Palona AI Ditches 'Thin Wrapper' Chatbots for a 'Digital GM' to Run Restaurants

IN BRIEF

Palona AI, a startup founded by Google and Meta engineering veterans, is pivoting hard from general-purpose sales agents to a specialized AI operating system for the trillion-dollar restaurant industry. With today’s launch of Palona Vision and Palona Workflow, the company is offering a blueprint for how to build a durable AI business on the “shifting sand” of today’s large language model ecosystem.

“You’re building a company on top of a foundation that is sand—not quicksand, but shifting sand,” said Palona AI’s co-founder and CTO Tim Howes. He’s referring to the central challenge for AI founders today: the unstable, rapidly evolving ecosystem of foundation models.

This reality is what drove his Palo Alto-based company to make a decisive vertical push. Today, Palona AI launched Palona Vision and Palona Workflow, transforming its multimodal agent suite into a real-time operating system for restaurants that spans cameras, calls, and coordinated tasks. It’s a strategic pivot from the company’s debut in early 2025, when it emerged with $10 million in seed funding to build emotionally intelligent sales agents for broad D2C enterprises.

By narrowing its focus, Palona is moving beyond “thin wrappers” to build a deep, multi-sensory system that solves high-stakes, physical-world problems. “We built an orchestration layer that lets us swap models on performance, fluency, and cost,” Howes explained, revealing their strategy for navigating the LLM churn.

The 'Digital GM' That Never Sleeps

For a restaurant owner, Palona’s new offering is designed to function as an automated “best operations manager” that’s always on. Palona Vision uses a restaurant's existing in-store security cameras—no new hardware required—to analyze operational signals like queue lengths, table turnover rates, prep station bottlenecks, and cleanliness.

Palona Workflow then acts on these insights, automating multi-step processes like managing catering orders or running through opening and closing checklists. By correlating video signals from Vision with Point-of-Sale (POS) data and staffing levels, it ensures consistent execution across locations. “Palona Vision is like giving every location a digital GM,” said Shaz Khan, founder of Tono Pizzeria + Cheesesteaks, in a press release. “It flags issues before they escalate and saves me hours every week.”

Lessons in Going Vertical

Despite a leadership team with a storied pedigree—CEO Maria Zhang was a VP of Engineering at Google and CTO of Tinder, while Howes co-invented LDAP—the team’s first year was a lesson in focus. Initially serving fashion and electronics brands with “wizard” and “surfer dude” sales agents, they quickly realized the restaurant industry was a unique, trillion-dollar opportunity that was “surprisingly recession-proof” but “gobsmacked” by inefficiency.

“Advice to startup founders: don't go multi-industry,” Zhang warned. By verticalizing, Palona went from a thin chat layer to a “multi-sensory information pipeline” processing vision, voice, and text. This focus unlocked access to proprietary training data, like prep playbooks and call transcripts, that generic web scraping could never provide.

Four Technical Lessons for AI Builders

Palona's journey offers a masterclass for developers building enterprise-grade AI.

1. Build on ‘Shifting Sand’: To avoid vendor lock-in with a single provider like OpenAI or Google, Palona’s patented orchestration layer allows them to swap models on a dime. They use a mix of proprietary and open-source models, such as Gemini for computer vision, to optimize for performance and cost. The lesson: Don’t let a single vendor become your product’s core value.
2. Go from Words to ‘World Models’: Palona Vision marks a shift from understanding language to understanding physical reality. It identifies cause and effect in real-time, recognizing a pizza is undercooked by its “pale beige” color or alerting a manager to an empty display case. “In words, physics don't matter,” Zhang explained. “But in reality, I drop the phone, it always goes down... we want to really figure out what's going on in this world of restaurants.”
3. The ‘Muffin’ Solution for Memory: In a restaurant, memory is the difference between a frustrating interaction and a magical one. Finding that an open-source tool produced errors 30% of the time, the team built Muffin, a proprietary memory management system. A nod to web “cookies,” Muffin handles four layers of data: structured (allergies), slow-changing (preferences), transient (seasonal cravings), and regional (time zones).
4. Ensure Reliability with ‘GRACE’: An AI error in a kitchen can mean a wasted order or a safety risk. To prevent chaos, Palona uses its internal GRACE framework: Guardrails, Red Teaming, App Sec, Compliance, and Escalation. “We simulated a million ways to order pizza,” Zhang said, using one AI to act as a customer and another to take the order, all to measure and eliminate hallucinations.

PRISM Insight: Palona’s pivot is a playbook for the next wave of AI startups. As the gold rush for building thin wrappers on major LLMs ends, the new frontier is vertical AI—deep, domain-specific systems that solve high-stakes physical problems. The winners won't be those with the best general model, but those who can build a proprietary 'information pipeline' of vision, voice, and operational data that a general-purpose AI can't replicate.