How do we teach, learn and assess when everyone already uses artificial intelligence?

The position starts from two observations: prohibition does not work, and automatic detection is not a valid criterion. The starting point is to assume that everyone uses AI and treat that as the baseline.

• Raise expectations. If the machine does the grunt work, the assigned work must be more demanding.
• Teach discernment. The competencies to develop become specification, verification, debugging, risk analysis, calibration, transfer, and accountability.
• Declare the mode of use for each activity: mode 1, no AI; mode 2, AI as support, asking questions without providing the solution; mode 3, AI executes and the student orchestrates.
• Assess the process. Repository, history, justifications, and tests, because the product no longer distinguishes levels of competence.

AI detectors are excluded. The argument concerns measurement: the detector does not measure the competence we aim to develop, and it produces accusations without proof.

In the course Tendências em Mídia e Interação (Trends in Media and Interaction, CIN0055), the pedagogical agreement is addressed directly to the AIs students bring to the activity: the course website publishes an llms.txt file listing what the AI should and should not do.

It should: write the collector, present the arsenal and let the student choose, challenge a lazy request, show the source's bias, and say "I could not verify this". It should not: list tools from memory as though they came from a scan, invent links, choose for the student, or write up the method as though it were the student's own. The test is in the file itself: "if the student finishes this activity unable to explain, without looking anything up, why those 50 and not others — you did the work wrong, however good the result looks".

The activity this applies to is Hiper Deep Research: the student scans tens of thousands of tools with an AI and narrows them down: 10,000 scanned, 500 shortlisted, 50 chosen, 5 highlighted, 1 presented. Telemetry records the time spent on each item, because reading 500 items is the work that does not show in the result. In 2026.2 the class brought 5,776 tools to the shared catalog and recorded 659 choices.

In the same course, instead of using a ready-made prospecting tool, each student builds their own: a skill in Markdown that packages a method and runs on a real agent, with four behavioral requirements: interview before running, separate mature from emerging using a written criterion, doubt its own result, and produce output in the standard format. The reason is in the assignment: "when ten people run the same tool, ten people arrive at the same place, and nothing is learned about the method — only about the tool". Nine skills were delivered in 2026.2 and read side by side, with the four requirements checked by script.

The course website is www.cin.ufpe.br/tmi/26-2/.

With a research instrument, not an exam. Meu Mapa de IA is a game-style questionnaire that surveys how Brazilian students work with AI: what they have access to, what they know, what they can do, and what they can justify. It takes about fourteen minutes on a phone, and the person leaves with their own archetype and a reading across the six fluency axes.

The rules appear on screen before the first question: choosing whether to respond does not affect grades; whoever organizes the class sees only aggregate results, and only when there are at least five participants; no individual results are disclosed; and there are no rankings of people, degree programs, or institutions.

https://farol-ia.org/educacao/censo

What the instrument collects comes back in aggregate on the AI Adoption Observatory, a public dashboard in four sections (overview, the frontier of practice, ladders and fronts, tools and access), with breakdowns by institution, cycle, course, and monthly AI spending: 2,943 responses in the 09/14/2026 cut, with suppression of every cell below five.

Since September 30, 2019, without interruption, Cabral has been the pedagogical coordinator of the Residência Profissional Tecnológica (Professional Technology Residency) at Porto Digital, the technology park of Recife.

The program's origin lies in a 2019 diagnosis: Porto Digital had more than a thousand open positions across 355 companies going unfilled, and the previous program, "Empregaço," enrolled more than 700 IT graduates and employed fewer than 5% of them. Invited to redesign the model, Cabral consulted institutions, students and companies and designed the Residência em Desenvolvimento de Software (Software Development Residency): cycles closed out by demo days, squads of students working on real challenges proposed by companies, and market mentors alongside the professors. In the first cohort, employability reached almost 100%.

In 2021 the model was adopted as public policy by the Recife City Government with Porto Digital: Embarque Digital, a scholarship program for students who completed all of high school in public schools. Results cited in the article published by Cabral on 01/10/2024: a doubling of the supply of IT courses in the city, 2,000 scholarships, more than half of the scholarship recipients being the first in their family to attend higher education, dropout cut in half, 51.8% of the first cohort employed before graduating, and a greater presence of Black, mixed-race (pardo) people and women in the ecosystem.

