Typeform meets AI. That’s kind of what I ended up building.
You lucky person can use for free (and w/o an account) at juttu.co
But what I wanted to share, what I think is something most people don’t talk about is: Fake hype and AI agents.
In short: What is an AI agent
It’s a systematic way to use LLMs (or other ai models) and normal code together. The intention is to produce a system that combines the benefits of code (it always does what its told) and the benefits of AI (it creates and understand ambiguous stuff).
Why everything is kinda an AI agent
In every product you use, if there is AI involved the developers working on it will have a major issue. It’s that the AI does unpredictable stuff, and that the user does unpredictable stuff.
Two parties that doe unpredictable stuff is too much to handle for a code base. Too many edge cases.
But imagine now doing several AI steps after each other, where the second step depends on user input + the output of the first AI operation.
Then it gets really wild and unpredictable.
So we need to control the AI. And as soon as we attempt to do that, we are already building an “AI system” and not just a prompt. Looks a little like this:
So, the complexity differs, but wherever you see AI embedded, it’s probably some form of AI system like this - which people call agents.
Now, there are purists that somehow define agents as where the AI autonomously does everything without any human intervention. In which case, AI agents do not exist yet, so it’s kind of useless to talk about them.
Designers, Product People you need to study this
Now, while there is a ton of hype, building ai systems is going to be part of almost any product in the coming years. Not cause of the hype, but because it’s useful.
Just try out juttu.co, which is still early on, and tell me that you’ll not want survey analysis done by ai first. Sure, you can download the csv .. but why not have AI summarize the findings first?
And building my first 3 AI systems I’ve learned a lot.
I built
a survey generation system with 5 steps and multiple branches.
a conversational survey system that has to work “on the fly” as the user responds to user questions but needs to be context aware
a survey analysis system that needs to extract trends, insights from complex data.
And these learning are not just about the code, they influence the UI/UX. How does it flow? How do we control user expectation, user input? How do we balance speed and quality (important!).
So I can only say: Go learn these things, play with these interfaces! They will be everywhere. Look through the hype.
This Anthropic post is a great read on it, that’s very understandable.
And shoot me a message if you want to chat about this :)
niko@juttu.co
Hope you are as excited as I am about how much better products this is going to enable!
Happy building,
Niko