Barcelona Code School
Career guide · August 2026
Five routes into AI work, and the one decision that separates people who get paid from people who keep learning: knowing when a rule is enough, when a model helps, and when a system really needs an agent.
Start here
There is no single way to work in AI. You might use AI to become better at the job you already have, automate business processes, build AI applications in code, work with models, or lead AI projects inside a company. The right route depends on the work you want to do — not on the most impressive job title.
How to use this guide. Read the five routes, choose the one that sounds closest to the work you want, and follow the three learning stages. You can change direction later. The goal now is simply to make a good first choice.
Section 2
Three shifts that decide your trajectory
Generating code, copy and first-draft analysis is now fast and inexpensive. What stayed expensive: deciding what to build, connecting it to real business data, proving it works, and being accountable when the output is wrong. Careers are moving closer to specification, integration, evaluation and verification than to production.
AI can now handle some of the routine work that used to be given to junior employees. Junior roles still exist, but employers may expect stronger basics and clearer proof of practical ability. A small working project can help, but it does not replace technical foundations, professional experience or a good interview.
Businesses rarely shop for people who "know AI tools." They have a specific pain: work that eats hours and produces errors. The person who can map that process, decide which parts should stay deterministic and which benefit from a model, build it, and keep it reliable is the one who gets paid — whether their title says engineer, operations, marketing or consultant.
Section 3
These routes lead to different kinds of work. For each one, look at four things: what you will do, what you need to learn, which tools you may use, and what you should be able to show at the end. The table gives you the whole map at a glance; the cards after it give the detail.
| Route | Best for | Coding | First result | You end up |
|---|---|---|---|---|
| 01 — AI-augmented professional | Anyone who wants to stay in their profession and work faster | None | Days | More valuable in the job you already have |
| 02 — AI agent & automation engineering | Career changers, operations people, founders, consultants | None to start; technical literacy to get paid | 2–3 months | Building AI systems for businesses |
| 03 — Agent developer / AI application engineer | Developers, and people ready to learn programming properly | Yes | Months | Inside an engineering team |
| 04 — ML / AI engineer | People who like mathematics, data and experiments | Yes | Longest | Working with data and models |
| 05 — AI lead inside a business | People who already lead teams, products or budgets | None, but hands-on literacy required | Weeks, after one build of your own | Deciding what gets built and owning the result |
A useful rule: choose the simplest approach that solves the problem. Agents are more flexible, but they are also less predictable, more expensive and harder to test.
Route 01
You keep your current role and rebuild how you work.
What you doUse AI for research, reports, content, email, client communication and other recurring work. Your role stays the same, but you can do more with less routine.
What to learnHow to give clear context, organise your knowledge, connect everyday apps and turn a repeated task into a simple workflow.
Possible toolsClaude or ChatGPT, Notion or Google Drive, Calendar and email connectors, and simple automation tools.
Your resultA personal AI workspace and one repeated task that now takes less time. Measure the hours you save.
Not for you if you want a new job title in AI. This route increases your value inside your current role; it doesn't by itself make you hireable as an AI specialist.
Route 02
You design and build reliable AI-powered business systems.
What you doTurn real business processes into working systems. You connect apps and data, decide where AI is useful, add human approval where needed, and make sure the system works reliably.
What to learnProcess mapping, workflow logic, how apps exchange data, basic APIs and JSON, knowledge bases, testing, permissions and error handling.
Possible toolsn8n or another automation platform, Airtable or Supabase, AI models, APIs, Telegram or other business apps.
Your resultA complete business workflow you can demonstrate from start to finish, together with evidence that it saves time or improves quality.
Job titles vary. You may see AI Automation Specialist, Automation Consultant, AI Solutions Builder or n8n Developer. You can begin without programming, but paid work usually requires comfort with APIs, JSON, data and technical troubleshooting.
Route 03
You build agentic applications in code and own the software around them.
What you doBuild AI applications in code and take responsibility for the software around them: data, authentication, tests, deployment, performance and cost.
What to learnProgramming, APIs, databases, application architecture, connecting AI to reliable sources, testing, deployment and monitoring.
