In the fall of 2022, Olga Topchaya was home with a new baby, staring down the worst tech recession in a decade. Hundreds of thousands of people in the industry were being laid off. She was one of them.
She had spent the previous ten years in marketing and product roles in tech. She’d gone on parental leave, and by the time she was ready to come back, there was nothing to come back to.
“Instead of sending off my resume into a void and crying myself to sleep in a corner with rejection letters,” she says, “I’ll take a step back and hang out with my baby and reassess.”
This was months before ChatGPT launched. Most people in tech had never heard the phrase “large language model.” But Topchaya stumbled onto generative AI through a model called DaVinci 2. By today’s standards, it was primitive. She opened the playground, started experimenting, and something clicked.
“Holy shit, this is incredible,” she remembers thinking. “Why is nobody talking about this? Where is the news? Developers aren’t talking about it, companies aren’t talking about it. Nobody’s talking about this.”
She started reaching out to anyone who would talk to her about the technology. Random influencers, professors, people lurking in obscure corners of Discord servers. She joined her first hackathon, despite not being a developer. She built connections with a small community of people who understood what was coming.
She also told her husband, who is a developer.
“This is gonna replace Google, you know that?” she told him.
His response: “Lady, you are crazy. This is a large language model.”
“I was like, ‘No, I do not know what a large language model is, but I’m gonna learn,’” she says. “And he’s like, ‘No, a large language model can’t do that. It’s not what you think it is.’ And I’m like, ‘Watch. This is gonna happen.’”
She was right. And by the time the rest of the world caught up, she was already working in the space.
Her first foothold came as a prompt engineer at Copy.ai, one of the early startups building on generative AI. That was a contract role, and when it ended, another contract followed at Trustwise AI, a company focused on safety and explainable AI. Then a third contract, doing prompt engineering for an insurance company. By the time the third engagement wrapped, the pattern was clear.
“It’s crazy to be a 1099 consultant just wasting my money on taxes,” she says. “It just made more sense to own a company.”
That was the beginning of Lapis AI.
But the company didn’t stay where it started. Topchaya realized quickly that prompt engineering and consulting alone weren’t enough to compete. She had the contacts, the relationships with AI engineers she’d built during those early Discord and hackathon days. So she brought them on as contractors and started offering custom AI software development.
That still wasn’t enough.
The gap she kept seeing wasn’t technical. It was human. Companies would invest in solid AI strategy and well-built software, and the projects would still fail because the people who were supposed to use the tools weren’t using them, or weren’t using them well.
“You can have really great strategy, really great software,” she says, “but if your people are not using it, or they’re not using it properly, your projects are doomed to failure.”
So Lapis added AI coaching and training. Not as an afterthought, but as a core part of how the company delivers work. Before anything gets built, the team needs to understand how people will actually use it. Not in a lab, not based on what an executive imagined, but in the field, in the hospital, in the user’s actual day-to-day.
“I’m not thinking about technology first,” Topchaya says. “I’m always thinking about people first, business first.”
That philosophy shows up in her sales conversations, too. When a potential client comes to Lapis and says, “Can you build me this thing?” her first question is not about requirements. It’s simpler than that.
“Why? What are you trying to achieve?”
She credits her marketing and product background for that instinct. She thinks in KPIs, target audiences, and user behavior. It’s a different starting point than most AI shops, which tend to lead with the technology and work backward.
“I see all these projects that are just constantly failing and stuck in POC mode,” she says. “And that’s because they put the technology first instead of the business first and the people.”
Today, Lapis AI works primarily with mid-size companies, though engagements have ranged from small firms to enterprises with six thousand employees. The industries include medical, architecture and construction, professional services, and increasingly financial services. The company offers custom AI software development, strategy consulting, and the coaching layer that ties it all together.
One of Topchaya’s sharpest observations is about how companies actually waste time. She’s done the math across client engagements and breaks it into rough thirds: about ten percent of an employee’s time goes to what she calls “looking for your keys” (finding a calendar invite, tracking down what someone said in a meeting, locating a piece of information), another ten percent goes to repetitive tasks that could be automated, and another ten percent goes to administrative work that has nothing to do with what the person was hired to do.
“Get rid of all that stuff, because that’s not what you hired your people to do,” she says. “Take it away, give it to AI, because that’s the kind of stuff where AI succeeds. And allow your people to become more productive.”
The companies that get this wrong, she says, are the ones that treat AI as a headcount reduction tool. Fire everyone, replace them with software, cut costs. Lapis takes a different position: cut the busywork, reallocate the hours, and let people do the work that actually requires a human.
“Put your people in roles where humans are successful,” she says. “And the thing that you actually hired them to do.”
She’s equally direct about the risks. Hallucination gets most of the attention, but Topchaya is quick to point out it’s just one of many risk areas. Tool sprawl is a bigger problem than most companies realize. She’s watched firms that once required six rounds of review before a single RFP could go out the door turn around and give AI tools free rein across the organization with no oversight at all.
“You wouldn’t let me put a Facebook ad up,” she says, recalling her days in financial services marketing, “but AI is just like, ‘Here you go. Do what you want.’”
The newer models bring their own challenges. Topchaya and a colleague describe them as “almost too arrogant to use.” The latest generation of AI models is designed to be autonomous, which sounds good in theory but creates real problems at scale. A model might rewrite an entire document when all you asked was why it made a specific choice. At the individual level, that’s an annoyance. At enterprise scale, with thousands of users and real cost implications, it’s a business problem.
“The newest and the latest and the greatest isn’t always the best,” she says. “You have to choose the right model for the right job. And it is a hill that I will die on.”
What comes next for Lapis is research. Topchaya hasn’t made the official announcement yet, but the company is building out what she’s calling Lap Study, a research arm focused on developing proprietary products. One area of early exploration is medical AI. She’s careful to call it research, not launch.
Her champions are close to home: her husband, who once called her crazy and now helps push her forward; her mother, who is constantly at her side; and the development partners who’ve been with her since the early days.
When asked what advice she’d give another founder, Topchaya keeps it simple.
“You’re gonna have some good days and you’re gonna have some bad days,” she says. “And so long as the overall trajectory is upward, just keep that in mind. You’re gonna be okay.”
Olga Topchaya is the Founder of Lapis AI, a full-service AI consultancy offering custom software development, AI strategy, and coaching for organizations integrating generative AI and large language models into their operations. Her background spans more than a decade in marketing and product roles in the tech industry. Lapis AI serves mid-size and enterprise clients across medical, architecture and construction, professional services, and financial services.