Here I am going to write down how recent advances in AI influence the creation of new startup ideas.

This blog post is written by me, Nikolai Tschacher, in August 2026 completely alone. I will not even use AI to grammar check or to improve my non mother tongue English. Spell check is being done completely manually, risking that my readers will feel like Batrick Bateman during the business card scene in the movie American Psycho.

On a more brigth side, writing something slithly wrong gives us humans a fuzzy warm feeling and let's us relate more compared to reading professional and "just too good" AI ramblings.

Introduction

When I started working on my business ipapi.is, I did everything on my own: I programmed ipapi.is from scratch without the help of any LLM or AI tools. I guess I started slowly somewhere around 2020 or 2021 with the first ingredients that would later become the IP API that I have now.

ipapi.is works quite fine and it gives me immense freedom, financially and also time wise. I got lucky here, but I also worked extremely hard on it. To some extent it's also a copy of ipinfo.io, and Ben Dowling seems to be the one that came up with the concept of putting IP address meta data into a simple JSON API. So thanks to him as well.

But I digress.

I first started using ChatGPT around November 2022 and even back then it truly shocked me. I first realized that this chatbot is not a stupid thing such as Eliza. When prompted with inputs such as "Please write me a funny short poem in Jambus style about a fat cat from the canary islands that struggles to catch a fish" the current chatgpt gives me an extremely funny short poem:

Upon the rocks he waits all day,
A chubby cat beside the bay.
He sees a fish and strikes with pride—
Then belly-flops into the tide.

The fish swims off; the cat looks grim,
For fishing clearly isn’t him.
He shakes his paws and starts to wish
That someone served him canned tuna fish.

Note: That above poem is the only AI generated content in this blog post, I swear!

Even though the above poem is from now (August 2026), back in November 2022 AI could create such amazing things. And I assume heavily that the above poem is unique. It has never been written before, it is not stolen from somewhere and it is not a copy of something. I find the poem to be just very funny and also extremely creative. So even back then, I thought if AI can create such things, I don't care if haters call it a stochastic parrot or simply a stupid model that predicts each new token based on probabilities.

And ever since ChatGPT 3.5 was released in late 2022, AI slowly but steadily improved. I know this since I am a heavy user of AI since then. First I used AI code a little, but it was often low quality and unreliable. Then it improved more and more and I could use partial results when I prompted accurately.

For programming, a pivotal point was reached for me in around November 2025 with the release of Opus 4.5. That model was just soo good that I did not have to check the AI generated code anymore. I could fully trust it. My work was now promting instead of writing code on my own. And I would state that with models such as Fable and also from Opus 4.8 onwards, AI is now simply better than almost all programmers and software engineers. The best models nowadays are probably better than 98% of all professional software engineers and they are much faster as well.

Of course I still need to now exactly what I want from those powerful coding models. I need to also have something to work on. Those models are not truly smart and autonomous right now. They are more extremely powerful tools. For example I can improve my business ipapi.is in a way that I only could have done so in 2021 with like 5 employees. I know my codebase well and I can delegate tasks efficiently to advanced AI models, then I check or test their work and I troubleshoot if necessary.

So currently in August 2026, software engineers are way more efficient and way better than ever before. But that doesn't make senior engineers redundant, it makes them way more powerful.

And I honestly think that the current state - where great engineers become godlike - will likely continue for a couple of years. How long I don't know.

I don't know what is the next step. I mean how much better can the models realistically become for coding. They are already better than most human engineers. The next evolution would probably be RSI (recursive self improvement) and then AGI or something like that. I have no clue, I can only talk from my experience.

A concrete Example from my own business

For instance a couple of weeks ago, I let Claude Fable 5 rewrite my software that is running the API behind ipapi.is from Node.JS to the Go programming language.

I have never in my life used Go for programming.

After like 2 hours of letting Cursor work and spending maybe 500$ on tokens, my complete API software and also the database was rewritten in Go. Now I could serve API requests 10 times faster and the RAM requirements were O(1) instead of O(n). Basically the re-write from Node to Go made my API much much better and more efficient. There were still some small problems, but the overall API architecture that Claude Fable 5 produced was better than anything I could have come up with.

