How I actually use AI every day

If you live in Spain as a foreigner, you eventually develop a specific kind of dread. It arrives in the form of a registered letter from the local tax authority or town hall. My family and I moved from Sweden to Marbella in 2018. We originally meant to stay two years, but we are still here, which means I still get these letters delivered to our house. Before, I would stare at the dense legal Spanish, type it awkwardly into Google Translate, and panic because the output suggested I had committed a serious crime and owed a fortune. Now, I just point my phone camera at the paper and ask my AI assistant a simple question: "Do I owe them money, or do they just want me to confirm my address?" Usually, it is the latter. That is where this technology actually lives in my daily routine. It is a very practical tool for removing friction. I spend my days building Substrat, an infrastructure project for AI builders, and Anchor, a platform meant to help alienated fathers. Both require navigating massive amounts of information. I do not use the software to do my thinking for me. Let me explain. I use it to clear the brush so I can see what I am actually supposed to be thinking about.

The first reader

Whenever I get a fifty-page technical document or a long commercial contract, I no longer read it line by line on the first pass. I upload the whole thing and ask the model to tell me where the traps are. I ask it to summarise the obligations, flag any unusual clauses, and point out what is missing. It acts as an incredibly fast, slightly pedantic junior assistant. Back when Fredrik Bage and I were running Mediapilot in Stockholm, I spent late nights squinting at paperwork, terrified I was missing a detail that would cost us dearly. We grew that business over seven years and created more than a hundred jobs, but I am fairly certain I lost a year of sleep purely to administrative paranoia. Later, when I was working on the commercial side at Happeo, I had to digest complex enterprise agreements on tight deadlines. Having a machine do the initial sweep today does not mean I sign blindly. It means I start reading the document with a map, already knowing which paragraphs need my actual attention.

Arguing against myself

I also use it constantly as a sparring partner. When I have a thesis for a new feature for Anchor, or an angle on a small angel investment I am considering, I will write out my rationale. I dump all my thoughts into a prompt, including the messy, half-formed ideas. Then I give it a single instruction: "Tell me why this will fail." I ask it to find the holes in my logic, to point out the assumptions I am making, and to argue against me as aggressively as possible. I want the pessimistic view. I used to rely entirely on friends or colleagues for this sort of feedback. The problem is that human beings naturally want to be polite. Even the most direct Swedish colleague will soften the blow if they think an idea is stupid. Software has no such filter. Just last week, I pitched the model an idea about structuring a specific data set for Substrat. Within seconds, it outlined three distinct reasons why the approach would break down at scale. Two of the reasons were nonsense, but the third was a fatal flaw I had completely missed. It saved me weeks of building in the wrong direction.

The machine does not care about my ego. It will happily tell me my logic is flawed, which is exactly what I need before I take an idea into the real world.

The hard boundary

But there is a line I will not cross. I categorically refuse to use AI for anything that involves building trust or delivering bad news. If I have to tell a partner that a deadline is slipping, or if I need to have a difficult conversation with someone I work with, I write every single word myself. Better yet, I pick up the phone. I have noticed a trend recently where people use software to draft apologies or difficult emails, smoothing out the rough edges so they do not have to feel the discomfort of the situation. I think that is a massive mistake. The discomfort is the entire point. When I sit down to review my written life plan, which I have done every year since around 2010, I rate my family and friends, my health, my finances, my work, and my personal time out of ten. If the scores are low because I have been working too much and neglecting my kids, or playing tennis badly and ignoring my joints, those failures are entirely my own. You cannot outsource accountability. A machine can translate a tax bill or find a flaw in a business model, but it cannot take responsibility for a failure.

I'll be honest with you, I haven't got this figured out perfectly. I still catch myself sometimes treating the prompt window like a search engine out of pure habit, rather than talking to it like a reasoning engine. I still waste time asking it questions I should just answer myself. I am sure in five years my current workflow will look entirely primitive. Right now, though, I am just happy I no longer break into a cold sweat when the Spanish postman knocks on my door.