the market wants experience, uruguay counts certificates

the answer to ai cannot simply be more courses. someone has to pay for the work of learning before the experience can be sold.

in the first half of 2026, 75.9% of the it job ads tracked by Advice Uruguay that specified experience required at least three years in similar roles. searches for junior profiles fell 28.5% compared with the same period the previous year. searches for seniors barely changed.

the market wants people who have already done the work. meanwhile, demand is shrinking for those who need to start doing it.

the easy answer is to open another course. put artificial intelligence in the name, announce places, show enrollment figures. the course may be good. it may teach something necessary. but it cannot, on its own, resolve the contradiction: we are trying to produce experience with instruments that produce training. the two are related. they are not the same thing.

if uruguay wants to keep selling services that require knowledge, it has to address the stretch between learning something and being able to take responsibility for doing it. that stretch does not materialize when a diploma is handed over.

cuti's education compilation, published in april 2026 with data labeled 2024, records 1,779 ict qualifications awarded. broken down, that means 615 technical qualifications, 702 undergraduate degrees, and 462 postgraduate qualifications. technical and undergraduate qualifications add up to 1,317. the scope includes disciplines beyond software, and the records do not necessarily represent distinct individuals.

these are not 1,779 new developers waiting for someone to hand them a computer. a postgraduate program may be helping someone who already works to specialize. someone may earn a degree after years at a company. and a program classified under ict does not automatically certify the ability to implement ai systems.

meanwhile, Advice Uruguay recorded 7,255 it-related job ads in the first half of 2026. these are advertisements, not actual hires or vacancies that necessarily went unfilled.

subtracting annual qualifications from six months of job ads does not produce a shortage estimate. it produces a number that looks like an answer. it mixes periods, people, experience levels, and units that cannot be subtracted from one another.

the distinction matters because a bad calculation can produce bad policy: train more people with the same profile and wait for the problem to disappear.

what these data allow us to say is more specific. hiring searches favor experience, the entry route is weakening, and the education statistics do not measure how many graduates can take on those responsibilities. nor do they allow us to attribute the entire decline to ai. job ads do not isolate its effect from economic activity or competitiveness.

that is enough to reconsider what we are training people for. it is not enough to announce that we are short of exactly so many thousand professionals.

the problem does not end with software companies, either. in 2025, uruguay exported us$1.992 billion in business and professional services and us$1.350 billion in computer services. a great deal of work beyond writing code is at stake. those categories contain very different tasks; they are not amounts we can simply declare replaceable by ai.

imagine a company that keeps a contract and fulfills it with twenty people where it previously needed thirty. it can improve its margin, maintain revenue, and deliver a perfectly acceptable result to its client. from the company's perspective, a productivity gain. from the country's perspective, ten fewer jobs or ten hires that never happened.

exports do not have to disappear for their ability to generate employment to deteriorate.

that possibility changes the education question. preparing people to enter a team structure that may need fewer people is not enough. we have to ask which responsibilities clients will keep buying, which we can create, and how people learn to take them on.

i would not bet on training people around a tool. i would train them around problems someone is willing to pay to solve.

an accountant who understands why a reconciliation is wrong has something you do not acquire by learning to ask a model for an answer. a developer who can diagnose a failure has something different from the ability to generate code. an operations person who understands which exception could stop a service has knowledge that should inform the design of the automation, not be left outside it.

Uruguay XXI identifies 1,399 graduates in finance and accounting in its selection of university business fields for 2023. they are not digital specialists, nor a number to casually add to ict qualifications from a different year. they are, however, an educational base worth considering when discussing where talent can come from.

i would not ask every accountant to start over as a junior programmer. i would give them tools to work with data, understand an automation, and verify its output while retaining their specialist knowledge. from someone with a computing background, i would demand technical depth, but also the ability to understand why something is being built and what happens when it fails.

the profile i would propose has a real specialty and can use data and ai without giving up judgment over the result. they can explain what they did, detect when it does not work, and take responsibility for fixing it. this is not someone who knows everything. it is someone who knows something deeply enough to work with others without outsourcing their understanding to an interface.

take a concrete assessment. give a team duplicate invoices, partial payments, two currencies, and incomplete documents. ask them to automate part of the reconciliation. then change a rule and introduce an error that was not in the examples.

that reveals who understands the process, who can integrate the systems, who finds the fault, and who can explain the remaining risk to the client. those are different responsibilities. they do not have to reside in the same person.

the certificate should follow that demonstration. it should not replace it.

and there are foundations we should not skip. in the monitor from advice and Cuti - Chamber of Uruguayan IT Companies covering november 2023 to october 2024, 58% of it positions requested english. for a client-facing role, i would assess it through explaining a decision and discussing a problem, including assisted translation where appropriate. i would not assume a tool makes comprehension unnecessary.

the same goes for reading, mathematics, and writing. before using those abilities as a filter to exclude people, i would offer bridging instruction. but i would not certify autonomy for someone who still cannot explain why an answer is wrong.

