The first chance
When a convincing application becomes easy to produce, getting trusted may become harder.
The first client has to believe something that later clients can check: that you can do the work.
Until you have a record, you offer whatever evidence you can. You explain the problem unusually well. You notice something others missed. You write a proposal that makes it clear you have thought about this particular job. For someone without the right connections or credentials, that effort can be a way in.
AI changes the cost of producing such an application. A good proposal can now reflect an applicant’s judgment, an excellent writing tool, or both. The employer still has to decide who will deliver.
The danger is a market in which it becomes easier to present yourself and harder to get your first chance. If employers respond to abundant polished applications by trusting only people with a record, the people already inside gain an advantage over those trying to enter. Avoiding that outcome requires a way to create credible evidence of ability before a person has an established reputation.
What a good application tells you
A useful distinction is between making ability easier to see and making an applicant look able. Correcting confusing language can reveal experience that an employer would otherwise miss. Generating a persuasive answer can also conceal whether the applicant understands the question. The same polished paragraph can perform either function.
One large experiment shows the benefit of removing a writing barrier. Emma van Inwegen, Zanele Munyikwa and John Horton studied algorithmic résumé assistance for nearly half a million jobseekers. Assistance increased the probability of being hired by 8%, with no evidence of lower employer satisfaction. This was writing assistance focused on improving résumés, not a test of agents inventing an applicant’s qualifications. Clearer presentation helped people obtain work. The experiment
On Freelancer.com, researchers studied an AI cover-letter tool in two job categories. Letters became better tailored to job descriptions, but tailoring became less predictive of callbacks and offers. Past-review ranking became more predictive of callbacks, with no detected increase for final offers. This working paper concerns one platform; it does not establish that newcomers lost work. Cui, Dias and Ye
An employer cannot solve this simply by asking applicants to try harder. If the extra effort produces another document that the same tool can readily generate, the employer has added a hurdle without necessarily learning more. Applicants may work harder to stand out while employers become less willing to believe them.
Nor does banning assistance answer the central question. If the job will be done with AI, seeing what a person can accomplish with it may be more useful than seeing how they perform without it. What matters is whether the assessment reveals a capability the employer actually needs.
Who pays for the first piece of evidence?
A first job produces two things: work for the customer and information about the worker. The customer pays for the work. Future customers benefit from the information.
That creates a reason for each employer to wait for someone else to take the initial risk. An established worker arrives with evidence already paid for; a newcomer asks the employer to help produce it. A market can have capable people looking for work and employers looking for capable people, yet still struggle to bring them together.
Amanda Pallais tested this problem by hiring 952 randomly selected workers in an online marketplace and varying how much information their public evaluations contained. Both the opportunity to work and more detailed evaluations improved subsequent employment outcomes. The experiment predates the current AI boom. It matters here because it demonstrates that producing credible information can change access to work, even without a new way to train people. Inefficient Hiring in Entry-Level Labor Markets
AI could make this old problem more important wherever it weakens the inexpensive signals newcomers once used. The institutional response I would like to see is a better market for first opportunities: short, paid assignments that produce useful work and a record another employer can inspect.
For example, a new analyst might be paid to reconcile a small, messy dataset. The record would include the brief, the checks performed, the errors found and the client’s acceptance of the result. An applicant who used AI well should be able to show that. A fluent explanation attached to an unchecked answer should not count for the same thing.
This is a proposal, not a claim that an assessment business has solved hiring. It depends on jobs having small pieces of work that are genuinely informative. Some do not. A trial can also be gamed, its assessor can be careless, and a successful narrow assignment may reveal little about working with a team for a year.
A record you can take with you
There is another design choice hidden inside the proposal. If only one platform can verify the work, the platform owns much of the value of the resulting reputation. Leaving it could mean becoming a beginner again.
A useful record would therefore need to travel: what was done, who checked it, under what conditions, and what the evaluator actually observed. Portability does not mean publishing confidential client material. It means that the worker can carry enough verified evidence to avoid rebuilding trust from nothing each time they change marketplaces.
Someone still has to fund the first assignments and their evaluation. Employers may pay when the work itself is useful. Platforms may pay when a better supply of proven workers benefits future transactions. Either way, the price of admission should not quietly become weeks of unpaid labour that only well-supported applicants can afford.
The strongest alternative is that none of this becomes a major constraint. Employers might develop better screening, or AI might improve applicants’ actual performance enough that taking a chance becomes less risky. That would be a welcome outcome. The relevant evidence is whether capable newcomers reach paid work, not whether their applications become more fluent.
A society with widely available expertise should have more ways for people to begin. Whether it does will depend partly on the institutions that turn unfamiliar ability into something others are willing to trust. Giving everyone a better introduction is useful. Giving someone a first piece of work, and evidence they can keep afterward, changes their position in the market.