The practice · open
pAIr
It is a way of working. You can meet someone and practise pAIr freely: no permission, platform or contract with nualart.cat is required.
An open practice · two humans + AI
Two people.
One real project.
AI at the table.
pAIr adapts pair programming: two people work together on the same project and use AI as their main tool to think, decide and build.
pAIr is the open practice. pAIr Sessions is the professional service.
2 people+ AI1 real project→ tangible result
Pair programming, beyond code
01 / WHAT IT IS
What is pAIr?
It is not “writing a prompt”.
It is thinking together while doing.
Pair programming, consolidated in the second half of the 1990s within Extreme Programming, made conversation, decisions and mutual learning part of the act of programming itself.
pAIr brings this idea to real projects in any discipline. Two people share context, screen and decisions; AI is the main tool and an active intelligence: it contributes knowledge, generates alternatives, challenges and produces.
The service · paid
pAIr Sessions
Professional support from nualart.cat: preparation, facilitation, technical judgement and shared production around your challenge.
See how it worksBuilding means more than making it work
02 / TECHNICAL JUDGEMENT
What can fail when AI programs what nobody understands?
An app can work
and still be dangerous.
AI lets people build without knowing code, servers or databases. It can also leave invisible errors that only surface when real users, data or attackers arrive.
There is no rigorous figure for “all apps made by non-programmers”. The closest evidence is this: in a study of 200 real, deployed vibe-coded applications, 90% contained at least one vulnerability; over three quarters were high severity or critical.
of 200 deployed vibe-coded apps contained at least one vulnerability.
of generated samples in a test of more than 100 models failed security tests.
tasks in a user study produced less secure solutions with an AI assistant.
Permissions that do not protect
A screen may look private while the server allows access to or changes in somebody else’s data.
Exposed secrets
API keys, passwords or personal data may end up in the browser, repository or public logs.
Inputs turned into attacks
Unvalidated forms can open the door to injection, cross-site scripting or malicious content.
Unsafe dependencies
AI may suggest packages that do not exist, are obsolete or vulnerable, turning installation into an entry point.
No plan for the day after
Without backups, logs, updates and rollback, a working demonstration is not yet a maintainable product.
The answer is not to stop building.
It is to combine AI’s power with technical judgement: understand the architecture, inspect the code, test the limits and take responsibility before publishing.
Sources and scope
- Deng et al. (2026), Understanding the (In)Security of Vibe-Coded Applications: 200 deployed apps, 1,471 vulnerabilities, 90% with at least one.
- Veracode (2025), GenAI Code Security Report: over 100 models; 45% of samples failed security tests. This is not a percentage of complete apps.
- Perry et al. (2023), Do Users Write More Insecure Code with AI Assistants?: controlled study with 47 participants and five security tasks.
What can we bring to the table?
03 / CHALLENGES
An urgent project,
a postponed idea,
a problem that costs you every day,
an invention…
You do not need to arrive with a solution or know how to use AI. You need to explain what you want to change, what you already know and why it matters.
Build
A website, prototype, automation, visualisation or internal tool.
Solve
A slow process, repetitive task or problem that consumes time and money.
Think
A difficult decision, business idea or deep, challenging ideation session.
Review
A product, text, experience, strategy or campaign before publishing it.
Research
A complex topic, the available alternatives and the evidence that should guide a decision.
Learn by doing
A technology or method applied directly to your own project, not a generic exercise.
From challenge to result
04 / PROCESS
We arrive prepared.
We build live.
A pAIr Session may be exploratory, productive or both. It should never be a passive AI demonstration.
- 01
Orient
Understand the intention, context, constraints and what a good result would mean.
- 02
Agree
Recommend preparation, duration, pace, deliverables and budget before starting.
- 03
Prepare
Study materials, research and prepare context so shared time is productive.
- 04
Do pAIr
Talk with AI, disagree, make decisions and produce together on the same object.
- 05
Continue — or not
Close with a result and next steps. If more work is needed, agree another session or a separate commission.
pAIr Sessions · professional service
The challenge decides the format.
Not the other way round.
We do not sell a package of hours before understanding what you want to achieve. We begin with a short conversation, recommend a format and prepare a clear quote.
Tell us your challengeFocus Session
Clarify a problem, review a proposal, challenge a decision or unblock one specific point.
Possible resultA map, sound judgement, a decision or a first useful piece.Build Session
Work continuously and build a first version: a prototype, narrative, campaign, process or tool.
Possible resultA tangible result that can already be tested.pAIr Day
An intensive challenge that needs exploration, decisions, production and review in a single day.
Possible resultA solid version and clear next steps.pAIr Cycle
Two or more sessions separated by testing, feedback, reflection and further building.
Possible resultIterative development, not an improvised marathon.Preparation may include reading, research, tests or technical setup. It is assessed according to the challenge.
We can agree review, documentation or a feedback session. It is not added automatically.
If specialised production is needed outside the shared conversation, it is quoted as a separate commission.
Where does it come from?
pAIr is an initiative by nualart.cat
A proposal to make the encounter between human knowledge, technical judgement and artificial intelligence useful.
Jaume Nualart is its initial creator. He brings more than twenty years connecting data visualisation, interface design, research, code and digital projects. pAIr names a practice anyone can adopt; nualart.cat offers the professional service to prepare and facilitate sessions when the challenge requires experience, structure or technical responsibility.
Necessary questions
Before doing pAIr
Can I do pAIr without hiring nualart.cat?+
Yes. pAIr is an open practice: meet another person, choose a real project and work with AI as your main tool. The paid service is professional support, not the right to practise pAIr.
What do I hire in a pAIr Session?+
Preparation, facilitation, technical judgement and shared working time. We agree the intention, duration and possible result beforehand. Pure execution work is separated.
Is AI a tool or a third intelligence?+
Both: the main tool and a participating intelligence that can know more, reason, propose and produce. This does not remove human responsibility for goals, verification and consequences.
Is it only for programming?+
No. It originates in pair programming, but the practice can research, write, design, develop a campaign, review a product or transform a working process.
Do I need to know AI or programming?+
Not to begin: you need to know your context or have a real question. But if the result is software to be published, somebody must understand its architecture, code, servers and security.
When does it become freelance work?+
When production no longer needs shared conversation and decisions, but hours of specialised execution. We separate it from the session and quote it independently.
pAIr Sessions · professional service
What do you want to build,
solve or rethink?
Tell us the challenge. We will tell you whether we recommend a session, a cycle, a different commission —or no service at all.
Start with 20 minutes