Possible toolsPython, LangGraph or a similar framework, a database, GitHub, cloud hosting and evaluation tools.
Your resultA deployed AI application that people can use, with tests, clear limits and a way to see what happens when something fails.
Before choosing this route: you need programming foundations. Experience with route 02 can help you understand AI workflows, but it does not replace learning Python, databases, testing and software engineering.
Route 04
You build and improve systems based on data and machine-learning models.
What you doWork with data and models: prepare data, train and compare models, evaluate their quality and put them into production.
What to learnPython, statistics, machine learning, neural networks, data pipelines, model evaluation and deployment.
Possible toolsPython, SQL and a data warehouse, PyTorch or TensorFlow, Hugging Face, notebooks, experiment tracking and MLOps tools.
Your resultA model you trained, evaluated honestly and deployed, with a clear explanation of the data and the choices you made.
This is usually the longest route. It requires programming, statistics and sustained practice with data. Not every ML engineer trains large models; many adapt, evaluate, deploy and maintain existing ones.
Route 05
You decide what should be automated, who builds it, and whether it worked.
What you doChoose where AI can create value, set requirements and quality rules, select tools or vendors, manage risk and own the business result.
What to learnProcess mapping, basic AI system design, evaluation, privacy, cost, risk and how to recognise a strong implementation.
Possible toolsProcess maps, simple prototypes, evaluation scorecards, cost models and the tools used by routes 01 or 02.
Your resultOne improved process with clear before-and-after numbers: time, cost, errors or speed.
This is usually a destination, not a first AI job. It grows from leadership and industry experience. Build practical AI literacy through route 01 or 02, then apply it inside a business you already understand.
If some terms above were new, here is the short version. None of them are as complicated as they sound.
| Term | What it actually means |
|---|---|
| API | The way one app lets another app ask it for something or tell it to do something — how your form talks to your CRM without a human copying data across. |
| JSON | A simple text format apps use to pass data to each other: name, email, budget, each clearly labelled. You read and edit it, you don't program it. |
| Deterministic / rule-based | The same input always produces the same result, because you wrote the rule. The opposite of asking a model and hoping. |
| Knowledge base & retrieval | Your documents, stored so the system can find the relevant piece and answer from it instead of inventing an answer. |
| Structured output | Making the model return tidy fields — name, score, next action — instead of a paragraph, so the next step can use them. |
| n8n | A visual tool where you connect apps and steps into a working process. You build by dragging and configuring, not by writing an application. |
| Baseline (route 04) | The simplest solution you compare everything against. If your clever model can't beat it, it isn't clever. |
| Deployment | Moving something from your laptop to a place where it runs on its own and other people can use it. |
Section 4
Start with your current experience and the kind of work you want. You may fit more than one route; choose the best starting point, not a permanent identity.
| If this sounds like you… | Start here | Check before committing |
|---|---|---|
| I want to use AI in the work I already do. | Route 01 | Choose one repeated task you can improve now. |
| I enjoy processes and want to build business automations without becoming a software developer first. | Route 02 | Are you willing to learn APIs, JSON, data and debugging? |
| I already code, or genuinely want to become a software developer. | Route 03 | Are you ready to learn software engineering as well as AI? |
| I enjoy mathematics, experiments and working carefully with data. | Route 04 | Do you have time for the longest technical foundation? |
| I already lead people, products, operations or budgets. | Route 05 — supported by 01 or 02 | Which business result will you own? |
Got your route? Get the one-page start. Email office@barcelonacodeschool.com with the number of your route and we will send the one-page first-week checklist for it: what to build, what to avoid, and what "done" looks like. No call required.
A test that beats any quiz. Imagine spending a normal working day on this route. Would you rather improve your own work, map a business process, write and test application code, experiment with data and models, or make decisions about value and risk? Choose the work you want to repeat, not the title you want to announce.
