Intuit was an early pioneer within the utilization of agentic AI, however its path to success has hardly been a straight line.
At VB Rework 2026, Intuit VP of AI Nhung Ho described how the corporate rebuilt its agent structure twice within the span of about 4 months, first shifting from a fleet of specialist brokers to a central orchestration layer, then abandoning that layer for a abilities and instruments primarily based system as soon as the orchestrator itself began failing underneath its personal complexity. The complete second rebuild took 60 days, with a primary working model in underneath 20.
The failure mode that compelled the second rewrite was particular. Brokers within the orchestrated system handed outcomes to one another in pure language, and every handoff misplaced context the following agent wanted to behave appropriately.
"If in case you have 10 brokers and so they all are passing to one another, each time that go occurs, error compounds," Ho stated.
Why the orchestration layer broke down
Ho stated the unique push towards specialist brokers got here from an easy buyer grievance. A fleet of succesful brokers continues to be one thing a buyer has to handle, deciding which agent to make use of for which activity. Intuit's reply was a system that might take a activity and route it internally, with out asking the client to choose an agent themselves.
That orchestration layer held up for about three months, which Ho described solely half joking as roughly a 12 months within the compressed timeline of agent improvement in 2026.
It broke for a structural motive slightly than a capability one. Passing outcomes between brokers in pure language meant every downstream agent needed to infer how the upstream agent reached its conclusion, and that inference degraded with every further hop. A ten agent chain didn’t fail often, it compounded errors by design.
That prognosis is what despatched Intuit again to a abilities and instruments structure.
The 60-day rebuild, and what it took to get engineering buy-in
Rebuilding a manufacturing agent system in 60 days required greater than an architectural choice. Ho stated the more durable drawback was inner, convincing each management and the engineers who had constructed the unique brokers that scrapping latest work was the suitable name.
The pitch to management relied on proof slightly than argument. Ho's staff constructed a demo of the brand new structure utilizing actual buyer queries pulled from manufacturing, then confirmed it performing higher than the prevailing system on the identical duties.
"The very best proof, at the very least my perception, is what are clients making an attempt to do? And no matter system you construct wants to deal with these issues," Ho stated.
Profitable over engineering required a unique case. Lots of of engineers exterior Ho's core staff had constructed the specialist brokers being retired, and the ask was to take their brokers aside into particular person abilities and instruments as a substitute.
Ho stated the motivating argument was scale. A standalone agent solved one slim drawback, whereas a shared talent or instrument constructed into the brand new structure might serve each buyer who touched that a part of the product. That shift additionally modified what companion groups had been answerable for daily, shifting their focus from constructing brokers to working evals, since evals turned the one technique to measure whether or not the brand new structure was truly working.
Bringing a human into the loop, and suggestions at a unique scale
The clearest buyer going through results of the rebuild is a function that lets a stay agent dialog pull in a human — although it's at present in early testing, stay to about 1% of Intuit's buyer base. "We're going to be scaling it up within the subsequent few weeks," she stated.
Ho stated a buyer can herald an Intuit product help particular person mid dialog, or their very own accountant, or one in all Intuit's personal bookkeepers, and that particular person joins with the complete context of what the agent has already achieved.
Ho drew a direct distinction with how most AI chat merchandise deal with the identical scenario. A normal objective assistant answering a tax query usually ends with a disclaimer to seek the advice of an expert. Intuit's system is constructed to attach the client to that skilled immediately, inside the identical dialog.
That human handoff sits alongside a permissions mannequin constructed for monetary information particularly. Each motion an agent takes on a buyer's monetary information requires specific permission first, although Ho stated that requirement can ease over time as clients construct belief within the system. Intuit retains an audit log of every part an agent does that may be reversed if wanted.
Suggestions within the agentic AI period
The rebuild additionally modified how Intuit gathers and makes use of suggestions, a shift Ho stated is qualitatively totally different from what got here earlier than.
"Suggestions up to now was very, very sparse, and it was additionally very bimodal," Ho stated. "Both they beloved it or they hated it, and often it tends in the direction of the damaging."
In a chat primarily based system, each dialog features as suggestions, which Ho stated moved the corporate from roughly 0.3% of consumers ever giving specific suggestions to one thing near 100%.
Ho stated she has returned to writing code herself particularly to construct fashions that analyze that suggestions quantity systematically, searching for the place the system is falling brief at a scale no guide evaluate course of might sustain with.
That quantity comes with a tone most product groups aren't used to listening to immediately. Prospects inform the agent precisely the place it failed, in plain phrases.
"They straight up let you know, 'You suck. I hate this. This isn’t proper,'" Ho stated. "However they're additionally prepared to present the methods grace and proper it as nicely, and so the onus is on all of us to reap this new piece of suggestions and kind of suggestions, and truly enhance the system."
