Category Glossary
THE WORDS THE UPSTARTS USE.
What they actually mean.
Every platform in the restaurant AI category runs on a short list of words: autonomous, agentic, AI-native, predictive, always-on, contextual intelligence. The words describe a product. The product shipped is often a different product. This glossary reads each term honestly — how the category uses it, how we use it, and why the difference matters to an operator writing the check.
Autonomous AI
// Category use
The marketing claim implies action without human approval. The architecture ships alerts, insights, and recommendations that still require an operator to act.
Read →
Execution Layer
// Category use
An execution layer is the architectural component that takes action based on signal, within defined parameters, without a human approval step. It is not a feature. It is a design choice made at the architecture stage.
Read →
AI-Native
// Category use
In practice, AI-native in this category means the platform was built using modern data infrastructure (data warehouses, embeddings, frontier model APIs) and ships an AI-generated narrative layer over the output. The underlying architecture — data ingestion, processing, dashboarding — is the modern data stack that any technology company uses.
Read →
Agentic AI
// Category use
Agentic implies an AI agent — a system with the capability to take action within an operational context. The architecture shipped by most platforms marketed as agentic is advisory: the agent produces outputs that humans act on.
Read →
Predictive Scheduling
// Category use
Predictive scheduling in 2026 is a commodity ML capability. Demand forecasting, labor optimization, and shift suggestion are well-understood problems with published solutions. The accuracy of the predictions has improved. The architectural model — system predicts, human approves — has not changed.
Read →
Proprietary Corpus
// Category use
A proprietary corpus is a training dataset that competitors cannot replicate by buying access to data providers or integrating with standard tech stacks. For behavioral intelligence in hospitality, a proprietary corpus would include observed behavioral decisions at scale in environments where those decisions could be studied — stadiums, theme parks, large venues — over many years.
Read →
Frontier Model Wrapper
// Category use
A frontier model wrapper takes a public LLM API — OpenAI, Anthropic, Google — adds a domain-specific system prompt, and presents the output as proprietary AI. The system prompt is configurable; the model is not. When the underlying model is deprecated or changes behavior, the wrapper product changes with it. That roadmap is not owned by the wrapper company.
Read →
Always-On AI
// Category use
Always-on describes the availability of the platform. It does not describe proactivity. A platform that is always available to be asked is not a platform that initiates action. The continuous monitoring produces continuous output. The operator receives the output when she has bandwidth to process it. That is often not during the moments the output is most operationally relevant.
Read →
Contextual Intelligence
// Category use
Contextual intelligence is a UX layer over existing analytics. The same underlying signals are surfaced with better prioritization and fewer irrelevant notifications. This is a real improvement over undifferentiated alert firehoses. It is not an execution layer.
Read →
Empathic Intelligence™
// Category use
Empathic Intelligence is the architectural principle that any system serving a human operator must model what the operator is carrying — her cognitive load, her physical location in the operation, her current constraints — and calibrate its output to her capacity. Not just classify a signal by urgency. Classify it by who can receive it, act on it, and at what cost to her bandwidth.
Read →
Data Broker Test
// Category use
The data broker test asks: could a well-funded competitor purchase or aggregate equivalent training data for your AI platform through commercial data providers? If yes, the training corpus is commodity. The AI layer is differentiated only by model access and go-to-market. That is a reversible advantage.
Read →
Compliance by Design
// Category use
A compliance-by-design system treats labor laws, break requirements, minor restrictions, and similar constraints as non-overridable structural inputs. The system does not produce a schedule that violates them. The approver cannot choose to publish a non-compliant schedule because the option is not presented.
Read →
Start Here
The Three Words That Separate
the Category.
If you read nothing else: autonomous, execution layer, and data broker test. The first is the word the category abuses. The second is the architecture the category did not build. The third is the diagnostic that separates defensible AI from commodity AI.
Application Review
Most operators who apply
will not be selected.
We work with operators whose operation, culture, and competitive position fit what we built this for. We review every application individually. We select from the backlog.
If you are reading this because a competitor sent it to you, they may already be in production. We don’t confirm or deny active deployments.
Applications reviewed individually · Not all are accepted