We're cofounders who left Amazon to build our own startup. We learned the hard way why Big Tech habits don't always translate.

• Two former Amazon employees share what they had to unlearn to build an AI startup.
• The biggest challenge was working with a frugal mindset and limited resources.
• AI tools helped them save costs, speed up research, and reduce the need to scale head count.

This as-told-to essay is based on conversations with Shalini Aggarwal, a 50-year-old CEO in San Jose, California, and Andy Ratsirason, a 37-year-old CTO. The two left Amazon at different times and reconnected as cofounders of Tenfali, an AI startup.

They share the main processes they had to unlearn in order to successfully grow their startup. The following has been edited for length and clarity.

Shalini Aggarwal: Andy and I began working together in 2015, after I relocated to the US from India. He was a dev engineer, and I worked on the product and program execution side.

Andy Ratsirason: I joined Amazon for the first time in 2014 because I wanted to be part of the Silicon Valley ecosystem. I always wanted to be an entrepreneur, so I tried to tailor my career to suit that goal. I left Amazon, came back in 2020, and left again in 2023.

After I left for the last time, I made multiple pivots in founding a startup. I realized I needed someone else to join my team. Shalini started showing up regularly on my LinkedIn feed, liking and commenting on posts about startups and taking risks, so I reached out to her. She had just left Amazon, and we reunited to cofound Tenafli.

Aggarwal: We quickly realized that we took a lot of systems and tools for granted when we were in an enterprise company. The startup world is completely different; here, we have to build from scratch, and there was a lot about our mindset we had to unlearn.

The AI boom made leaving Amazon easier

Aggarwal: I stayed at Amazon until 2024, and my last nine months were spent working on AI projects, specifically on music recommendations.

Ratsirason: In 2023, I was almost three years into my return to Amazon, and I had a choice: either stay comfortable with those Big Tech paychecks or take a leap and launch my own startup. I submitted my resignation in February and then began to think about what was next.

Aggarwal: My decision to leave Amazon was more of a gradual process. During the COVID pandemic, my father retired, and we lost my mom. I could see how lonely my father was, and the biggest challenge was how to fill his days. That planted the seed.

Through working on personalized AI recommendations, I thought of a product that could serve as an AI companion for older adults, providing personalized recommendations and scheduling activities based on their typical daily routines.

My 50th birthday was coming up, and the AI boom was happening; if I didn't take the leap then, I knew I would never. In September 2024, I put in my two weeks' notice and left Amazon.

We had to unlearn the Big Tech habit of building before finding demand

Ratsirason: At Amazon, I didn't really think about why we were building a product or if people would use it. The 'build-first' mindset meant focusing solely on building a good product, knowing the customer was already there.

A clear moment that this mindset wouldn't work at an early-stage startup came after we spent months building, and launch day arrived, almost nobody showed up organically for a few weeks. That was the wake-up call: shipping isn't the finish line when we didn't have demand.

After that, we shifted to doing customer conversations earlier and running small distribution tests, including waitlists, partnerships, and community posts, before over-investing in product.

Leaving Big Tech forced us to relearn how to be frugal

Aggarwal: We applied and were accepted into several startup resource programs, including AWS Activate, Nvidia Inception, and Google's Cloud Credits program.

Ratsirason: We received a few thousand in AWS credits, but before we realized it, we lost almost all of the credits in the first two months. We over-provisioned capacity, and during AI testing, we left a few resources running longer than intended, which quietly accumulated costs over time.

Once we noticed the issue, we set up AWS budget alerts, added cost monitoring, made shutdowns part of our testing checklist, and simplified certain aspects of the architecture to match our stage.

To reduce costs further, we also bought a small local machine to run some of our AI experiments on, so we only use the cloud when we truly need it for scale, managed services, or production.

Using AI for research freed us up to spend more time with customers

Aggarwal: The time we save with AI enables us to allocate our energy to attending summits, forums, and programs, where we can learn about the work of others in the field. We also spend more time with prospective customers doing interviews.

Ratsirason: We used to spend hours reading long articles and research, and trying to keep up with the latest news in the field. The cycle felt heavy and slow, pulling us away from customer conversations and shipping.

We now use AI to scan a large set of content, surface the most relevant ideas for us, and summarize the few pieces worth reading. Having to be frugal, we've learned to spend only where learning happens. Talk to users first, then build the smallest thing that can prove value.

Certain roles we don't need due to AI

Ratsirason: AI acts as a junior engineer, handling a lot of the coding for us with the requirements we set up.

Aggarwal: It's also taught us where we won't need to scale. I know I don't need to hire a user interface designer. If I understand the requirements of something, I can quickly draft it with the help of AI and receive feedback on it myself.

Ratsirason: A few years ago, launching a usable version of our product, Agefully, would have required significantly more capital and head count. We just need two engineers and subscriptions. I'm grateful to be part of this AI era.

Our Big Tech backgrounds helped — but they also held us back

Aggarwal: We've seen at scale how things work and what processes we can implement to avoid chaos later on. At the same time, the biggest disadvantage has been overcoming the mindset that tools and infrastructure are readily available. Until we learned how to work through that, we weren't allocating our energy properly.

Ratsirason: Coming from Amazon, the name itself has some positive weight, but it can also work against us. People might assume that since we worked at a big company, we don't know how to run a small startup because we don't have a lot of resources.

The hardest part was overcoming the fear of shipping something imperfect. Coming from Big Tech, we were accustomed to high-polish expectations and assumed anything less would turn users away. We learned that in early-stage startups, the real risk isn't rough edges, it's building something people don't need.

Do you have a founder story to share? Contact this reporter, Agnes Applegate, at aapplegate@businessinsider.com.