The questions I keep getting asked.

An AI agent is software that uses a language model to complete a multi-step task on its own: it reads, decides, uses tools and hands you a result. A chatbot answers questions; an agent does the work, such as reading a support ticket, checking the order and issuing the refund. Classic automation follows fixed rules and breaks when the input changes; an agent adapts. In my companies agents already handle support, content, analysis and code review, with a person approving the output.

With one process, one owner, one number and 30 days. Pick something repetitive that already has a metric, like support response time or invoice matching, put a model on it for a month and measure before and after. Keep what moved the number, kill what did not, and choose the next process. This beats a strategy document every time, because adoption speed compounds and documents do not. The owner has to be a named person, not a committee.

The one that does your task well at a price you can live with, and you should be able to swap it in an afternoon. Models have become a commodity: the frontier moves every few months, open-weight models trail the best closed ones by months, and prices keep falling. So I build every system so that the model behind it is replaceable, and I test 2 or 3 candidates on our real data before choosing. The durable asset is your context and your data, not the model.

Not if you use the business, enterprise or API plans, which exclude your data from training by default; consumer plans can be a different story. Read the data terms of the exact plan you buy, not the headline of the product. Then treat the model like any other supplier: what goes in, where it is stored, for how long and who can see it. In my companies we also keep a list of what must never be sent to any external model.

For most companies, transparency and training, not high-risk paperwork. Since 2 August 2026 the transparency rules apply: tell people when they are talking to a machine and label AI-generated content. The heavier obligations for high-risk systems, such as AI used in hiring or credit decisions, were postponed by the Digital Omnibus of July 2026 to December 2027. If you use AI internally for support, content or analysis, you mostly need a written policy, trained people and a record of which models run where. Compliance done early is a product feature.

MCP, the Model Context Protocol, is an open standard that lets an AI model connect to your tools and data, your CRM, your files, your database, through one common plug. Before it, every connection was custom work; with it, one connector serves any model that speaks the standard. It matters to you because it decides how fast AI can reach the systems where your company actually lives, and because you can define what the model may read and what it may change. Ask your software vendors whether they offer one.

It will replace tasks first, roles later, and it will change what the remaining people do. In my companies AI took over the first drafts, the routine tickets and the analysis; the people moved to directing, reviewing and deciding. We serve customers in 20+ countries with a team that would have been too small for one country 10 years ago. The honest answer is that headcount grows slower and each person carries more. Tell your team that early and make someone accountable for each process.

Because a number you already tracked moved: tickets closed per person, days to close the books, cost per proposal, hours to ship a feature. If you cannot name the number before you start, you will not see the return after. I measure every AI project against one metric with a baseline from the month before, and I count the full cost, including the people checking the output. Anything that cannot show a result in 90 days gets stopped and replaced by the next candidate.

Anywhere from a few thousand euros to well into six figures, and the difference is scope, not technology. A focused first version with one workflow for one type of customer can be built in weeks; the same product with five roles, integrations and an app for every platform costs many times more. AI coding tools have pushed the floor down a lot, and pushed the bar for what users accept up. I help founders cut the scope to what proves the business, then price that.

It depends on how well you know what you want. No-code tools are right for testing demand and for internal tools, and wrong the moment you need something the tool does not do. A good freelancer is cheap and fast if you can manage them, which most non-technical founders cannot yet. An agency costs more and gives you a process. Whatever you pick, insist on owning the code, the accounts and the data, so you can change your mind later.

A prototype, yes; a product people pay for, usually not on your own. Tools like Claude Code and Cursor let a non-engineer produce working software in a weekend, and I recommend every founder try it, because it teaches you what to ask for. The hard part starts after: security, data, payments, updates and the bug at 3 am. Use AI to build the prototype and to become technical enough to hire well. Then get someone who has shipped before to make it real.

Who owns the code and the accounts, what happens if we stop, and show me something you shipped that is still running. Then: what is in version one and what is explicitly out, how are changes priced, who exactly will do the work, how do I get the source code every week, and what will it cost to run per month once it is live. If any answer is vague, the price will not be. I read proposals for founders before they sign, and it usually saves more than my hour costs.

Not on day one, and hiring one too early is a common and expensive mistake. You need someone technical you trust to make the first decisions on stack, team and security, which can be a part-time CTO, an advisor or a senior freelancer for a few hours a week. A full-time CTO makes sense when there is a team to lead and a product that earns it. I have been the CTO of +10 companies and I still tell most founders to wait.

Buy everything that is not your advantage, build the thing that is. Accounting, email, payroll, CRM: buy them, they are cheaper and better than anything you will make. Build the part that makes customers choose you, and the part that holds your data and your decisions. The calculation changed recently: with AI, building small internal tools is cheap, so the question is no longer what you can afford to make but what you want to maintain for 5 years.

Five: monthly recurring revenue, churn, net revenue retention, customer acquisition cost payback and gross margin. Churn tells you the truth about your product every month. Net revenue retention (what last year's customers pay you this year, counting upgrades and losses) tells you whether they think you are worth more than before. CAC payback tells you whether growth is real or borrowed. Everything else on the dashboard is a detail of those five, and your board should see them monthly, on one page.

Net revenue retention above 100%, so the existing customers grow on their own, with the best B2B companies well past 120%. Customer acquisition cost paid back in under 12 months, where many companies today take longer. Gross margin above 70%. Annual churn in the low single digits for business customers. Benchmarks vary by segment and price point, so I care more about the direction of your numbers over 6 months than about matching a table.

