Don’t Panic.

Yes, you’ve heard it everywhere - software’s dead (since 1999, according to Marc Benioff) - now it’s even deader. The three-headed culprit: vibe-coded startups that re-built your software in a weekend, the junior financial analyst Claude CoWork, and your users who are now supposedly all software engineers writing their own tools.

Yeah, sure.

Don’t get me wrong: We’re witnessing and co-creating the biggest tech revolution since the internet. Things are changing fast, and the old B2B SaaS playbook no longer works.

The real reason why today’s successful software companies won’t do well in future markets (which currently arrive ca. every 3 weeks) isn’t that they haven’t bolted on an AI chatbot yet. It’s not the competition from outside; it’s how long it’s been since they last shipped something that genuinely surprised their users.

The latest trends (super-fast builds and a new way to think about products entirely) simply magnify these problems.

I’ve built two new products in the last couple of months, so let’s look at one of them as an example of how it went.

The first thing I did was to learn about the users’ pain points - I was lucky enough to have been working with them for a while, so I knew where they struggled and what drives them. We collected five ideas or so and picked one to focus on (the most troublesome), which was generating business reports and the necessary data collection.

I built a first, fully functional version of an agent with chat input in a couple of days (not weeks), deployed it, and let them use it - while making sure to have my observability tools in place so I could collect all their input and trace each step the agent took.

Based on all that data, I could now fine-tune the agent iteratively, mostly via prompting and some context engineering, so it could finally answer their questions consistently.

Notice anything revolutionary about this process? Neither do I.

On the tactical level, yes, things changed - new tech, new user experience and interfaces, and of course, software development is now a very different game that requires a different developer mindset and adapted processes. There’s a bunch of new metrics, and critically, you now need your input and real-life data to fine-tune anything that uses LLMs under the hood.

Yes, the competition is real, the pressure from three sides is real (remember from above: startups as challengers, LLM-based chatbots, self-built software), and the tech is here to stay. It’s time to act.

The companies that’ll lose aren’t the ones without some AI feature crammed into their product or with the least elaborate team of coding agents, but the ones that forgot how to build great products in the first place.