Builds an end-to-end agent execution stack on Google Cloud
Uses Compute Engine and Google Kubernetes Engine (GKE) for stable execution
Standardizes secure cross-system transactions through Universal Commerce Protocol
Drives multi-tool decisions with Gemini Enterprise (formerly Vertex AI) and the Pokee-Isaac model
Advances integration with Gemini Enterprise, Google Workspace, and BigQuery to support scalable enterprise workflows
Google Cloud enables Pokee AI’s agent commerce stack for scalable end-to-end transaction execution.
Google Cloud enables Pokee AI’s agent commerce stack for scalable end-to-end transaction execution.
Solving the execution gap for long-horizon
enterprise workflows
AI agents have made significant progress in information discovery and understanding user intent. Yet in real-world business settings, they still face a critical limitation: they can understand a task but often struggle to complete it reliably.
This challenge is especially acute in ecommerce, financial services, and enterprise workflow automation. Even a seemingly simple purchasing decision can involve complex contextual information, and agents may lose track of product comparisons, multistep purchases, cart configurations, payments, or order management. As workflows become longer and more complex, traditional large language models are more likely to fail because of limited context, incorrect tool selection, or inconsistent state management.
Ecommerce illustrates the problem clearly.
Generating even one slide could nearly exhaust the token budget, exposing the limits of context capacity. Enterprises need complete, long-running workflows, not short outputs. Our goal was smaller models, longer context, and deeper execution. Google Cloud gave us the reliability, cost efficiency, and system capabilities to build agents faster and more robustly.
Bill Zhu
CEO, Pokee AI
Connecting an agent to a merchant has traditionally required custom development, continuous synchronization of product catalogs, pricing, and inventory, plus separate security, permissions, and compliance reviews. As merchant networks expanded, this model became costly and difficult to scale. Agents could help users discover products, but often could not reliably complete transactions.
Pokee AI was built to close this execution gap. Its enterprise AI agent models and infrastructure are designed to execute complex business processes with extremely long context, not simply respond to prompts. A reinforcement learning-powered execution engine and multimillion-token context capabilities help agents maintain consistency across long, multistep workflows. The technology supports enterprise automation, investment-report generation, and ecommerce strategy execution while reducing token costs by approximately 10 times.
To support this architecture, Pokee AI works closely with Google Cloud, running its applications and proprietary inference models on Compute Engine and Google Kubernetes Engine (GKE). Google Cloud provides the reliability, scalability, security, and cost efficiency required for enterprise AI workloads.
Google’s Universal Commerce Protocol (UCP) adds another critical layer by providing a common standard for agent-driven transactions. Instead of building and maintaining a separate integration for every merchant, Pokee AI can connect to a standardized transaction layer. This helps close one of the industry’s most persistent gaps: AI reasoning has advanced rapidly, but reliable execution has not kept pace.
Deploying an agentic commerce execution stack
for seamless transactions
In ecommerce, Pokee AI uses Google Cloud to power a complete agentic commerce execution stack, enabling agents to manage the entire journey from product discovery through payment rather than stopping at recommendations or conversation. Its hybrid AI architecture combines long-context reasoning, multi-tool orchestration, reinforcement learning-based decision-making, transaction execution, and security controls in one system, allowing AI agents to complete complex, real-world business tasks more reliably.
Within this architecture, GKE serves as the core platform for the agent runtime and reinforcement learning model deployments, enabling large numbers of agents to operate reliably and scale dynamically under high concurrency. Compute Engine provides the underlying compute and transaction-processing layer, supporting high-throughput data workloads and real-time ecommerce traffic.
In ecommerce, product details, bundles, discounts, and transaction requirements create highly complex workflows. UCP helps standardize interactions with merchant systems, while Pokee Isaac preserves critical information across long contexts to support more reliable execution.
Bill Zhu
CEO, Pokee AI
Pokee AI also deploys the Pokee-Isaac model and its reinforcement learning engine on Google Cloud to handle complex semantic understanding and multistep planning while maintaining consistent decisions across extremely long contexts.
The UCP adds another critical layer. Traditionally, product information, inventory, and payment logic are distributed across separate merchant systems, requiring custom integrations and complex retrieval and indexing pipelines. UCP standardizes these capabilities, allowing agents to load complete merchant catalogs and supported actions directly into context and execute transactions without traditional retrieval workflows. This reduced merchant onboarding time from six to eight weeks to approximately three days.
By combining Virtual Private Cloud isolation with Google Cloud’s enterprise security capabilities, Pokee AI integrates UCP’s payment handler architecture with virtual card primitives issued through a licensed partner. This enables agent-controlled spending that is secure, auditable, and compliant while protecting sensitive financial information.
Ultimately, this architecture increased the completion rate for end-to-end, multistep transaction tasks from approximately 41% to 92%, while also driving an approximately 23% increase in conversion rates. These results demonstrate the real-world business impact of combining a unified protocol, long-context capabilities, and a reinforcement learning-based execution system.
Scaling operational efficiency through
unified enterprise ecosystem integration
Following the success of its ecommerce transaction use cases, Pokee AI is expanding its collaboration with Google Cloud beyond infrastructure into broader enterprise ecosystem integration. In this model, Google Cloud is not only the platform for running applications and models, but also a unified environment connecting AI reasoning, enterprise data, collaboration tools, and commercial transactions.
At the enterprise level, Pokee AI is working with Google to advance integration with Gemini Enterprise, Google Workspace, and BigQuery, with the goal of enabling agents to access transaction systems, enterprise knowledge, and analytics simultaneously. Through this combination, AI agents could go beyond executing isolated tasks to understanding business context, using enterprise data, and completing end-to-end business workflows. This capability is especially valuable in high-impact industries such as financial services, where institutions need to access data across systems while meeting strict compliance requirements. By further connecting with the Google Cloud ecosystem, Pokee AI has the potential to help enterprises reduce data silos, improve operational efficiency, and strengthen decision support.
Pokee AI is also expanding into new regions and industries. The company has launched commercial service partnerships in multiple countries, marking a shift from individual merchant pilots to regional-scale deployment and expansion into high-value sectors such as financial services, procurement, accounting, and enterprise services.
For developers and engineering teams, Google Cloud also provides a more efficient, AI-native development environment. Gemini Flash supports low-latency, high-efficiency workloads, giving teams greater flexibility to balance speed, cost, and reasoning performance based on task complexity.
Looking ahead, Pokee AI aims to help agents operate continuously in real-world business environments, scale reliably, and deliver measurable value through true end-to-end execution.
Looking ahead, Pokee AI will continue working closely with Google Cloud to advance product development, enterprise deployment, token cost reduction, and ecosystem expansion.
Bill Zhu
CEO, Pokee AI
Pokee AI builds private frontier agentic AI models for VPC, private server, and on-device deployment, powering enterprise agents across finance, ecommerce, legal, procurement, support, and regulated workflows.
Industry: AI
Location: United States
Products: Google Cloud, Compute Engine, Google Kubernetes Engine (GKE), Universal Commerce Protocol, Gemini, Gemini Enterprise, Gemini Flash, Google Workspace, BigQuery
