‘I thought I wasn’t qualified for the role’: Google engineer reveals what helped him transform his career

When Emrick Donadei joined Google, AI was not part of his job description. His work, like that of many software engineers at the company a few years ago, had little to do with large language models or AI safety.“I wasn’t qualified for an AI role,” Donadei, now 32 and based in New York, told Business Insider. “I didn’t have the right credentials, and when I spoke with other teams, I felt a disconnect.”Today, Donadei works as a software engineer on AI safety at Google. His transition did not come from a formal degree in AI or a sudden promotion. It came from a decision to participate in a hackathon at a moment when the company itself was shifting direction.

From traditional engineering to LLMs

Donadei says the turning point came when ChatGPT’s release pushed Google to accelerate its focus on LLMs. As internal opportunities to switch roles opened up, he became curious about AI but doubted his eligibility.“Fundamentally, it’s similar to my last role,” he said, describing his current work to Business Insider. “But instead of building software, I’m building LLMs, which requires data, training, and compute.”At the time, he had not worked hands-on with AI products. That gap, he said, made conversations with AI teams difficult. He believed demand played a role in his eventual transition, but not without effort.“I was at the right company at a time when demand was high,” he said. “But also because I decided to participate in a hackathon.”

Why hackathons mattered

Donadei is direct about what changed his trajectory. “I think hackathons are the best way for everybody to get into AI,” he told Business Insider.He had prior experience. He participated in Amazon-led hackathons in 2018 and 2019 and won both. When Google announced its annual employee-only hackathon in 2024, which ran for seven days, he saw it as a chance to compete on what he called “the hottest topic in the industry.”During the hackathon, he built a small prototype and demoed it at the end. The goal was not novelty, he said, but exposure.“I didn’t do anything revolutionary,” Donadei said. “I built a small prototype that wasn’t super useful, but it was a good way to get started.”The process forced him to work with unfamiliar tools and concepts. He learned about infrastructure for LLMs, agentic workflows, and model fine-tuning. “The things that are less sexy,” he added.More importantly, the experience showed him that he could build in a domain he had previously avoided.

What came after mattered more

Donadei is clear that participation alone was not enough. “You can’t just do a hackathon and stop there,” he said. “You have to actually leverage that experience.”Thousands of employees participate, and only a few win. His team did not treat the prototype as a one-time exercise. Instead, they talked about it. A lot.Donadei reached out to group technical leads across Google to present the project. “They tend to have a high-level view,” he said, explaining that a 30-minute discussion was often enough to determine whether an idea fit a team’s work. If it did not, he was usually redirected to someone else.Those conversations helped him build connections outside his immediate network. They also forced him to explain his work clearly and position it within Google’s larger AI efforts.The momentum continued into 2025, when he participated in a second hackathon. That work later led to a public technical disclosure with Google.

Learning outside formal roles

Alongside internal projects, Donadei continued learning independently. He said he relies heavily on AI tools to accelerate his understanding.He uses Claude Code to read code and documentation, Gemini and ChatGPT Deep Research for case studies, and NotebookLM to process large amounts of information. He also watches Andrej Karpathy’s YouTube lectures and runs a podcast with a friend focused on software engineering and AI.“We’re doing it mostly because it’s the most proactive way for us to keep learning,” he said.

Not late, but deliberate

Looking back, Donadei does not frame his transition as luck alone. He sees it as proof that timing matters only if paired with action.By giving him access to internal tools and exposure to decision-makers, the hackathon showed him that the gap between traditional engineering and AI work was bridgeable.“By granting me unlimited access to frontier technologies and a direct line to key decision-makers,” he said, “the hackathon proved that I wasn’t late to the AI revolution.”For engineers watching AI roles from a distance, his story offers a narrower lesson. Credentials matter, but participation matters more. The shift begins not with confidence, but with showing up where the work is already happening.