Intuit learned to build AI agents for finance the hard way: Trust lost in buckets, earned back in spoonfuls
Building AI for financial software requires a different playbook than consumer AI Intuits the latest version of QuickBooks provides an example.
The company announced Intuit Intelligence, a system that orchestrates specialized AI agents across the QuickBooks platform to perform tasks including sales tax compliance and payroll processing. These new agents complement the existing accounting and project management agents (which have also been updated) and provide a unified interface that allows users to query data through QuickBooks, third-party systems, and uploaded natural language files.
The new development follows years of investment and improvements in Intuit’s GenOSallowing the company to build AI capabilities that reduce waste latency and accuracy improvement.
But the real news isn’t what Intuit built – it’s how they built it and why their design decisions will make AI more useful. The company’s latest AI rollout represents an evolution that builds on hard-won lessons about what works and what doesn’t when deploying AI in financial contexts.
What the company learned is sobering: Even when the auditor improved transaction categorization accuracy by an average of 20 percentage points, they still received complaints about errors.
“The use cases we’re trying to solve for customers include tax and finance; if you make a mistake in this world, you lose customer trust in buckets and we only get it back in spoonfuls,” Joe Preston, Intuit’s VP of product and design, told VentureBeat.
The architecture of trust: real data queries over generative answers
Intuit’s engineering strategy revolves around a fundamental design decision. For financial questions and business information, the system requests factual data, rather than generating answers through large language models (LLMs).
Also crucial: that data isn’t all in one place. Intuit’s technical implementation allows QuickBooks to ingest data from multiple different sources: native Intuit data, third-party OAuth-connected systems like Square for payments, and user-uploaded files such as vendor price list spreadsheets or marketing campaign data. This creates a unified data layer that AI agents can reliably query.
“We’re actually interrogating your real data,” Preston explained. “That’s very different than just copying and pasting a spreadsheet or a PDF and pasting it into ChatGPT.”
This architectural choice means that the Intuit Intelligence system functions more like an orchestration layer. It is a natural language interface for structured data operations. When a user asks about expected profitability or wants to run payroll, the system translates the natural language query into database operations based on verified financial data.
This is important because Intuit’s internal research has revealed the widespread use of shadow AI. When questioned, 25% of accountants using QuickBooks admitted they were already copying and pasting data into ChatGPT or Google Gemini for analysis.
Intuit’s approach views AI as a mechanism for translating and orchestrating queries, not as a content generator. This reduces the hallucination risk that has plagued AI implementations in financial contexts.
Explainability as a design requirement, not an afterthought
Beyond the technical architecture, Intuit has made explainability a core user experience for all its AI agents. This goes beyond just providing correct answers: it means users need to see the reasoning behind automated decisions.
When Intuit’s accounting agent categorizes a transaction, it doesn’t just display the result; it shows the reasoning. This isn’t marketing copy about explainable AI, it’s an actual user interface displaying data points and logic.
“It’s about closing that trust loop and making sure customers understand the why,” Alastair Simpson, Intuit’s VP of design, told VentureBeat.
This becomes especially critical when you consider Intuit’s user research: While half of small businesses describe AI as useful, nearly a quarter have not used AI at all. The explanation layer serves both populations: building trust for newcomers, while giving experienced users the context to verify accuracy.
The design also enforces human control at crucial decision points. This approach extends beyond the interface. Intuit connects users directly to human experts, embedded in the same workflows, when automation reaches its limits or when users want validation.
Navigate the transition from forms to conversations
One of Intuit’s more interesting challenges involves managing a fundamental shift in user interfaces. Preston described it as one foot in the past and one foot in the future.
“This isn’t just Intuit, this is the market as a whole,” Preston said. “Today we still have a lot of customers filling out forms and going through tables full of data. We’re investing a lot in exploring and questioning the ways we do this in our products today, where you’re really just filling out form after form, or table after table, because we see where the world is going, which is really a different form of interaction with these products.”
This creates a product design challenge: how do you serve users familiar with traditional interfaces while gradually introducing conversational and agentic capabilities?
Intuit’s approach was to embed AI agents directly into existing workflows. This means that users are not forced to adopt entirely new interaction patterns. The payment agent appears next to the billing workflows; the accountant improves the existing reconciliation process rather than replacing it. This step-by-step approach lets users experience AI benefits without abandoning familiar processes.
What enterprise AI builders can learn from Intuit’s approach
Intuit’s experience deploying AI in financial contexts highlights several principles that apply broadly to enterprise AI initiatives.
Architecture is important for trust: In domains where accuracy is critical, consider whether you need to generate content or translate data queries. Intuit’s decision to treat AI as an orchestration and natural language interface layer dramatically reduces the risk of hallucinations and avoids using AI as a generative system.
Explainability should be designed into, not bolted on: Showing users why the AI made a decision is not optional when trust is at stake. This requires conscious UX design. It can limit model choices.
User control maintains confidence during accuracy improvements: Intuit’s accounting agent improved categorization accuracy by 20 percentage points. Still, retaining the ability to override users was essential to adoption.
Gradually move away from familiar interfaces: Don’t force users to leave conversation forms. First, integrate AI capabilities into existing workflows. Let users experience benefits before asking them to change their behavior.
Be honest about what is reactive versus proactive: Today’s AI agents primarily respond to prompts and automate defined tasks. True proactive intelligence that makes unsolicited strategic recommendations remains an evolving capability.
Address staff concerns with tools, not just messages: If AI is intended to augment rather than replace employees, equip employees with AI tools. Show them how to use the technology.
For companies working on AI adoption, Intuit’s journey provides clear guidance. The winning approach prioritizes reliability over capability demonstrations. In areas where mistakes have real consequences, that means investing in accuracy, transparency, and human oversight before sophisticated conversations or autonomous action.
Simpson summarizes the challenge succinctly: “We didn’t want it to be a bolt-on layer. We wanted customers to be in their natural workflow and agents doing work for customers, embedded in the workflow.”