Section 5
Not every route needs the same technical depth. Start with four skills that help everyone use AI responsibly. Add the next five if you build systems for other people.
| Skill | You can do this when… |
|---|---|
| Give useful context | You explain the goal, relevant background, limits and desired result clearly. |
| Check the answer | You verify important facts instead of treating confident language as proof. |
| Protect information | You know what data should not be shared and which actions still need human approval. |
| Judge the value | You can say what improved: time, quality, cost, speed or user experience. |
| Skill | You can do this when… |
|---|---|
| Connect tools and data | You understand where information comes from, where it goes and what format the next step needs. |
| Choose the right approach | You can explain why a step needs a rule, an AI model or an agent. |
| Test and improve | You try normal, difficult and incorrect inputs and record where the system fails. |
| Make failures visible | You can see which step failed, what it received and what should happen next. |
| Keep the system safe to run | You treat incoming text as untrusted — an email or document can contain instructions aimed at your system, so the tools it can reach are limited and risky actions still need a person. You also know which data must stay inside the company, and what your client's obligations are under EU data-protection rules (GDPR). |
One more thing clients ask about early: money and speed. Before you propose a system, you should be able to say roughly what one run costs in model and platform fees, how long it takes, and what happens to both if the volume grows ten times. A workflow that costs more than the work it replaces is not an automation — it is an expensive demo. Tools will change. The ability to understand the problem, check the result and build something reliable will transfer to the next platform.
Section 6
Every route moves from foundation to practice to proof, but the work inside each stage is different. Move forward when you have the result — not because a certain number of days has passed. The time ranges below assume 6–10 focused hours a week and are there for planning, not as a promise: routes 03 and 04 take considerably longer if you are starting from zero.
You are ready for stage 2 when: you can complete one small exercise without copying every step, and explain why you made each choice.
You are ready for stage 3 when: another person can use or evaluate the work, and you know its limits, cost and main failure cases.
This stage is complete when: you have relevant proof for your route and can explain both what worked and what did not. Ask an employer, client, user or experienced practitioner for specific feedback.
A useful checkpoint. Ask yourself: What works now that did not work before? What broke, and why? What can I show another person? If you cannot point to something concrete, stay with the current stage.
Section 7
A project makes your skills easier to assess, but it is only one part of the picture. Employers and clients may also consider your foundations, previous experience, communication and interview performance. Strong proof usually includes:
Instead of several disconnected demos, you can build one connected business system: automations, AI steps and, where useful, agents working with shared data. Barcelona Code School calls this project a Virtual AI Department. Its strength is not the number of agents, but the quality of the decisions:
A second AI model checking the first is not enough. A stronger project combines several simple checks:
"I tested it" means nothing until you can show the cases. For a lead-qualification step, a first test set is about ten saved examples with the answer you expect:
For each case you write down what the system should return, then run all ten after every change. Now "it works" has a number behind it — nine out of ten, and you know which one fails and why. That single sheet is often the difference between a demo and a system someone will pay for.
Two paths, and they behave differently. Employment — you join a company in an automation, operations or engineering role, and your project plus your previous experience are what get you the interview. Salaries vary so much by country, industry and seniority that any number printed here would mislead you; check the actual openings in your city for the route you picked, and treat that as your real market data.
Independent work is usually where the first invoice appears, because it starts small: one process, fixed scope, one owner who feels the pain. The pricing logic is worth learning early, because it is the same logic that gets you hired:
A process eats 20 hours a month. At an internal cost of €25 an hour, that is €500 a month leaving the business. A system that removes 70% of it returns roughly €350 a month — about €4,000 a year. Against that, a one-off build priced at two to three months of the return is an easy decision for the buyer and real money for you. Then add the honest costs: your build time, model and platform fees, and the maintenance nobody remembers to budget for. If the numbers don't work at this scale, they won't work at any other.
Notice what makes this conversation possible: the before/after numbers from stage 3. Without them you are negotiating about your time. With them you are negotiating about their money.
Section 8
You can learn any of these routes independently through documentation, tutorials and your own projects. Plenty of people do. It is worth knowing where self-teaching usually breaks, so you can plan around it either way:
What a taught programme buys is a sequence, a group, a deadline and feedback from someone who has already broken these systems. What it cannot buy is your domain knowledge, your discipline and the conversations you have afterwards. Those stay yours either way.
