting employers in Australia, the UK, and North America with professionals across Africa has brought him into contact with assumptions about what makes someone a credible candidate.
In our conversation, Gerard described how some employers initially believed they needed to hire locally. As they gained confidence in the hiring process and the working environments supporting distributed teams, geography became less important. What mattered more was whether the professionals they hired could deliver.
The shift is revealing. Geography had functioned as a proxy for confidence in the working relationship. Once employers had more reason to trust the process, that proxy lost some of its influence.
Gerard also believes employers can overvalue pedigree: familiar universities, well-known employers, and conventional career paths. And as AI makes CVs easier to polish, he questions how much confidence recruiters should place in the document alone.
His distinction was simple: “The CV should be a ticket into the process, not the evidence that settles the decision.”
Gerard says he increasingly asks candidates a practical question: What’s your favourite AI tool, and what have you actually built with it?
The value of that question isn’t the candidate’s choice of tool. Nor is it the fact that they have built something with AI. It is the conversation that follows.
What problem were they trying to solve? What did they discover along the way? What choices did they make? What failed? What would they change if they started again?
A candidate who can explain those details provides information that a list of technologies cannot.
Consider two descriptions of an AI-powered customer support application.
The first says that the candidate built an application using Python, FastAPI, and a language model.
The second explains that the initial retrieval approach struggled as the knowledge base grew, that the candidate investigated the failure, compared alternatives, changed the architecture, and measured the resulting improvement.
The second description offers more useful information because it connects a problem to a series of decisions and an outcome.
Of course, AI can help write both descriptions. A detailed account is not automatically truthful, and impressive metrics are not automatically reliable. The difference becomes meaningful when the candidate can explain the work, establish their own contribution, discuss the trade-offs, and respond thoughtfully when the interviewer explores the details.
The aim is not to reward people who can describe their work most convincingly. It is to understand what they actually contributed and what that contribution demonstrates.
But evidence is not proof
There is a catch.
A practical assessment can reveal more than a CV, but it does not eliminate uncertainty. A polished portfolio can be misleading; so can a successful work sample.
A candidate may perform well on a short exercise but struggle with the scale, ambiguity, collaboration, or operational responsibility of the actual job. Someone else may perform poorly in a timed assessment despite being effective in a real working environment.
Research supports the usefulness of work samples, but not the idea that they are perfect predictors. A 2005 meta-analysis by Roth and colleagues examined the relationship between work-sample test performance and job performance, finding a meaningful but imperfect association. The study is available through its publication record.
That distinction matters. An assessment is valuable because it provides relevant evidence, not because it removes the need for judgment.
There are also questions of fairness.
A lengthy unpaid assignment may favour applicants with more free time. A task requiring expensive equipment may exclude capable people who cannot access it. A live coding exercise may reward confidence under observation more than the ability to solve problems in the conditions of an actual job.
And when employers repeatedly ask candidates to complete substantial work without compensation, they risk shifting the cost of their hiring uncertainty onto applicants.
Replacing CV screening with a demanding assessment process does not automatically make hiring better. It can simply create a different set of barriers.
The answer is to design assessments around the capabilities that genuinely matter for the role, keep the process proportionate, and evaluate candidates against consistent criteria.
There is another consideration: the evidence required depends on what the employer is trying to establish.
For an experienced engineer, it may be reasonable to examine previous architectural decisions, production incidents, code reviews, and the outcomes of systems they have operated.
For a graduate, the absence of an extensive record of professional delivery should not be treated as proof of limited potential. Their reasoning, ability to learn, approach to unfamiliar problems, and response to feedback may be more informative.
Past performance, present capability, and future potential are related, but they are not interchangeable.
A sound hiring process recognizes the difference.
What this means for people trying to stand out
For professionals, the implications are not that CVs no longer matter or that everyone must publish a portfolio of impressive projects.
Presentation still matters. Credentials can matter. Experience can matter. The challenge is to ensure that these things communicate something real rather than becoming substitutes for it.
When describing a project, explain the problem, not just the tools. Describe the decisions you made, not just the tasks you completed. Where possible, show the outcome and how you evaluated it. Be honest about what you contributed, what you learned, and what you would do differently.
A small project that demonstrates thoughtful decisions may tell an employer more than a long list of technologies. A candidate who can explain why an approach failed may reveal more than someone who presents a flawless demo without acknowledging its limitations.
This is particularly important for people early in their careers.
They may not have a prestigious employer on their CV or years of experience to point to. That does not mean they have nothing meaningful to demonstrate. They can show how they approach a problem, how they learn, how they respond when something doesn’t work, and how they use tools to make progress.
They should not have to pretend to be experienced to be taken seriously. They need opportunities to make their current capabilities–and their potential to develop–visible.
For employers, the corresponding responsibility is to avoid confusing familiarity with merit. A candidate whose background resembles that of previous successful hires may be a sensible choice, but familiarity should not become the only route through the hiring process.
A less conventional background is not evidence of greater talent, either. It is a reason to look carefully at what the candidate can demonstrate rather than deciding in advance what their background means.
The signal and the thing itself
There is a useful parallel between this problem and what AI is doing to technical work.
As AI makes it easier to produce software, writing code becomes a less complete measure of a developer’s value. The ability to define the problem, make sound design decisions, verify an implementation, and understand its consequences becomes more important.
Something similar is happening in hiring.
As AI makes it easier to produce the signals associated with professional competence, the ability to produce those signals becomes a less complete basis for judging talent. Employers must pay closer attention to what those signals demonstrate–and what remains unknown.
Neither shift means that the old skills or signals have become irrelevant. Code still has to work. CVs still have to communicate. Credentials can still matter. Experience remains valuable.
What changes is how much we can reasonably infer from any one of them.
The temptation will be to search for a new shortcut: a better assessment, a more sophisticated AI detector, a portfolio requirement, or another universal measure of ability. Some tools may help in specific contexts. None will remove the need to understand the work, the person, and the decision being made.
Hiring will always involve uncertainty. The aim is not to eliminate it through a perfect test. It is to make better decisions by gathering more relevant evidence, interpreting it carefully, and acknowledging what we still cannot know.
When everyone can produce a convincing signal of qualification, presentation alone becomes less useful as a differentiator.
The harder–and more valuable–task is to make capability observable.
Because the future of talent evaluation may depend less on how convincingly someone can present themselves as qualified, and more on how effectively we can understand what they can actually do.
When everyone looks qualified, the task is not to find the person who looks most impressive. It is to get better at recognizing what makes someone good.
If you enjoyed this article, you might also like:
If AI Can Code, What Should a Programmer Know?
We have spent decades making software development more abstract. AI is simply taking that abstraction further. The question is not whether we should use it, but what expertise remains essential when implementation becomes cheap.
Samuel, this hits small businesses even harder. A 10-person company can't afford a bad hire, and it usually has no technical interviewer to ask the follow-up question. The owner I'd want to hire for is the one who can explain what they'd do when the AI gets it wrong. That's still hard to fake. Have you seen any simple tests a non-technical owner could run?
'Why did you choose that approach?' is the whole thing. AI can produce the artefact but not the judgement behind it. Same with leaders: trade-off thinking is the new signal.