Anchor: building AI for alienated fathers

You are sitting at a kitchen table looking at a stack of printed emails, court summons, and letters from solicitors. The house is quiet. Your children are somewhere else, and you do not know when you will see them next. I am lucky enough not to know this exact feeling first-hand, but I have spoken to enough men who do. The overwhelming theme is not anger, though that is often present. It is isolation. The shame of being an alienated father is heavy, and the administrative burden of fighting for contact is crushing. You are expected to be highly organised, perfectly rational, and legally precise at the exact moment your life is falling apart. Most people cannot do it.

The midnight shift

When we started working on Anchor, I had to look closely at how family courts actually function. The reality is deeply bureaucratic. Judges, social workers, and mediators deal with evidence, timelines, and communication logs. They do not have time for your emotional truth; they want facts. If a father wants to maintain or regain contact, he needs to submit clear, unemotional records. He needs to reply to hostile, sometimes provoking messages with boring, factual text. Human beings are terrible at this when they are hurting. If you draft a reply to your ex-partner at 2am, it will probably be defensive. If you try to organise three years of WhatsApp messages and missed calls into a coherent timeline for a judge, you will quickly drown in the details.

We are building a system that stays awake, stays calm, and strips the emotion out of a terrifying process.

This is where artificial intelligence is genuinely useful. We are building Anchor to do the structuring. A man can dump a chaotic, angry draft into the system, and the AI will rewrite it to be polite, brief, and focused entirely on the logistics of the children. It acts as a filter for your worst impulses. It can also pull dates and facts from hundreds of unstructured documents to build a clean timeline for a solicitor. More importantly, it is available when everyone else has gone to sleep. Friends and family are crucial, but they eventually get tired of talking about your court case. Solicitors charge by the hour and do not answer the phone on a Sunday night. Anchor is simply there to act as a barrier between a father's late-night panic and the permanent legal record he needs to maintain.

What we refuse to do

I have spent a lot of time looking closely at AI infrastructure with my other company, Substrat. If there is one thing I know, it is that large language models hallucinate. They confidently invent facts. Because of this, we had to set strict boundaries for Anchor early on. The system does not give legal advice. If a user asks what their chances are in front of a judge, or what a specific statute means, the software stops and tells them to speak to a qualified professional. It is incredibly dangerous to let an algorithm play solicitor with someone's family. We are dealing with real children and real custody agreements, not a software subscription. If an AI gets a line of code wrong in a tech startup, a website crashes. If it gets legal advice wrong here, a man might lose access to his kids forever.

We are also careful not to present this as a replacement for human connection. I have kept a written life plan since around 2010, updating it every year and rating my family, health, work, and personal time out of ten. I like structure. But men going through family court usually have their structure completely destroyed. They need real therapists, support groups, and good friends. I play tennis with a few fathers here in Marbella, and the value of just standing on a court, hitting the ball badly, and talking openly about life is high. No software can replicate that. Anchor is a machine. It does not feel empathy. It simulates a calm response based on training data. That distinction matters deeply to us. We are not building a synthetic friend to keep lonely men company. We are building a utility to help them navigate a legal system that is hostile to chaos.

The work ahead

Building software for people in crisis is slow, deliberate work. When Fredrik and I ran Mediapilot back in Stockholm, we were selling B2B services. We grew it over roughly seven years, created over a hundred jobs, and the stakes were ultimately commercial. A bad month meant lost revenue, maybe some hard conversations, but life went on. At Anchor, a bad feature release could genuinely harm someone's custody case. We test constantly, limit the feature set deliberately, and review exactly how the tool handles distressed prompts. I haven't got all the answers on how to perfectly integrate AI into family law, and we will probably make mistakes along the way. But when I look at the current alternative, which is thousands of fathers sitting alone in the dark trying to draft legal documents through tears, I know this is worth trying to get right.