Behind the scenes in Cornell’s College of Agriculture and Life Sciences, Ido Efrati is helping shape how AI gets adopted. He does so by starting not with the technology, but with the people and problems it might serve.

In practice, that means sitting down with a person or team and asking them to walk him through their day: where they hesitate, where they repeat themselves, where something feels harder than it should be.

“What matters first is learning their world,” said the programmer analyst. Only after he understands their workflows and roadblocks can he begin a conversation about potential solutions – AI-focused or not.

To dig into that institutional experience, he’s been developing a structured approach to AI discovery, including an intake process with questions designed to reveal the problems, bottlenecks, and judgment calls already in place.

From Problem Triage to Practical AI with Broad Applications

“Understanding the problem is only half of it; the other half is clarifying what's technically possible,” said Efrati.

Part of that is understanding which part of the AI spectrum applies, because it isn’t one thing: it ranges from personal productivity tools to agentic systems that automate complex workflows. Getting that match right and being honest when AI isn't the right fit at all, is itself a skill he has been developing.

Creating space for Efrati to explore innovative solutions and develop strategic approaches to AI adoption are his key supporters, CALS OIT Director Chris Hufnagel and Mike Barre, Efrati's direct supervisor. Their belief that innovation is worth investing in, even when outcomes aren't guaranteed, encourages Efrati to keep pushing his pilots forward.

“Even though none of the pilots are finished or even guaranteed to succeed, we are making progress,” he said. “The bigger goal is to build a repeatable model for how CALS identifies, evaluates, and adopts AI: a college-wide strategy grounded in our real operational needs, structured with Resilient Cornell objectives in mind, and designed to build lasting institutional capability.”

AI as an Extension

At a recent AI Demo Day, Efrati heard an opening speaker make a point that stuck with him: AI isn't replacing people at work, it's giving them more options and more things to work on. That reframing shaped how he thinks about what AI does.

Efrati said, “What I've found is that, fundamentally, AI extends our reach. It lets us take on complex problems we previously didn't have the resources for or navigate technical domains where an expertise gap would previously have been a hard stop. Even non-IT staff can experiment and bring a proof of concept to their IT colleagues. Exploring AI together opens up new conversations and working relationships that didn't exist before.”

He believes AI also changes the feasibility equation, enabling us to reexamine problems that were always worth solving, but were too complex, or too dependent on bandwidth that was simply unavailable. Those problems are now worth revisiting, and that's what makes this moment feel genuinely cutting edge.

Thinking Big, Starting Small, and Changing Direction

Even if the limits of feasibility and reach are extended or removed, the problem at hand still requires a practical solution—one that starts with small changes and builds. To structure what can feel like unlimited exploration, Efrati builds in checkpoints.

He said, “Before anything gets built, there's a planning phase: what's the core assumption, what would prove or disprove it, and what's the smallest thing we could build to test it. For example, a hundred forms a year, manually copied field by field between two systems: can we supply an agent that will eliminate that entirely? A person at the budget office manually running queries and compiling reports by department: can we automate that entire process? Each phase in those experiments either confirms you're onto something or tells you what to adjust before you've committed too much.”

Knowing that not every idea leads where the engineers expect is also part of the structure, and that's informative, too. For example, a TeamDynamix (TDX) ticket classifier. The proof of concept worked, but what it revealed was more useful. The IT staff didn't want it applied to every ticket automatically; they wanted to invoke the classifier for specific cases, and to feed it context they know but the system doesn't.

That feedback reshaped the pilot direction and raised a better design question: how do you give a system access to institutional knowledge dynamically, and let it draw on that in real time? Sometimes what a proof-of-concept reveals is that the real problem is different from the assumed one, and the redirect is often more valuable than the original path.

Return on People

Even though Efrati is enjoying experimenting with AI, he is more excited about shaping how CALS uses it. He pairs his urge to "just start building things" with developing skills and methodologies that would hold up at institutional scale.

“The same qualities that make AI useful - it's fast, it's confident, it produces things that look right — can also make it risky if you're not paying close attention. You need to design thoughtfully, and recognize when something's wrong,” he said.

“The biggest shift for me was moving from thinking about return on investment — is this worth the dev time? — to thinking about return on people. Can this give someone back hours in their week? Reduce a stressful manual process? Let them focus on work they care about? That reframing changed how I identified opportunities and made the work feel much more purposeful.”

That reframing led Efrati to volunteer as a tech lead in the AI Innovation Hub, where he works alongside colleagues to develop structured AI solutions for real university problems — and to better understand, together, where AI genuinely helps and where it doesn't.

“Now, what excites me most are the problems people had stopped asking about—problems that are suddenly worth asking again. The back-burner list is getting shorter. That's what makes this feel like a real frontier.”