I’ve been building Frontier AI (FAI) at Lilly for the past 18 months now. Frontier AI is a boutique team meant to rethink how we do molecule discovery at Lilly. The seeds of FAI were planted when I started building out Kernel Lilly, our first research agent system, in 2024, when tool-using LLMs were starting to take off and the scientific community had seen ReAct-style agents in the form of Chemcrow and Co-scientist. It was clear that, beyond improvements in base models, connectors and runtime environments for the models to execute code would be key (what the cool kids now call the harness). Since then, we’ve expanded FAI beyond specific agent applications to include bespoke agent harness design and started to build capabilities to connect the agents to key laboratory automation platforms (agentic labs).

Building FAI from scratch has been a fun experience. Hiring at the frontier of AI and life science was uncharted territory, especially in late 2024, when agentic AI was in its infancy. It took a lot of rethinking and recalibrating how hiring and talent identification happen. Everyone who is now part of FAI has been selected meticulously by me with the help of my collaborators, who are also end users of the systems we build.

A few weeks into hiring, we quickly realized that conventional interviews suck and are not meant for the coming AI age. Coding interviews can be gamed, and “culture” rounds can be faked. Graham Duncan has written beautifully about hiring and his experience in building teams. His writing inspired a lot of our interview and hiring decisions. Through our interview process, the hope was to reveal the character of the person, the way they think and operate, and whether FAI is the right place for them.

As we continue to hire, I wanted to share a few things that I personally look for in candidates:

  1. Integrity. I place huge value on this virtue, more than on technical competency. If a candidate is willing to say things just to get hired, it is an automatic rejection for me. That does not mean you need to know everything; self-awareness and not being afraid to look stupid are important.
  2. Passion. You don’t need to fit a cookie-cutter mold of computational chemistry or bioinformatics—or, more recently, AI engineering. When I came to Lilly, I had never seen a protein structure before. I am looking for passion, a flame of genius that can be stoked. We are hiring for how you think rather than what you’ve thought about. Explain an idea you are passionate about in simple terms. Share how you implemented it and why you made those decisions. Who helped you along the way? A simple, clear example devoid of any buzzwords is a good signal that you are really passionate about it.
  3. Relentlessly resourceful. This follows passion closely. People can be passionate but still be armchair philosophers. Are you going out of your comfort zone to bring about change? How do you learn and incorporate feedback? Ideas are cheap; execution is expensive. At FAI, no ask is too small or too big. Keeping your personal bandwidth in mind, you are expected to wear multiple hats while working at the frontier of what’s possible. This does not mean you vibe code and expect someone to make it engineerable. Having the discipline to understand the spectrum from prototype to production is important. This also dovetails with the ability to communicate what did not work while having a bias toward action: trying different ideas quickly with a clear intention behind each.

Lead by example through your work. Ask good questions. Care for the scientist and the science you build tools for. Be genuinely curious.

A near-perfect team is made up of smart, passionate, imperfect people who complement each other. If these ideas and the mission move you, please apply. I look forward to chatting soon!