Adding an AI agent to established software is not the same as giving users a new chat box. Toast’s approach starts with the work restaurant operators already do, then places AI where it can answer a question or help complete a task. An AI agent is a tool designed to carry out tasks, not just provide information. For a platform that handles transactions and daily operations, the key product decision is where that added ability helps—and where existing habits and safeguards should remain intact.
Start with the operator’s day
Toast’s restaurant platform spans payment terminals, handheld devices, kitchen coordination, inventory, scheduling and ordering. Its point-of-sale system, or POS, records restaurant purchases. A manager may use these tools while walking the floor during a shift or while doing prep work in a back office. Those settings matter: a useful AI entry point has to fit the task and the moment, not require an operator to stop work and explore a separate interface.
Restaurant operators also work with slim margins. That makes labor time saved, sales gained and expenses reduced more meaningful measures of an AI feature than the novelty of its responses. Product teams need to understand where a restaurant earns and spends money before deciding which tasks to automate.
The existing platform presents another constraint: customers already know how to use it. Replacing familiar flows all at once could make routine work harder. Toast has instead put contextual entry points into the app, with prewritten starter questions tailored to the page a user is viewing and the time of day. Those prompts address the blank-page problem: an empty chat field gives users little indication of what the system can do.
▲ AI access during restaurant work
Test new behavior before broad release
Toast began developing its AI capabilities in late 2023 without locking in a final feature specification. It recruited ten restaurant customers as design partners and kept a direct feedback loop with them in a WhatsApp group. An initial mobile reporting assistant went to that small group, allowing the team to revise it through roughly a year of close iteration.
Underlying AI models improved enough by early 2025 to support a wider rollout, and Toast IQ became generally available in October 2025. The sequence shows a practical way to manage risk in mature software: use a contained group to learn what operators actually ask for before exposing a developing feature to the broader customer base.
That caution matters because a slow, inaccurate or buggy feature can damage trust in software a restaurant relies on every day. It also applies after launch. As AI capabilities and connections to other parts of the platform expand, users may need fresh guidance even when the interface looks much the same.
Move from answers to useful work
Toast IQ illustrates the progression from retrieving information to helping an operator act on it. In a demonstration environment, a performance request returned $5,323.75 in net sales from 87 orders and 97 guests on Monday, September 21, 2026. The average check was $61.19. A follow-up request compared sales with the prior Monday:
| Date | Net sales |
|---|---|
| September 14, 2026 | $4,140.92 |
| September 21, 2026 | $5,323.75 |
| Change | +$1,182.83 (+28.6%) |
The comparison attributed the gain to larger average checks despite lower guest volume. Conversational access to those figures can spare an operator from navigating several reporting screens. Early use also showed that less tech-savvy operators took to the assistant quickly: asking a question in ordinary language was easier for them than finding their way through software menus.
The next step was closer to an agent’s task. Toast IQ identified leading revenue items, including Pre-Paid Dinner and Pepperoni Flatbread, then retrieved eight existing flatbread items when asked about the menu. After suggesting a spicy sausage and roasted pepper flatbread concept, it produced a menu-item card with fields for the item, its menu group and its price. The price was entered at $14.99 and then changed to $15.99.
Crucially, the card staged the change for operator approval before publication to the POS. The AI could prepare the work without silently making a live menu change. That boundary lets an established workflow gain a faster route to action while keeping the operator in control of the final decision.
▲ Operator review before a menu change
Toast IQ Grow applies the same outcome-focused approach to marketing. It helps restaurants create email and SMS campaigns and connects those communications to purchases recorded through the POS. One customer using it reduced spending on outside marketing agencies by more than 70%. That figure comes from a single case and is not an outcome every restaurant should expect. The product’s value lies in connecting the work to a business outcome an operator can assess.
Keep the product team close to the work
Adding agents changes what teams must understand. Consumer AI tools can raise expectations for what business software should do, but restaurant workflows still demand reliability. Product leaders need to observe kitchen and front-of-house operations, understand the financial pressures behind a task and stay close enough to the technology to judge what a new capability can safely handle.
Toast’s product leadership approach includes periodically taking on individual contributor work and, for critical efforts such as Toast IQ, working directly with the frontline team. That hands-on involvement may help leaders see how AI changes development speed and product behavior rather than relying only on plans and status reports.
What to change—and what to preserve
For teams adding agents to established software, the priorities are concrete:
- Identify a frequent customer task with a measurable cost or revenue consequence.
- Test an early approach with a small group before changing the experience for everyone.
- Put clear, contextual starting points inside the workflows customers already use.
- Let AI prepare actions, but keep consequential changes subject to user review.
- Return to customers’ workplaces as capabilities change, and explain what the product can now do.
Toast IQ’s example suggests that the most useful shift is not from old software to an entirely new interface. It is from familiar tools that only display information to familiar tools that can help operators complete work—without giving up the reliability and control those operators depend on.