👋

For somebody who started writing his first lines of code at least 25 years ago, now (post-summer 2026) feels like a new era has arrived. Non-coders might experience a glimpse of the current AI capabilities, but coders see them fully in effect with the agentic loop closed: planning, scoping, coding, designing, debugging, reviewing. The AI agents are relentless to execute your prompt instructions.

Today the frontier models have replaced all my previous meticulously crafted manual code writing activity. Which is surprising given I actually enjoy my time coding. Coding feels like a craft of writing efficient functions, keeping the big picture while also being focused on a minor detail. It feels productive. But seeing the agents perform, shows me that era is ending soon. A human crafting code is "not an economically productive enterprise" as DHH stated in his keynote.

I'm no longer even questioning if I should spend time writing lines of code. I prompt several agents to release features for Mailbook that have been ideas on the shelf for 5 years. Instead of taking into account how much time a certain feature would take, I can fast-forward into actually testing the feature in production, with my own audience. It is great (and frightening)!

I am no longer emotionally invested in the functions and features that is produced. Previously taking stuff out that took me weeks of work? Please no. But now I just let the agent strip it out. I'll add it back later (if ever) with a few good prompts.

Building a new web game idea? Sure, why not. But why not 15 games, or a whole platform hosting these web games? Allow anybody to submit games. Game play statistics. Parental controls. Profile for each family member. Ad-free. Crazy, no? Done!

From the CTO's point of view these capabilities are even more impactful. Arguing with the PM about which features to do in Q4 - just do 'em all. Limiting the scope of a promising feature - launch it full-featured. Using a third-party package - have it build a perfectly tailored version from scratch.

Every previously hard-earned lesson on how to ship quality software is being turned upside down. Focus has to shift to (agentic) testing, making the AI align with the user's goals, preventing the AI from going down the wrong path, pointing out alternative directions, etc. On the other hand, we gain fun new capabilities like having the AI build two (three, four) versions of a solution to compare.

What previously held me back from going all in with agents taking over my codebase was the idea of being reliant on OpenAI or Anthropic. But with DeepSeek offering almost the same quality, there will always be a fallback. I don't need to worry that humans will end up having to figure out the stuff agents wrote. Only agents need to efficiently understand the code agents wrote.

What does it all mean?

In 2009 I wrote the post "In the near future everybody develops" (read it!). Although I failed to see the distinction between "will" and "could", the current AI models now allow anybody to craft a software solution. The technical bottlenecks will evaporate with the AI not only coding, but also offering a complete production-ready release to anybody who prompts.

The effect is that software and its features themselves will become commoditized. The implication is that as a software business the emphasis is more on understanding problems and solutions, gaining trust in relationships, growing a brand and investing in sales. Which is an uncomfortable conclusion for someone who prefers tech over sales any day of the week (literally).

But I see the enormous potential. Accepting that code is a commodity forces me to invest in offering a different type of value to clients. First, with vastgoed.ai the move is away from a one-size-fits-all SaaS model to one where customers get bespoke, tailored features. Next is positioning the platform as an AI harness for real estate, so agents do the work not only on our platform, but also inside the systems our customers already work in. Initially for the Netherlands, but rolling out to multiple countries. Something previously deemed too massive an undertaking, now a fun weekend exercise for the agents.

This post is a commitment to my challenge of not writing a single line of code manually anymore. It will force my attention to more productive and impactful parts of building a business.