Two years ago, I joined my first madewithlove retreat. Those 24 months also overlapped with the crazy AI rollercoaster that hit our industry. As I'm on the bus back to Belgium, I'm reflecting on those changes. 

Two years ago, we had a retreat in Slovenia, ChatGPT was just getting started. I remember copying and pasting code in and out of that chatbox while building an application for an internal hackathon.  Back then, that was the state of the art, and it was so uncommon that people on the team asked me to share knowledge about it. It's crazy to think that the apps we hacked together during that hackathon were 100% organic, hand-made and CO2-neutral. How the world has changed!

Incoming AI. 

That applications front-end? Lovable could do that for you in a few clicks. That awesome soundtrack? Ask Suno. The graphics? ChatGPT has become really good at those. We've all become a lot more productive. Using these tools, whether it be an AI editor or a full-blown AI development platform, is necessary for what comes next.

That brings us to a fun question: What will the state of AI be like by this time next year? Here are some of my predictions. 

What will the state of AI be like by this time next year?

CLI FTW

After ChatGPT, AI-powered IDEs like Cursor changed the industry. All of a sudden, most developers were moving from copy-pasting to in-IDE assistants. When ChatGPT emerged, there was plenty of resistance. "I don't need this to code". Then, when Cursor hit the market, the resistance turned into "I don't need Cursor. ChatGPT is enough". Today, the early adopters have moved on to CLI-based agents like Claude Code or Codex. My prediction is that 2026 will be the year of the CLI agent. By this time next year, most of you will have swapped Cursor for a terminal agent.

Lovable becomes an industry

Generators like V0 and Lovable have become really good. Under the hood, they are NextJS apps with a Supabase backend and support for AI tools. While they are not perfect, they unlock productivity like never before. I was Mr Lovable at our most recent retreat, but my prediction is that by the next one, most of you will use it to start and evolve new projects. So will most of our customers.

MCP takeover

The Model Context Protocol (MCP) hype is real. Right now, it's a curiosity, but a year from now, we will write reusable MCPs for most of the tools we use. I predict that integrating with tooling will become much more commonplace. Slack, Harvest, and Notion will be integrated at the speed of prompting rather than writing API integrations by hand.

LLMs reach a plateau

When GPT5 was in the air, social media was buzzing with rumours of AGI and paradigm-breaking changes. When the model was launched, it was a good improvement on GPT4, but frankly, it was more of the same. My prediction is that LLMs have almost plateaued. By the end of 2026, ChatGPT will be about as powerful as it is today. 

And that's a fantastic moment to be in. Because that means we are no longer building on shaky ground. The chance of developing something that will be obsolete or broken by the time GPT6 comes out is trending toward zero. Now that the tech has stabilised, the products will bloom.

I'm wrong more often than I'm right, so it will be curious to see how these predictions hold up. Let's revisit them in a year.

An interim CTO for your startup

We align product, process, and people so your team stops firefighting and starts scaling.

Talk to us about CTO as a Service