AI teams retain people when the work stays intellectually stimulating and the company supports learning, collaboration, and growth.
2
Diverse teams bring different language, cultural, educational, and lived perspectives that can expose model and data blind spots.
3
Companies attract AI talent by communicating tangible impact, offering access to conferences and publishing, and building visible technical communities.
Summary
This panel focuses on the people systems behind growing AI teams. Olga Beregovaya, Ashley Antonides, and Shailvi Wakhlu discuss how to retain technical staff through interesting work, learning opportunities, supportive managers, and visible customer impact. They describe team structures that reduce friction between developers, data scientists, product groups, and governance functions. The panelists connect diversity to model quality, giving examples of language, cultural, educational, and lived experience that can reveal issues other team members miss. They also explain how companies can attract candidates through meaningful work, conference participation, publishing, internships, university partnerships, and useful public content. Human-in-the-loop work receives attention because repetitive evaluation and labeling tasks need good tooling and a possible growth path. For fast-moving teams, the panel recommends protected time for research and experimentation alongside clear project goals. Retention depends on compensation and benefits, but the discussion gives more weight to team culture, intellectual curiosity, career development, and the ability to see that work affect customers.
Interesting work and learning culture give AI employees reasons to stay
Olga Beregovaya says AI teams retain people when the work remains intellectually stimulating. Ashley Antonides adds that technical ability alone is not enough. Teams need a learning culture where people read papers, ask questions, collaborate, and build on each other's energy. Diversity also belongs in team design because AI systems affect people from different walks of life. The panel treats learning as a daily team practice rather than a benefit added after hiring. Internal reading groups, workshops, cross-training, and access to new research help people keep developing while they work on company projects.
Team leadership must connect development, machine learning, product, and governance
Ashley describes friction between developer teams and machine learning teams, especially when organizations separate them into different groups with different workflows. She suggests an overall leader, such as a chief product officer or CTO, who understands both perspectives and can build shared infrastructure for model delivery. The panel also discusses AI ethics and governance roles. Technical teams can see an ethics officer as an auditor or adversary when the role is kept outside the team. Bringing that person into technical discussions makes requirements clearer and lets governance become part of the delivery life cycle. Olga adds that large companies need coordination because different departments can independently launch AI initiatives.
AI responsibility should exist across every department
Olga argues that AI work now involves more than the technical group. Development, product, marketing, compliance, and human resources each need someone who understands the department's role and helps maintain alignment. The panel mentions an AI digital ethicist at Ikea who looks after ethical deployment and the distribution of responsibilities. Olga also supports a centralized governing office that can coordinate company activity while leaving room for creativity. The concern is that easy access to language model APIs can lead to scattered projects, such as separate efforts in product and digital marketing. Coordination helps the company understand what each group is building and why.
Diverse teams can find model problems that homogeneous teams miss
Olga says diversity is needed across data collection, model development, and model implementation because otherwise teams can produce one-sided views. She gives the example of a Chinese chief data scientist and young mother in a language AI team. Her language background helps her notice hallucinations that colleagues with different language backgrounds might miss, while her experience talking with a toddler informs how she thinks about giving precise instructions. Ashley says hiring managers should ask candidates how they think about diversity and how it will affect their role. The panel also expands diversity beyond demographic traits. People from boot camps, self-taught backgrounds, other industries, and different educational paths can bring useful domain knowledge and lived experience.
A clear account of impact helps companies compete for AI talent
Olga says companies need to explain what their work will change for customers or society and where the organization is going. Candidates are more likely to engage when the goal is concrete and the work has visible impact. She also describes using interviews with management, senior management, and peers so candidates can meet the people they would work with and judge the culture. Ashley says research-focused teams should explain opportunities to attend conferences, publish papers or blog posts, receive training, and join internal reading groups. Smartling encourages conference submissions and publishing because its practical AI work gives researchers access to real-world data and case studies.
Human-in-the-loop work needs good tools and a credible growth path
Olga says models still hallucinate and data remains biased, so language AI systems continue to need people for assessment, ranking, validation, post-editing, and fact checking. These tasks can feel repetitive or like small side jobs. People are more likely to stay engaged when the tools make ranking prompts and assessing outputs easy. Ashley describes three approaches to data curation: hire people into the company, use an internal crowdsourcing system, or work with an external labeling partner. An internal route can give data creators or labelers a path toward machine learning engineering. Having the machine learning team label some examples also helps them understand edge cases and write better guidelines.
Companies can source talent through networks, universities, public work, and internships
Ashley starts with personal and company networks, then points to university partnerships, conferences, recruiters, and head hunters. She says recruiters work better when the company clearly explains its requirements and use case. Olga says conference presentations and useful LinkedIn content can create inbound interest from people who want to work on the problems being discussed. Internships can also convert into full-time hires, although neither speaker suggests that senior specialists appear without sustained recruiting work. Shailvi Wakhlu adds that a loved consumer product can make hiring easier, while B2B companies may need to create more public opportunities for collaboration and show potential candidates what the work looks like.
Fast-moving teams need protected exploration alongside measurable goals
The panelists recommend internal webinars, paper discussions, workshops, and small research projects to help teams keep up with rapid changes. Ashley describes a 'micro sabbatical' where someone spends a week or two researching a topic and returns with a white paper or prototype. Olga supports setting research tracks without chasing every new model or method. She also describes allowing personal projects for part of the week while holding people accountable for the rest of their team commitments. Clear KPIs help people protect delivery work while still making room for experimentation. Motivation rises when teams can see that a prototype improved metrics or helped a customer meet its own goals.
"I think it's all about an interesting job in this modern time and day, it is so exciting to be in the area that actually just keeping people intellectually stimulated."Olga Beregovaya02:13
Who should watch
You are building or reorganizing an AI team and need practical ideas for dividing responsibilities across engineering, product, research, and governance.
Your hiring process struggles to attract people with different backgrounds, or you want to turn conferences, internships, and public technical work into recruiting channels.
Your team relies on human evaluation or labeling and needs to keep that work useful, motivating, and connected to career development.
