steffen.ing
Writing

My beliefs on where Software/SaaS/AI SaaS is heading (July 2026)

2 min read

From what we developed at Nukipa Labs, I believe the following will happen around software and SaaS.

1. Most things will become MCP servers

Building intelligence into your own product is way too expensive. The big labs want you to use their agents, not their APIs — hence something like a 20x difference in token cost between agent and API.

We could not build a proper product with the latest models included. So I believe most software will become MCPish.

2. Vibe coders are gaussian distributed in every industry

In every industry you find three types of people:

  • One end: the people who build their own software with Claude and the rest.
  • Other end: the people who do not even use agents yet.
  • Middle: the people who have an agent, but no real clue how to put everything together.

I think it is important to build for the middle. They already put things into ChatGPT or Claude, but do not have the time, the will or the knowledge to understand how context management, MCP servers and the rest actually work.

They will be happy to pay for something that just makes “Claude do XYZ”.

3. Moats I see

There are fewer moats in software businesses, but they still exist.

  • Brand & GTM. Getting to an existing customer base faster than newcomers.
  • Unique data. Data points that have not been trained into the base LLMs and cannot be easily scraped. Proprietary data, or data collected over time — a Facebook scrape from 2015 is not something anybody will replicate.
  • Deterministic behaviour. The more stable and predictable your product is at helping agents do something, the more likely people keep using it.
  • Speed of shipping. Are you able to ship at near real time?
  • Customer context. Are you able to get your customer and industry context much better than your competitors? I think customer context also gives you the biggest lock-in into the software you are selling.

4. Frontends — I don’t know

I am not sure yet how, or if, we will need a lot of frontends in the future. I think most will be done via MCPs and agent interfaces.

But certain actions will be faster in a normal, classical interface. Quick yes/no questions. Short navigation on a map.

It might also be that frontends just build up on the fly, so that every user of your product has a self-built one. I actually believe that scenario is highly likely.

These things are likely already outdated by the end of 2026. But you need to make some predictions to at least not build totally into the blind.