THE BIG IDEA
The biggest advantage AI-native startups have isn’t technology — it’s a blank slate. AWS believes enterprises can capture some of that advantage by adopting first-principles thinking: questioning which processes, products and customer experiences still make sense in an AI era instead of simply automating them. The lesson is that AI delivers its greatest value when it drives business redesign, not just process automation.
For years, enterprises have adopted new technologies by layering them onto existing systems.
AI-native startups are taking a different approach.
Rather than asking where AI fits into existing workflows, many are redesigning products from the ground up, questioning assumptions that have guided software development for years.
That shift in thinking — more than any particular AI model or coding tool — may be the biggest lesson large enterprises should take from the newest generation of AI-native companies, according to Deap Ubhi, global director of technology for startups at AWS.
“The big advantage startups have is that they are starting from ground zero,” Ubhi said. “No legacy assets, no preconceived notions. This is truly what you would call first-principles thinking.”
Need more clues? Ask the Sherlock chatbot in the lower right corner to summarize this story, explain technical concepts or answer other questions.
The approach is helping AI-native startups move far faster than previous generations of software companies, according to AWS’ recent report on AI-native startups, “Engines of Growth.” While the report found that 78% of AI-native startups are ready for agentic AI and robotics, Ubhi said that readiness reflects how these companies are being built rather than simply their willingness to adopt a new technology.
“In a lot of the startups that we see today, agents are the primary interface between them and their customers,” he said.
First-principles thinking
Enterprises looking to become nimble should not ask how AI can automate an existing process. Rather, they should determine whether the process should exist at all. This is an exercise in first-principles thinking.
“I’m not suggesting that enterprises should tear everything down and actually start from scratch,” he said. “But just the first-principles thinking exercise will allow them to understand what their hard constraints are and what their soft constraints are.”
Hard constraints are those that are non-negotiable while soft constraints refer to those instituted perhaps by habit or tradition. He encouraged organizations to distinguish between actual business requirements and practices that have simply become accepted over time.
Ubhi applied first-principles thinking to customer service, in an example.
Most large organizations possess years of customer interactions across phone calls, chat sessions and email exchanges. Ubhi said that historical data can become training material for a language model that would eventually power AI customer service agents.
Communication from customers accumulated over many years will start to show patterns, which language models are good at recognizing and automating. A company could train a voice model as well with that data. The result would be an AI text or voice chatbot that interacts with customers.
“You’ve got data, you’ve got language models that you can use off the shelf that you can then train with that data. You’ve got voice models that you can use off the shelf that you can also train with your voice data,” Ubhi said.
Discarding, not just adding
First-principles thinking also means being willing to discard things that may not be necessary anymore.
Ubhi said an executive might ask, “What are my customers asking me for more often than not when they’re chiming in to a customer service channel? Which channel do I even need anymore? Do I need phone? Do I need live chat? Do I need email? Which one of those do my customers enjoy the most?”
“If customers really hate the email experience of customer service, why is that? And is it even needed anymore?” he continued. “And if voice agents are a more scalable way to deploy both humans and non-humans to be responsive to your customers, now all of a sudden you can think about a better use of your resources – meaning you maybe don’t need to invest in all that infrastructure to support email customer service.”
Organizations might decide that first-level customer inquiries can be handled by AI voice agents while more complex issues continue to escalate to human representatives. Rather than simply adding AI to an existing call center, enterprises can reconsider which communication channels customers actually prefer and whether every existing workflow still serves a purpose.
AI-native startups begin with a blank slate, so they can design processes around AI rather than first dismantling established workflows. This enables them to benefit from AI faster than others.
According to the AWS report, they reach $1 billion valuations in 3.5 years on average, half the time it took pre-generative AI startups. Also, they see 156% average annual revenue growth compared with 65% for startups overall, and 55% generate more than $400,000 in revenue per employee.
Narrow but deep, with ‘taste’
Many successful AI startups also avoid trying to solve every enterprise problem at once.
“They are being very disciplined about, ‘I just want to solve this one problem,’” Ubhi said. “’It may be an inch wide, but it’s a mile deep. Therefore, I think I’m going to get traction that way.’”
He pointed to companies such as Sierra in customer service, Harvey in legal technology and Heidi Health for clinicians as examples of startups building highly specialized AI agents around specific business problems rather than broad horizontal platforms.
Despite AI automating more software development, Ubhi believes people will become more valuable, not less.
Asked whether capital, regulation or talent will become the industry’s biggest constraint over the next several years, he chose talent.
“I actually think that talent is going to be the key headwind,” he said.
More specifically, he believes the differentiator will be what many Silicon Valley founders now call “taste” — the ability to design experiences users genuinely enjoy rather than simply building functional AI systems.
“It’s become the new human, widely sought-after skill – people that just have a sense of what the experience should be like for the end customer,” Ubhi said. Even in an agentic experience, a person with “taste” can discern which touch points will delight the end customer rather than just deliver a utilitarian experience, he said.
“It could become the difference between two companies that are doing the exact same thing,” he added.
