Nano Banana, or how to score 20/20 on your next maths assignment

The scene is strikingly simple: a student takes out their smartphone, photographs their wave physics exam, and in a fraction of a second an AI developed by Google (Nano Banana) gives them not only the answer, but the step-by-step reasoning and fake ink marks on the paper.

This scenario is no longer science fiction: it is everyday life in lecture halls. For higher education and the Grandes Écoles, it raises an existential question: if knowledge and its technical solution have become commodities available in one click, what are we still selling?

If school only serves to validate the ability to produce a technical answer, it is already obsolete. The challenge now is to turn this technological break into an opportunity to redefine human added value.

A word of caution: no soft skills without hard skills

Before going further, I think it is crucial to make one fundamental point: school, and engineering school even more so, has many purposes beyond simple technical validation (socialisation, ethics, teamwork, intellectual maturity).

However, it would be dangerous to go to the other extreme and think that technical skills have become optional.

The ability to produce rigorous, accurate technical reasoning remains the absolute prerequisite of scientific higher education. It is on this foundation of hard skills that graduates are first recruited and recognised by industry. An engineer who can manage a team but has no intuitive grasp of the orders of magnitude in thermodynamics, because they have always delegated their calculations to an AI, is an industrial risk.

The goal is therefore not to replace technical skills with “general culture”, but to make sure the technique is genuinely acquired and understood by the human brain, not simply simulated by a machine. We cannot afford to award degrees that are illusions of competence.

I. The finding: the end of the traditional homework assignment

The widespread arrival of LLMs (large language models) such as GPT-4, Claude or Gemini has broken the unwritten contract of traditional assessment.

1. The illusion of competence: a student can get 20/20 on a fluid mechanics assignment without understanding the concept of viscosity. The tool acts as a cognitive prosthesis which, if misused, weakens thinking instead of enhancing it.
2. The commodification of the degree: if school merely checks technical knowledge that AI masters better than humans, the degree loses its value as a signal. Why hire a junior engineer if an AI does the same technical calculation instantly?
3. The failure of prohibition: banning AI is as futile as banning calculators in the 1980s. The tools are here, they cannot be detected by eye, and they will be part of students' working lives.

The problem is not the tool, but what we assess. An exercise asking students to “calculate the standard Gibbs free energy” is now a job for a robot.

II. The market imperative: when companies demand the “augmented” engineer

Today there is a major gap between the academic world and the harsh reality of the job market.

While some schools are still trying to ban LLMs from their classrooms, companies are integrating these tools into their processes at full speed. For an industrial recruiter or a consulting firm in 2025, AI proficiency is no longer a curiosity: it is an expectation of productivity.

A. The new productivity standard for juniors

Companies now expect junior profiles to be operational faster. A young developer is no longer asked to write every line of code by hand, but to know how to orchestrate GitHub Copilot to produce clean, secure code in half the time. A project engineer is no longer expected to spend three days summarising standards, but to know how to use an AI to extract the regulatory constraints in an hour. Refusing to teach these tools means sending graduates into battle with last century's weapons.

B. The risk of unsupervised self-training

If school does not teach the professional use of AI, students learn it “on the street” (or rather on Reddit and TikTok). The risk? That they develop bad habits: pasting confidential company data into public models (a security breach), blind trust in hallucinated results, inability to prompt with context. Higher education must take back control to turn a “cheating” use (hidden and guilty) into a “professional” use (open, secure and effective).

C. AI as the new hard skill

Knowing how to audit an AI-generated answer, understand a model's biases, or connect AI APIs into an industrial process: these skills are becoming as critical as technical English or project management. School cannot look away: it has a duty to deliver “AI-fluent” professionals to the market, able to bring this technological acceleration from their very first day on the job.

III. Solutions: rethinking added value

The goal is to move from a pedagogy of recall to a pedagogy of mental structure and critical thinking. According to SIRVENTIA, here are some answers to the question “How do we make this shift?”, from the simplest to the most structural.

A. Quick wins (immediate implementation)

These solutions require little budget but a change of stance from teachers.

• The return of the oral exam (the modern “khôlle”): it is the ultimate barrier against cheating. A five-minute exchange is enough to check whether the student masters what they wrote.Action: replace a written assignment with a three-minute video in which the student explains their reasoning to camera. AI can write the script, but the student has to own it and understand it to present it.
• Assessment in a “white zone”: to check that the fundamentals are acquired (they remain necessary to judge the AI's work), keep offline, pen-and-paper exams. It is the “sanctuary” of human cognition.
• AI as an adversary (critiquing an answer): instead of asking the student to solve the equation ΔrG° = −RT ln K, give them the answer generated by an AI (deliberately flawed or hallucinated) and ask them to correct and grade it.Target skill: critical thinking and expertise.
• Assessing the process, not the result: stop grading the final answer and grade the prompt engineering and the method instead.Exercise: “Here is the problem. Use an AI to solve it, but hand in your conversation history with the AI. I will grade the relevance of your questions and your ability to spot the machine's mistakes.”

B. Long-term solutions (structural, with resources)

These solutions aim to redefine the DNA of the Grandes Écoles to justify their cost and prestige.

• Hybrid learning and complex projects (PBL, project-based learning): AI excels at atomic tasks (solving an equation). It is poor at handling systemic and human complexity.Strategy: create cross-disciplinary projects (e.g. thermochemistry + environmental ethics + economics). Students use AI for the technical part (the commodity) so they can focus on complex trade-offs and decision-making.
• Back to the real world (hands-on): an AI cannot handle a test tube, solder a circuit or carry out field research.Investment: massively increase lab hours, practical work and prototyping workshops (FabLabs). The school's value becomes its physical infrastructure and the hands-on experience it offers.
• Prompt engineering and algorithmic ethics: make AI a core subject, not to learn to code (which will soon be automated), but to learn to direct and audit the machine.New curriculum: learn to break a complex problem down into sub-tasks an AI can handle, then reassemble and check the whole. That is the role of tomorrow's engineer: a solutions architect assisted by AI.

Conclusion: towards augmented intelligence

The Nano Banana AI model is not the enemy of education: it reveals what has become outdated in it.

If a Grande École merely checks that a student can do what a phone does in three seconds, it is dead. Survival lies in moving the cursor: we no longer train calculators (machines do it better), we train pilots, critics and creators able to make sense of the results the machine provides.

Academic excellence will no longer be measured by the amount of knowledge stored in the brain, but by the quality of the reasoning used to navigate infinite knowledge.