For the past few years, AI fluency has carried a certain professional cachet. Someone who knew how to prompt well, automate a few tasks or assemble an impressive prototype could appear unusually capable. That advantage is already thinning.
AI literacy is spreading through the workforce at remarkable speed. LinkedIn’s 2026 labour-market research, for example, found that US jobs requiring AI literacy grew 70% year on year. Familiarity with AI is becoming closer to spreadsheet competence or search literacy – useful, expected and widely available.
For founders, this is liberating. One person can research a market, draft copy, analyze customer feedback, mock up a product and automate parts of an operation with far fewer early dependencies. Specialist expertise still becomes essential as the company grows, yet its value arrives at a different point. Experts bring judgment, refinement, domain knowledge and the ability to make a promising system survive contact with customers, regulation and scale.
The same change is arriving in hiring. “AI skills” on a CV are no longer eye-catching. Employers will want to know what happens after the prompt.
Using AI and delivering with AI are different disciplines
A polished demonstration can be produced in an afternoon, but a useful business system has to keep working on an ordinary Tuesday three months later.
Tool fluency is needed, where the person responsible has to understand the business problem, know which data can be trusted, redesign the surrounding workflow and define how performance will be judged. They need to test edge cases, run evals, manage failure modes and communicate with the people whose work will change.
This is where much of today’s AI enthusiasm hits a wall. BCG’s August 2026 AI research found that just 6% of companies were seeing meaningful value from AI through lower costs or higher revenue. The separation between a compelling demo and a valuable deployment is therefore becoming one of the defining skills of the AI workforce.
Evals are worth discussing here. An AI system used for customer support, sales qualification or financial analysis needs a repeatable way to judge whether it performs well under real conditions. OpenAI’s business guidance describes evals as a way to translate business objectives into measurable performance, using real examples, edge cases and domain expertise. Quality can therefore improve over time, allowing businesses to decide when a system is ready for wider deployment.
Another way to think about it is a builder validating their hypothesis. A builder may think switching out the prompt or the AI model will make the workflow better, but that’s truly just a guess. Running an eval is needed to either validate that hypothesis objectively, or prove it wrong.
Companies need a better way to decide whom to trust
Millions of people now have access to the same AI tools. That makes access a weak measure of ability. A polished demo can be produced quickly, yet real value emerges when the system is used by customers, tested against imperfect data and judged against commercial results.
A case study gives employers far more to work with. It shows whether someone identified the right business problem, agreed on a measurable target before building and stayed with the project long enough to prove that the result held up.
The weeks and months after launch often reveal more than launch day itself. A system may save ten hours in its first week, then require substantial review later. An assistant may perform well on carefully selected examples, then produce weaker results across everyday customer requests. Employees may enjoy using a tool while the financial return remains modest.
The strongest evidence comes from sustained performance after deployment. It shows whether the chosen KPI improved, whether the gains lasted and whether the result justified the time and money invested.
Certificates < evidence
Traditional credentials can show that someone completed a course, studied a subject or passed an assessment. Employers hiring for AI work need stronger evidence of how that person performs when the work reaches a live business environment.
A portfolio gives them more to judge. It can show the problem someone tackled, the starting point, the KPI they agreed on, the result after deployment and the changes they made along the way. Over time, it can also show whether those gains lasted.
AI work also draws on skills far beyond technical proficiency. The World Economic Forum’s Future of Jobs Report 2025 ranked AI and big data among the fastest-growing skills, while analytical thinking, leadership, collaboration and resilience also remained high on employers’ agendas. The strongest practitioners will often be those who combine AI fluency with deep industry knowledge, sound judgment and the ability to explain their work to the people who rely on it.
Proof of work will become the strongest AI credential
The phrase “proof of work” may sound severe, but it describes where hiring is heading. Employers will value evidence that someone built something useful, deployed it successfully and produced a measurable return.
That evidence can take many forms:
- A founder might show that an AI sales workflow raised qualified meetings by 22% over a quarter;
- A finance lead might document a forecasting system that cut preparation time while preserving accuracy;
- A customer-service manager might show lower handling time alongside stable satisfaction scores.
A strong portfolio should connect the system to a measurable business result and show how the person reached it.
Results over several months carry far more value than a short build. They show whether someone can track performance, fix problems as they emerge, improve the system and keep stakeholders aligned over time.
Business leaders are already demanding this kind of evidence from AI spending. Deloitte’s 2025 survey of 1,854 executives across Europe and the Middle East found that satisfactory returns on a typical AI use case often took two to four years, while 6% reported payback within a year.
BCG also found in 2025 that just 40% of companies defined and monitored financial KPIs for AI value creation. Indeed, claims become credible when outcomes are measured.
Domain expertise becomes more valuable with AI
There is a temptation to imagine the best AI practitioner as the person who knows the newest tools first. In many businesses, the stronger candidate will be the person who understands the work itself:
- A lawyer who knows where legal review slows a deal can identify a better use case than a talented outsider chasing a flashy automation;
- A recruiter who understands candidate quality can judge an AI sourcing system against the right outcome;
- A sales leader who knows why prospects disappear after a demo can choose a better intervention than someone optimising activity for its own sake.
AI lowers the cost of building, but domain expertise improves the odds of building the right thing.
That combination also changes the role of specialists. AI gives founders and generalists the ability to reach further before hiring deeply specialised teams. Experts then contribute where judgment becomes expensive to fake – at the point where accuracy, regulation, customer trust, scale and commercial consequences become serious.
Start with one constraint and one KPI
Professionals who want to prove their AI capability should begin with a real business problem. Find the constraint holding a team back and focus the project there.
That could mean response time, conversion, support cost, research output or forecasting accuracy. From there, choose one KPI that defines success. The builder and the stakeholder should agree on it before development begins.
This gives the project a clear benchmark from day one. Everyone knows what success looks like, and the eventual result can be judged against an agreed starting point rather than enthusiasm around the technology.
Then document what happened. Record the baseline, the data used, how performance was evaluated, when the system was deployed and how the results developed over time. Changes made along the way are part of that record too. They show how someone responded when the first version met real conditions.
That evidence shows that someone can identify a worthwhile problem, build around it and prove that the work produced a measurable commercial result.
AI literacy will soon be commonplace across many professions. Proof of work will separate the people who can use the tools from those who can turn them into lasting business value.
