Five lessons for building AI apps

Five things we've learned the hard way about building AI-powered apps

We just reached 1,000,000 installs of Plus AI (so fun to type out the full number! 😆), so we wanted to look back and share some of our reflections on building AI apps over the last year.

Capitalize on the hype, but tune out the noise

Google search trend for “AI”

There is an insane amount of interest in AI right now, so you should absolutely lean in to “AI” as a core piece of your marketing message. 

But… it's important to identify the customers who need to solve a specific pain point using your product vs. people who just "want to try out AI." The latter group will be happy to give you lots of (distracting) feedback, but they won’t get out their credit cards to buy your product.

Decide how much time you want to spend educating customers

Reality rarely meets customer expectations in AI. If the average AI demo is ~60% possible when you actually use the product, the average customers' expectations are (at least) 120% of what's possible.

Depending on your GTM motion and the type of app you are going to build, decide how much time you want to spend educating your customers on what’s possible, when, and how to use AI.

If you are selling $100K+ deals, this is just going to be part of the process. If you’re building a PLG product, find the customers who already know what to expect and how to use your tool.

You probably shouldn’t build a chatbot

ChatGPT has been a huge success, but for most use cases, AI-enabled workflows and apps are a much better interface than AI chatbots. Chat is too open-ended and doesn't expose the right options and features to users.

Even when ChatGPT reaches “human level intelligence,” it will still be better to use structured apps and workflows to give AI instructions. That’s why we literally have dozens of different project management apps for humans to communicate with one another.

“ChatGPT wrappers” will be billion dollar companies

"ChatGPT wrappers" are the new "CRUD apps." Like a whole generation of incredible SaaS companies that "just" built CRUD apps, multibillion dollar AI companies will be built around amazing user experiences, data integrations, and pairing a great product with the right GTM model.

It’s not easy to build a great app, whether AI-powered or not, so don’t feel bad if an investor calls your app a “ChatGPT wrapper.” Just focus on solving valuable problems for customers.

Ship. Get real data, and use it to improve your product.

Data really is key now. It's hard to iterate on AI products without real-world usage of your product, but once you have it, it's way easier than ever before to make things faster and better.

Here’s a look at our subscriber chart - the 1s at the bottom of the chart are annotations for key product releases, model updates, etc. As we built up momentum and learned more about what our customers wanted, it made it easier and easier for us to continue improving the product.


It’s an exciting time to be building AI products. There is a lot of potential, and there are a lot of unknown unknowns. We think the best way to learn and shape the future of AI is to build something yourself, and that’s what we are doing at Plus!

Table of Contents
  1. Item text