that covers the educational design. now comes the part that needs a budget.

training someone inside an organization takes someone else's time. reviewing work, explaining decisions, letting them attempt something bounded, correcting it, and trying again. a mentor is not training anyone just because their name appears in a presentation. they are training when they have hours set aside to do it.

imagine every company makes the same decision: hire only experienced people and let another organization pay for initial training. each decision may make sense individually. the collective arithmetic does not. someone must have hired that person when they did not yet know how.

if we also automate the tasks through which they used to learn, we have to design another path. sitting them next to a senior and giving them an ai license is not enough. watching someone solve a problem is not the same as learning to solve it.

for the next five years, i would organize two separate tracks. one for reskilling people who already work and have useful business knowledge. another for professional entry, with paid practice, supervision, and responsibilities that grow in verifiable ways.

i would not send both groups to the same course. nor would i promise to manufacture seniors in nine months. the purpose of an initial placement is to develop autonomy in a bounded set of tasks, not to rewrite the participant's résumé to match a job ad.

companies should contribute projects, supervisors, and time. the state can cofinance training and reduce the risk of bringing in someone who is still learning. but it should fund those concrete conditions, not just enrollment.

the worker cannot be the only one paying for the transition, either: studying at night, working for free to gain experience, and also bearing the risk that there will be no job afterward. a training policy that only works for people who can survive months without income has already decided whom it excludes.

it would be convenient to say institutions are doing nothing. it would also be untrue. pixel, from UTEC - Universidad Tecnológica, provides up to 650 places for its 2026 cohort, in a three-semester program covering ai, data, cybersecurity, and challenges developed with organizations.

the program has digital citizenship goals as well as employability goals. it should be assessed against those aims, not expected to meet the entire demand for specialists on its own. but neither should 650 places become 650 available professionals or 650 jobs. none of those equivalences comes into existence just because a cohort has been announced.

for the part of the policy seeking employment outcomes, i would change the unit of measurement. not how many courses someone took. how many distinct people demonstrated an ability and are applying it six or twelve months later. that may be in a new job or in a different responsibility within the job they already had.

i would also measure who could not finish and why. otherwise, a program can improve its success rate by selecting only those who need the least support. the indicator looks beautiful. the gap remains where it was.

that requires following people's paths, recognizing existing abilities, and distinguishing initial training from reskilling. it is less eye-catching than launching another platform. it also lets us find out whether we are buying learning or simply administrative activity around learning.

i would not abolish long degree programs and replace them with modules on this month's tool. the point of fundamentals is precisely to avoid dependence on that tool. i would change the relationship between those fundamentals and the work in which they have to be tested.

one last confusion remains. better training is necessary. it does not create demand by itself.

if a company has no funded project, cannot sell its service, or has a client who decides to do the work in-house, sending the worker back to study does not solve the problem. before diagnosing a talent shortage, we need to distinguish it from a lack of business, unattractive conditions, or a hiring search nobody is actually willing to fund.

there is a very convenient version of the reskilling narrative in which every market difficulty becomes an individual's inadequacy. they lost their job because they did not keep up. they could not get in because they need another course. the diagnosis always ends with the need to study more. companies' commercial responsibility and the country's productive development policy stay out of the picture.

i would not accept that explanation as an automatic answer. nor would i accept the opposite: that preserving today's tasks is the only way to protect the people doing them. the aim should be to give them the means to take on different work and share in its value, not to guarantee that every procedure survives.

uruguay could end up with healthy exports and an increasingly narrow route into professional work. it could also use productivity to build new services and open other paths. neither outcome follows from the number of diplomas it hands out.

a country that wants to sell experience has to decide how it produces it. and who pays for the work of learning while that experience does not yet exist.

sources

advice / cuti, labor market monitor, july 2026, first-half 2026 data; pp. 6 and 12–13. the 75.9% comes from adding 50.4% senior and 25.5% mid-level among ads that specified experience. these are hiring advertisements, not hires or a causal measurement of ai's effect.
cuti, academic education in ict, april 2026, compilation labeled “2024 data”; table 1, p. 10, methodology, and institutional appendix. the 1,317 technical and undergraduate qualifications are the sum of 615 and 702. the records are not a census of available new workers.
uruguay xxi, services exports 2025, final tables, pp. 11–12. business and professional services: us$1.930 billion in professional and consulting services plus us$62 million in technical and other business services. computer services: us$1.350 billion, excluding telecommunications.
uruguay xxi, global export services 2025, table 9, p. 39 of the pdf; 2023 education data. this corresponds to the report's selection of university fields and is not added to the 2024 ict compilation.
advice / cuti's it observatory, labor monitor 2025, november 2023 to october 2024; languages section. requirements in job ads do not measure every graduate's english proficiency.
utec, píxel program, official description, 2026 cohort, duration, and objectives.