Zest puts restaurant discovery and booking inside text messages

What changed: Zest launched a personal AI restaurant concierge that works through text messages. Users can ask for recommendations based on cuisine, dietary needs, neighborhood, past preferences, or the occasion, then have the agent check availability and book through OpenTable, Resy, or SevenRooms without opening a separate app.

Why it matters: This is a practical example of agents becoming transaction interfaces rather than answer engines. Founders building consumer services should decide which parts of their product can work inside messaging, where users already coordinate with friends, instead of requiring another standalone app.

Try/watch: Test whether a conversational flow can complete one valuable transaction end to end, but monitor booking errors, user consent, and the quality of recommendations based on personal data.

DoorDash data suggests agent-led commerce is still at an early stage

What changed: DoorDash said it has received only “several hundred” orders from third-party AI agents, compared with roughly 3 million deliveries per day in its 2025 reporting. The company is experimenting with its own text-to-order and chatbot products while maintaining relationships with outside agents.

Why it matters: The headline around agents taking over online commerce is ahead of actual usage, at least on DoorDash. Buyers and operators should measure completed transactions, repeat usage, support costs, and conversion—not just the number of agent integrations or demonstrations.

Try/watch: Add agent-originated orders to analytics as a separate channel and compare their failure, cancellation, fraud, and repeat-purchase rates with ordinary app orders.

Waters connects AI agents to laboratory testing workflows

What changed: Waters introduced AI Link, software that connects large language models with laboratory software so scientists can request analytical testing tasks in ordinary language rather than navigating each function manually. The company is targeting laboratories handling increased workloads from AI-driven drug discovery.

Why it matters: This shows agentic AI moving into specialized operational environments, where value depends on reliable execution and integration with existing instruments—not simply better conversation. Technical buyers should evaluate whether agents can safely translate broad requests into validated procedures and preserve traceability.

Try/watch: Start with low-risk scheduling, setup, or data-retrieval tasks, and require approval gates and complete records before allowing an agent to control experiments or modify test parameters.