Intuit today announced the inaugural recipients of the Intuit Faculty Research Awards: five research collaborations with leading faculty at four universities — Stanford University, The University of Texas at Austin, the University of California San Diego, and the University of California, Davis — representing a combined $500,000 in unrestricted research contributions.
This inaugural cycle was funded under a single call for proposals, “Foundational AI for Autonomous Decision Intelligence,” issued in December 2025. The call drew a competitive pool of applicants from faculty across the nation who went through a structured, multi-stage review led by Intuit’s AI Science community panel, with responsible AI and ethics a formal part of the evaluation.
As Intuit continues to lead in AI that acts on a customer’s behalf — pairing artificial intelligence with human expertise — the bar for trust has never been higher. An agent that recommends a purchasing plan can be wrong without much cost; an agent that executes one cannot. Each of this year’s five awards attacks a distinct, foundational piece of that trust problem: verifying that an agent’s model of what a customer wants is actually complete (Udell), giving agents a way to adapt mid-workflow and prove their plans are safe before acting (Huang), measuring whether a sequence of financial recommendations genuinely helped rather than just correlated with a good outcome (Syrgkanis), teaching agents to earn greater autonomy only as they demonstrate reliability (McAuley), and building the evaluation-and-fine-tuning loop that makes agents provably better, and auditable, over time (Ghosh).
Together, these projects advance the scientific foundation behind Intuit’s mission to power prosperity for tens of millions of consumers and small and mid-market businesses who use TurboTax, QuickBooks, Credit Karma, and Mailchimp. Each addresses one of the three outcomes at the center of that mission — more money in customers’ pockets, less time spent on financial admin, and greater confidence in financial decisions — by closing gaps that stand between today’s assistive AI and tomorrow’s trustworthy, autonomous decision support.
This work is deliberately structured to advance the field, not to give Intuit a proprietary edge. Awards are unrestricted gifts, universities retain all IP, results are published and open-sourced, and every experiment runs on public or synthetic data — never Intuit customer data. That structure lets Intuit help set the research agenda for trustworthy agentic AI industry-wide, while building the university relationships and talent pipeline that will shape the next generation of this field.
“We selected proposals not just for scientific excellence, but for their ability to close concrete gaps in how we build trustworthy agentic AI.”
Kamalika Das, Head of AI Research, Intuit AI Research |
Meet the Inaugural IFRA Cohort
Recipients are listed in alphabetical order by last name.
Joydeep Ghosh
The University of Texas at Austin
Project: An Integrated Approach to Efficient Evaluation and Fine-Tuning of Trustworthy Agentic AI Systems
Summary: Builds a unified framework that evaluates and fine-tunes AI agents together, using “hindsight relabeling” to extract learning value from partial successes and failures so agents improve measurably and stay auditable.
Lifu Huang
University of California, Davis
Project: CAPA: Certified Adaptive Planning and Auditing for Trustworthy AI Agents
Summary: Builds a closed-loop framework that lets AI agents repair their understanding of a shifting workflow, plan with machine-checkable safety certificates, and automatically trace failures back to their root cause.
Julian McAuley
University of California, San Diego
Project: Continual Skill Maturation for Progressive Autonomy in AI Assistants
Summary: Teaches AI assistants to earn autonomy gradually — moving from recommending, to guiding, to acting with confirmation — and to pull back safely the moment conditions change.
Vasilis Syrgkanis
Stanford University
Project: Dynamic Causal Inference for Sequential Decisions
Summary: Creates automated statistical methods to reliably measure the true impact of a sequence of financial recommendations over time, even when hidden factors could otherwise bias the results.
Madeleine Udell
Stanford University
Project: Verified Decision Models from Conversation: Specification Completion and Verification
Summary: Develops methods to verify that an AI-compiled financial plan captures everything a customer actually meant — including the constraints they never thought to state — before the AI acts on it.
Looking Ahead
IFRA reflects Intuit’s ongoing commitment to advancing foundational AI research in partnership with academia — and it’s one of several ways we’re deepening our engagement with the research community driving this field forward. We’re excited to be working with this inaugural cohort, and we’re excited to share what we learn together.