Hard truths from building a startup in the AI boom
After spending four months in San Francisco as part of the South Park Commons community, I've learned a few things about building in the AI boom.
I recently walked away from my startup Hovi AI, which now focuses on AI agents for marketing, despite having a million-dollar funding commitment and runway to burn. On paper, everything looked good, but I knew it was time to call it quits.
There's a fundamental truth that the AI boom has made clear: Spinning up products may be painless, but turning them into successful startups still demands hard truths, counterintuitive insights, and relentless focus.
Stumbling blocks
When the startup I was working for shut down in April 2024, I decided to start my own, exploring ideas in the global B2B SaaS space.
Our first idea was AI voice agents for consumer research. After two months, we realized it was India-specific rather than global and moved on.
This led to our first pivot - we knew we wanted to help consumer marketing teams but not exactly how. After three months of reaching out to marketers with no strong insight, we found a unique opportunity in playable ads for mobile gaming. These interactive ads perform well but are expensive and slow to build.
We thought an AI agent that would create them in hours instead of weeks would unlock significantly more demand.
To validate this, we "did things that don't scale" and operated as an AI-accelerated service for three months. This meant we didn't wait for the product to be ready to work with customers but built the ads ourselves, with AI agents helping behind the scenes.
This confirmed that AI agents could reduce the time needed for ad creation, and that there weren't existing products that addressed this pain point. We raised our pre-seed from South Park Commons and flew to San Francisco for its bootcamp.
But after building the product and speaking to customers, we encountered market and technical challenges.
For a start, the mobile gaming sector was not in a growth phase but a consolidation phase. Along with that, we realized that faster ad creation wouldn't expand the market because real costs were in media spend and testing time on ad networks like AppLovin.
We had thought that if it took an hour to make a playable ad instead of two weeks, people would create more. But the real barrier was testing the ads for effectiveness, and this took significant time and money even with more efficient ad creation.
As for technical issues, reliable coding agents for gaming needed near-perfect accuracy and a tight feedback loop to test what they built - something the technology wasn't ready for back then.
Because of these setbacks, we decided to hit pause on this idea.
After exploring other opportunities for marketers, I realized that about 50 other startups were building in the same space and were ahead of us. This was our second "pivot hell," and I wasn't ready to go through it again, so I stepped away.
A successful founder once asked me: "What desire would make you grind for years, away from family on Sunday nights?"
Market opportunity isn't enough. You need an irrational desire to build a company, and I realized I didn't have it.
Rules of thumb for founders in the AI era
The experience taught me that while generative AI and AI agents are powerful, it's easy to get caught up in the hype. A tech wave doesn't automatically translate to a competitive advantage. You need a counterintuitive insight. Here are some rules of thumb for determining if you have one.
Audience → problem → promise → product
Start with who, not what. As software building costs decrease, go-to-market becomes more important.
Stick to the same audience while iterating on problems, which helps you go deeper into their needs. You need to love the audience you're building for.
In our case, we weren't sure about gaming marketers, so pivoting felt like starting from scratch.
Size matters
Trying to solve the obvious, large problems is not a good strategy for most startups. Finding a niche is important, especially within a competitive space like AI.
Have a clear hypothesis about whether that niche will grow because of your solution, if it's already growing, or if you have ideas for expanding beyond it. Test this hypothesis before going all in.
In our case, we didn't try to compete with Lovable or Bolt.new by creating general coding tools. We chose playable ads - an underserved segment.
The 0.1x wedge question
Can you find partners willing to use your product at 0.1x of its potential? Your vision might eventually get to 100x, but you need customers who are ready to use the basic version to determine if diving deeper will be worthwhile.
This is unique to AI, as you can get prototypes and early versions of your product into consumers' hands rapidly. But spinning them up into the full version still takes time and resources.
In our case, we started with AI-accelerated services to learn the problem space deeply before building the full product.
Visualization beats explanation
People respond to what they can see, not abstract descriptions. Build rough demos that show the concept early.
We were hesitant to build before validating, but demos aren't hard to create these days, and we should have kept building and upskilling continuously.
Accuracy reality check
Do your customers need your AI solution to provide 100% accuracy, or is 80% good enough? If customers tell you that they like your product but they need it to be perfect, you're probably in the wrong space.
I've come across multiple startups that pivoted from building products in areas like text-to-SQL - where a user asks a question in plain English and the AI returns a database query - because they require high accuracy.
We faced this with playable ads. We should have validated if marketers were willing to use an AI agent that would get them close to a workable ad that they would need to refine. But we found that marketers wanted a system that would deliver an ad that was ready for release.
Conviction over all
Getting your startup off the ground is still as hard as it ever was, even if Sam Altman says that "this is the best time to build a company." The best time is when you're excited and convinced by your idea space, not if you're driven by FOMO.
The AI boom has created incredible opportunities, but also a false sense of how easy it is to build a successful startup. Developing products has become easier, but finding product-market fit remains challenging.
Build because you have that irrational desire, not because the boom makes it look easy.
Palash Kala is a former founder fellow at South Park Commons.
First published in Tech in Asia on 25th Sep 2025: https://www.techinasia.com/hard-truths-building-startup-ai-boom