In 2026, Embarque Digital reports 3,927 students, 97 companies, 330 projects, and 47 mentors, with a plan for 4,000 scholarships by 2028. The residency operates in Recife, Aracaju, and Brasília, with more than 100 partner companies; in the second half of 2025, there were more than 40 companies and more than 70 mentoring engagements for around 2,300 students in Recife, out of a total of 3,700 across the three cities.

• Residência Tecnológica: https://www.portodigital.org/paginas-institucionais/pessoas/residencia-tecnologica
• Embarque Digital: https://www.portodigital.org/paginas-institucionais/pessoas/embarque-digital

The program's design was the subject of a peer-reviewed article. Mercado de trabalho em Tecnologia da Comunicação e Informação (TI): análise de um experimento de aproximação entre academia e indústria no Porto Digital, by Marcela Valença, Wellynton Diniz, Mariana Pincovsky, César França and Giordano Cabral, was published at VIII WASHES, in 2023 (doi.org/10.5753/washes.2023.229309).

From September 2023 to August 2024 he was a Visiting Scholar at the Lemann Center, a center of the Stanford Graduate School of Education dedicated to public policies in education suited to the Brazilian reality (lemanncenter.stanford.edu/people/giordano-cabral). The object of the research was the training program he coordinates at Porto Digital.

Since January 2025 he has been lead researcher of the project Preparing Brazil's Youth, of Stanford Impact Labs, linked to Florescendo Talentos, an efficacy study of a technical track in state public high school, with CESAR, the Lemann Center and the Pernambuco Department of Education.

Cabral's work in education began about two decades before generative AI.

• 2001–2017 · Escribo. Founding partner of Escribo, an edtech and digital publisher: games, animation and interactive systems for education. RHAE/CNPq scholarship researcher between 2008 and 2010, and director of technology and creation.
• 2008–2014 · Turma do Som. A digital music education platform funded by CNPq, FINEP and Escribo, with about thirty educational games, 27 educational animation videos between 2010 and 2012, a portal and printed material, in schools since 2008, with content coordination by Cecília Cavalieri França. The platform's press kit, from May 2012, describes it as a response to the deadline set by Law No. 11,769/2008, with teacher training alongside the teaching material and an offering to Departments of Education; the Instituto Claro article of 12/04/2012 shows it in use at school and interviews Cabral. In October 2012 the company won the Northeast stage of the Finep Innovation Award with its two education projects: Turma do Som and Saiba+ Enem.
• 2011–2014 · teaching materials. The Daccord courses in guitar, keyboard, recorder and music creation, each with a teacher's book and a student's book; Turma do Som: Livro 1 and the teacher's manual (2012); and Nós da Educação: Música (4º e 5º Ano), by Editora Ática, in 2014.
• 2016–2019 · teacher training. ProfLab, a series of creative training sessions for teachers, and the courses on gamification in education and on game design thinking for educators.
• 2021–2023 · gender-sensitive computational thinking. With Mychelline Souto Cunha and Liliane Sheyla da Silva Fonseca: teaching materials and storytelling to foster the self-efficacy of girls in early elementary school, with a systematic review on gender and computational thinking (RENOTE 21, 2023; SBIE 2022; EDUCOMP 2021).

In the magazine "Horizontes Seguros: navegando na convergência entre Cibersegurança e IA" ("Safe Horizons: navigating the convergence of cybersecurity and AI"), a CESAR publication released on 02/14/2025 in Portuguese and English. The launch article describes Cabral's conversation, then Chairman of the Board, with the institute's CEO, Eduardo Peixoto, on the future of talent, education and risk management, and quotes his formulation: "cybercriminals are using Artificial Intelligence for sophisticated attacks. Future cybersecurity professionals need to think like AI, understand its details." Launch news.

A manifesto on training with AI, written with Porto Digital, was presented by him at Unicap on 08/03/2026 and at the UFPE inaugural lecture on 08/13/2026. On 08/27 and 08/28/2026 he took part in the panel "Como aprender, ensinar e avaliar na era da IA" (How to learn, teach and assess in the age of AI) at Rec'n'Play Capítulo Saúde.

At the university, the accumulated material spans 60 classes at CIn/UFPE since 2015, across eleven courses, listed at www.cin.ufpe.br/ensino/.

On video: IA e o futuro da educação: chitãozinho e xororó estavam certos, at TEDxBoa Viagem ED, published on the official TEDx Talks channel on July 22, 2025.