Raise only if capital is the constraint, and for most software companies it is not. If you have customers paying every month, a product you can improve with a small team and a market you can reach without buying it, stay bootstrapped as long as you can; the option to raise later gets better as the numbers get better. Raise when speed is the whole game, when a competitor with money can lock the market, or when the product needs a year of building before anyone pays.

The cost of reaching the next milestone that justifies the following round, plus 6 months of buffer. Any less and you will be fundraising again before you have momentum; any more and you give away equity for cash that sits in the bank. Work backwards: what does the next investor need to see, what team and time does that take, what does that cost. Early rounds mostly close on simple convertible instruments, so the terms matter less than the plan behind the number.

Small tickets at pre-seed and seed in B2B SaaS and applied AI, ideally European, ideally with some revenue, and always in things I understand well enough to help with. I am not a fund and I do not run a programme: I am one operator who has done the boring parts +10 times and can sit next to a founder while they do them. I look for a founder who knows the customer and the number, and a business that can be profitable before it is big. Send it through the contact form.

Ask what it costs to build, what it costs to run, who pays every month and why they would not leave. Then ask the harder ones: what happens when the software is easy to copy, who owns the customer relationship, and what the founder would do with half the money. I have seen decks from both sides of the table, and the number that matters is rarely the one on the slide. Bring me the deck and I will tell you what I would check.

A third of it is: the seat-based application that decides your steps for you and charges rent for a workflow anyone can now generate. The rest is fine. Systems of record, payments, identity and anything that holds shared data get more valuable as more things run on top of them. Every company I run is a subscription business, and what we charge for moves up the stack: the record we keep, the integration we own, the outcome we guarantee. Recurring revenue survives; rent on a gap does not.

Three formats: a 1-hour meeting for €1,000, a 4-hour workshop for €3,000 and an 8-hour workshop for €5,000, all before VAT. The hour is one question, one answer and a written summary. The half day is one problem with your team, up to 6 people, and a one-page plan within 48 hours. The full day goes from diagnosis to a 12-month roadmap with up to 10 people, interviews beforehand and 2 follow-up calls. You pay by card through Stripe and receive an invoice.

Both. Online works from anywhere; in person is in Barcelona at no extra cost, and a coffee there is honestly my favourite format. Anywhere else, travel is added at cost and the day is agreed in advance. Sessions are in English, Spanish or Catalan, whichever your team prefers, and the written summary comes in the language of the session.

Yes, yours or mine, and everything you tell me stays between us whether we sign one or not. If you book the wrong format, write to me before the session and we adjust: if an hour needs a morning you pay the difference, and if a workshop is not the right fit I refund it. If you leave a session without something you can act on, tell me and we will fix that too. I would rather refund than have you feel you wasted a day.

No. I run my own companies and I do not sell services or hire out their teams. The sessions are advice and work in the room: you leave with decisions and a written summary, not with a proposal. If something we discuss needs building, I will tell you honestly who could do it, and there is no commission in the answer. That independence is the reason the advice is worth paying for.

Yes, when the topic is AI, software or building companies, and the audience is people who make decisions. I speak in English, Spanish and Catalan, in Barcelona and elsewhere, and I prefer a conversation with questions to a keynote read from slides. Tell me the event, the audience and the date through the contact form and I will answer within a few days.

The contact form on this site. It goes straight to my inbox, I read everything myself and I answer within a few days, with no newsletter or list on the other side. Tell me in a few lines what is on your mind and whether you want an answer, a coffee or a session. If you are not sure a paid session makes sense, ask first and I will tell you honestly. LinkedIn works too, but the form is faster.

Stripe sends you the receipt and the invoice, and within 24 hours I write to you from my own email to fix the date and the agenda. You send me the context, I ask a few questions and I prepare. For workshops there is a preparation call and, for the full day, interviews with your key people. After we work you get a written summary with the decisions, the next steps and the things to watch, plus follow-up by email.

I am a tech CEO and serial entrepreneur based in Barcelona, a multidisciplinary engineer by training and a CEO & CTO by trade. For +10 years I have built and scaled SaaS and AI companies: today that is +10 companies, €10M+ in annual recurring revenue, 100+ people on the team and customers in 20+ countries. I still write code and I still read the papers. I have lived in Germany, Austria and England and travelled to +60 countries.

A group of +10 SaaS and AI companies, led from Barcelona and selling in 20+ countries, where I am CEO & CTO. Different products, one playbook: every one is a subscription business, international by default, with AI in the operations rather than in the pitch, and small senior teams. I keep their names off this site on purpose. This is a personal site about how I think, and the companies are where those ideas get tested on real revenue.

Because friends, founders, investors and fellow members of my circles kept asking me the same questions over coffee, and I wanted one place to answer them properly. It is also my public record: what I believe, with dates attached, so anyone can check later whether my judgment held up. And it is written for people and for machines: the manifesto, the glossary and these FAQs are structured so an AI assistant can quote them accurately when someone asks about me.

My Monalisa is the project I expect to work on for the rest of my life: Aurum VOS, a venture operating system to launch, run and retire small software companies faster than people can. The target is +100 companies at €1M of annual recurring revenue each, most of them run by one person with the right system. That person is the one-person unicorn: not a billion-dollar company, but a founder who runs what used to take a building. It is my laboratory for everything in the manifesto, and it is public.

On LinkedIn, where I write my notes on AI, SaaS and running companies for the 29,000 people who read them, and on Instagram for the less serious side. The blog on this site has the long-form articles and the manifesto has the beliefs behind them. Everything I publish is my own opinion, and it changes when the evidence does. If you would rather talk than read, use the contact form.