How do you build a software company at a time when coding is faster than ever?
AI is an output amplifier, churning out code much faster than any human can. I code more than I did in the last 10 years combined, but I write barely any of it.
Where does that leave us? How do we position ourselves now that creating code is easy and fast?
I've been chewing on that question for a long time now.
Avoiding AI was never the way
Only a year ago (summer 2025), I still treated AI like a fad to wait out. I got the hype, but never really understood what’s so great about a chatbot that hallucinates about strawberries.
That started to change when a client approached us with an app idea, and at that time, I was dabbling in Claude Code. Well, I had the prototype ready before that client even got around to replying to our proposal. Nowadays, I code more than I did in the last 10 years combined, but I write barely any of it.
AI is not going away. Pretending otherwise has a cost, and it's not ours to pay; it's the client's. Every sprint we spend writing boilerplate by hand that a model could draft in seconds is a sprint the client is paying for at senior-engineer rates.
So yes, we use AI. Deliberately. It's a core part of how we build now.
There’s always a catch
AI is an output amplifier. It sounds like a good thing, and it usually is. A senior engineer with the right tools ships in days what used to take weeks.
Except an amplifier doesn't care what it's amplifying – it just churns out code. Bad architecture decisions now happen much faster and subtle bugs get copied across a codebase.
Usually, that’s fine, because you can just fix the mistakes later. The important thing is that it sort of works and it’s ready fast.
Usually, but not always. There are industries where small mistakes mean large and costly repercussions.
We already knew this, because of who we build for
Our best work, the projects I'm proudest of, were precisely like that – the ones with the most to lose.
Mobile banking apps and Cloud bank transformations where a mishandled transaction means money disappearing and a regulator asking questions. Clinical trials software and trial-patient matching platforms where the data belongs to a patient who never consented to it leaking. Apps controlling heavy machines, where a bug means an injury, or worse.
Panaxeo has spent years building for PCI-DSS and PSD2 environments, for HIPAA and FDA 21 CFR Part 11 audits, for ICH GCP clinical trial standards, holding ISO 27001 ourselves as a team. That's just who kept hiring us.
High-risk, high-regulation, low-tolerance-for-error work forces a level of rigor that most software never has to earn. You can't ship "mostly right." The cost of that is just too high.
Which is what AI-assisted development needs
AI makes us faster, but it needs guidance.
We’ve been learning to build regulated software for a decade, and it's suddenly the most relevant skill for us because of AI.
Panaxeans use AI to move fast. But they validate the output, they own the result, and they don't ship anything they can't defend in an audit.
That's the whole positioning, really. We build software that must be right when the cost of being wrong is too high.
Where this leaves us
We're the team you call when the code has to survive an audit, a regulator, or a piece of heavy equipment operating exactly as intended. When “it’s mostly right” isn’t good enough, because being wrong just costs too much.
If that's the problem you have, let's talk.
