Structuring a modern AI team

Linkov groups companies as technology companies, verticalized solutions or services companies, and tech-enabled companies. Each faces different challenges and makes different technology decisions. He argues that adoption and use of technology, rather than the technology itself, often limit progress.

Linkov says hiring an AI researcher is not always the first step: teams may need substantial work on data, workflows, product integration, and ROI before specialized model training pays off. He suggests identifying the company’s bottleneck, such as shipping, retention, reliability, or monetization, then hiring to address it. For Wisedocs, he would favor domain expertise and customer skills over five top lab researchers.

Drawing on teams he built in 2021 and 2024, Linkov describes setting different skill bars for model training, serving, and business acumen to fit the team’s needs and budget. As open source and commercial models advanced, his teams could use existing serving tools while placing more emphasis on domain knowledge in medical record processing. He sees generalists as useful early in a company’s AI strategy, with specialists becoming more valuable when the team needs further performance gains.

Linkov frames team work as inner loops of daily collaboration and outer loops of expertise that help a company stand out. He recommends building skills in prototyping, domain understanding, evaluations, and customer communication, including having engineers join customer calls. His team holds weekly 30-minute learning sessions to keep upskilling.

Linkov says people are needed to hold context and act on it, and that humans must verify work and remain accountable for the systems they build. He advises setting hiring plans around team composition and budget rather than trends about junior engineers or AI agents. He also recommends job-relevant interview questions instead of unrelated LeetCode exercises.