Sign InOpen Brain
AI EngineerVideoSource Linked

Why Off-the-Shelf AI Doesn't Understand Money — Udi Menkes, Intuit

Intuit argues that domain context alone does not create experience. Its approach learns from verified state-action-outcome histories, using frontier models only to propose candidates.

AI Engineer · Jul 29, 2026
Open Source Open MarkdownOpen JSON
Source Summary

In Intuit’s study of roughly **100,000 business situations and time frames**, frontier models reduced more than half of their advice to acquiring customers or increasing revenue. The talk shows recommendations that ignored cash-flow and supplier-concentration constraints despite receiving business context.

Practical Implication

Build domain agents from verified outcomes, not documents alone. Intuit converts records into **millions of state-action-outcome vectors**, uses an RL model to rank candidate actions, and then trains an LLM to express the grounded recommendation.

Agent-Ready Context
In Intuit’s study of roughly **100,000 business situations and time frames**, frontier models reduced more than half of their advice to acquiring customers or increasing revenue. The talk shows recommendations that ignored cash-flow and supplier-concentration constraints despite receiving business context.

Build domain agents from verified outcomes, not documents alone. Intuit converts records into **millions of state-action-outcome vectors**, uses an RL model to rank candidate actions, and then trains an LLM to express the grounded recommendation.

Observed outcomes are not automatically causal: a business can improve after an action for unrelated reasons. Similar-company comparisons help, but the validity of the advice still depends on data quality, matching assumptions, and the chosen outcome measures.
Connected Context · Feed7 Judgment

This shifts domain-agent grounding from retrieving authoritative documents to learning from verified state-action-outcome records, while narrowing confidence in those recommendations through causal and data-quality caveats. It reinforces deterministic controls around financial reasoning but adds an upstream requirement: the action-ranking evidence itself must be valid, well matched, and tied to suitable outcomes.

How Kepler Built Verifiable AI for Financial Services — Vinoo GaneshKepler’s deterministic handling of financial values complements Intuit’s outcome-grounded action ranking; together they constrain both the recommendation evidence and its numeric execution.Build for the Memo, Not the Demo — Shawn Chan, China Resources HoldingsClaim-level provenance and explicit uncertainty become implementation requirements when outcome correlations may not justify causal recommendations.CausalForge: A Formally Grounded, Self-Improving Agentic Framework for Automated Research in Causal InferenceBoth distinguish valid derivation from valid interpretation: formal or statistical machinery cannot ensure that assumptions and claims match the intended real-world question.Semantic Blindness: 500,000 Sensors Confused an LLM - Raahul Singh & Vanč Levstik, PhaidraBoth replace unconstrained model inference with structured, domain-specific resolution, but Intuit grounds actions in observed outcomes rather than deterministic entity lookup.
Context Map
agentdata#harness-engineering#agent-reliability#tool-use
Uncertainty
Observed outcomes are not automatically causal: a business can improve after an action for unrelated reasons. Similar-company comparisons help, but the validity of the advice still depends on data quality, matching assumptions, and the chosen outcome measures.