How the teaching actually runs at Barcelona Code School. Mornings are build-along: the instructor builds a working example live while you build beside them. Afternoons you get a business task and build your own version, with one-on-one help when it breaks — and tasks come at three levels, so nobody is left behind or bored. That is the part that is hard to reproduce alone.
Bootcamps since 2015 · 500+ graduates · groups of 6 · on campus or remote live · taught in English · certificate on completion · career support with portfolio, CV and interview preparation · Google Reviews 4.9, SwitchUp 4.8, CourseReport 4.8
| If your route is… | Programme | Format | Tuition |
|---|---|---|---|
| 01 — augment your current role | Work Automation with AI Agents | 8 hours — one-day workshop or 3 evening sessions. On campus or remote live. No coding. | 490€ |
| 02 — AI agent & automation engineering | AI Agent & Automation Engineering Bootcamp | 4 weeks full-time, 160 hours, groups of 6. Learn process mapping, automation, AI workflows, integrations, shared data, testing and monitoring. Build a Virtual AI Department using n8n. You can start without programming experience, but the work includes APIs, JSON, data and technical troubleshooting. | 3,600€ — 2,900€ early bird when you book at least 90 days ahead (in practice, the 2027 cohorts) |
| 03 — agentic applications in code | No single programme here is the whole answer | If you already code, the route 02 bootcamp adds the AI-system layer — workflows, tools, retrieval, evaluation — on top of what you have. If you don't, start with a proper programming path first; nothing on this page replaces Python, databases, testing, architecture and deployment. | — |
| 04 — models and ML | Machine Learning & AI Engineering | 13 weeks, part-time, mentored by an AI researcher. | 3,950€ |
All courses run on campus in Barcelona or remote live, in English.
Choose by the result you need. Route 01 may only require a short practical workshop. The Route 02 bootcamp is designed directly for business automation systems. For Route 03 it covers useful AI-system concepts, but you will still need a substantial programming path. Route 04 needs structured work with mathematics, data and models. Route 05 combines hands-on AI literacy with leadership experience.
| Programme | Starts | Format |
|---|---|---|
| Work Automation with AI Agents — route 01 | 29 August 2026 | One-day workshop, 9:00–18:00 |
| 15–17 September 2026 | Three evenings, 18:30–21:10 | |
| 20–22 October and 24 October 2026 | Three evenings, or one-day workshop | |
| AI Agent & Automation Engineering Bootcamp — route 02 | 7 September – 2 October 2026 | 4 weeks full-time, groups of 6, on campus or remote live |
| 5–30 October 2026 | ||
| 2–27 November 2026 |
Not sure which route is yours? Write two lines — the work you want to do and the experience you already have — to office@barcelonacodeschool.com, book a call at zcal.co/i/c9mEbhSA, or phone +34 936 639 807. Campus: Carrer de Paris 157, Barcelona.
You can begin routes 01 and 02 without programming. Professional work in route 02 still requires technical literacy: APIs, JSON, data, authentication and debugging. Routes 03 and 04 require programming. Route 05 requires enough practical understanding to evaluate the people and systems you lead.
Route 02 builds business processes mainly with automation platforms, models, APIs and existing services. Route 03 builds applications in code and owns the software architecture, testing, deployment and infrastructure. Some concepts transfer, but route 03 needs a much deeper programming foundation.
No, but the market is uneven. Demand depends on the country, industry, role and your previous experience. A focused route, strong foundations and relevant proof give you a better position than broad “AI knowledge”.
Some routine tasks and entry-level work are changing. At the same time, companies still need people who understand the domain, define the problem, check the result and take responsibility. No route is future-proof, so keep building both AI skills and durable professional judgment.
It depends on the route, your starting point and the time you can give it. Use the three stages as your guide: first make one task work, then make it reliable, then turn it into a real case.
The routes connect, but the transition is not automatic. Route 01 can grow into 02 or support 05. Route 02 gives useful context for 03, but programming and software engineering still need to be learned properly.