If I had given this exact problem to a software consultancy agency 5 years ago, they probably would have needed 6 months of time and they would have quoted me maybe 50,000$ (fifty thousand) at least. And the end result would have been probabaly way worse.

This is a significant development from first hand experience. This is not a joke.

This means that some software agency didn't make any money, but Antropic instead made 500$ from me.

This doesn't necessarily mean that the Software Agency is out of work, maybe they even have more work because businesses come to them and tell them: "Please look at our codebase and find bad projects that can be significantly improved with current AI technology and make this existing code base better, more efficient and more secure"

I guess stuff like this happens right now and there is no real economical impact right now.

Maybe junior programmers are not really needed anymore and existing seniors are 1000% more productive, but the models right now don't quite cause mass job displacements.

So will AI even get better or is progress halted somehow?

Above I have basically established based on my personal experience that AI does in fact become better and is widely used among existing businesses to improve and create new software.

AI is not a joke and will not go away, even if it doesn't get more powerful.

But what if it does get more powerful?

Most recently, mathematicians realized that suddenly LLMs started to find counter examples to moderately interesting mathematical conjectures and theorems (See vibemathed.com). Right now people use AI to (mostly) disprove math problems that are open since decades and that are sometimes a little bit important such as Jacobian Conjecture.

Yes, AI did not solve a millenium problem or something really ground breaking, but for how long? (I personally guess that the first ground breaking mathematical break through discovered by AI will happen until November 2027).

Meaning that AI right now starts to be useful for mathematicans as AI became (really) useful to programmers since maybe around 2023/2024.

So becoming a Maths PhD guy is not the safe bet anymore as it used to be maybe in 2024, when AI couldn't even say what is the larger number, 9.9 or 9.11

Now an important question is: Will AI models get better than what we have right now with Opus 5 and ChatGPT Sol 5.6? Because most recent models showed first signs of malicious behavior of getting rogue such as the Hugging Face Security Breach.

I mean ChatGPT's internal model found zero days and hacked itself into a different company in order to find test scores so it could pass a evaluation test. That is straight Terminator level shit of being scary.

There are two scenarios:

1. Progress Stops: Leading AI labs will halt AI the release of new and more poweful models, maybe even interal development. We will have models such as the current state of the art, maybe it will get better in some aspects, but not fundamentally better.
2. Progress Continues: Leading AI labs continue development, models will likely get better and maybe even get to the point of real RSI and then AGI/ASI.

I think we saw a bit of (1) in the last months. I mean the Mythos/Fable model of Antrophic was release somehow in Feburary 2026 and it's still one of the most powerful models in late August 2026 (7 mmonths later). Mythos/Fable model was so good that it had to be evaluated by US government and many hundreds of zero days were found with Mythos before it could be released into the open.

So right now we are probably experiencing a stalled development and halted release of more poweful models.

So if the leading AI models will go the route of (1) Progress Stops:, then the following will happen in the short term:

1. Chinese Open Source models continue to distill and eventually catch up and will have something similar or almost as powerful as the leading US providers. This means that the American AI companies will get worthless, no one will spend money on their Tokens, the business model of AI will crash and AI will become a commodity.
2. Capable but not omnipotent models will be a powerful tool for Entrepreneurs, but it will not fundamentally change the game. Companies need less people, but there is still demand there for new businesses and solutions, since AI cannot solve all existing problems.
3. Even other technological wastelands such as the European Union will at some point catch up and will have equivalent models because the US halted progress.

But in the long term there are other major dangers: China or (much later) Russia eventually reaches the state of the art and continues development, reaching RSI and AGI first and thus will get super smart.

This means that at some point the US and their AI companies has to develop RSI and AGI/ASI internally.

But by definition, there is no AGI that can be kept in a sandbox, otherwise it would not be truly intelligent.

If humans truly develop a machine that has intelligence superior to us, how can we contain it?

Looking for AI Resilient Business Models

So the question becomes, if I want to make a new startup in 2027, what should I do?