Google Cloud Deploys Clinical AI Agent at Seattle Children's 500K-Visit Network

Google Deploys Production Clinical AI Agent at Seattle Children's

Google Cloud is running a clinical decision-support agent called Pathway Assistant across Seattle Children's 40-location network, handling over 500,000 patient visits annually. The agent connects the health system's clinical effectiveness program into Google's Gemini LLM and Vertex AI infrastructure, surfacing internal guidelines and external medical evidence at the point of care.

The deployment marks a shift from experimental clinical AI to governed, production infrastructure. Hospital CIOs now face a choice: select a major cloud provider for clinical AI, or watch clinicians adopt ungoverned tools. Google's move directly challenges Microsoft's early lead through Epic integrations and AWS HealthScribe deployments.

What Pathway Assistant Does for Enterprise Buyers

The agent pulls from Seattle Children's internal clinical pathways and external evidence repositories, presenting recommendations inside clinician workflows. This is a reference architecture for hospitals that want to own their clinical logic while offloading LLM infrastructure to a hyperscaler.

Buyers will pay incremental Vertex AI usage costs—tokens, storage, API calls—rather than standalone SaaS licenses. Professional services fees will cover wiring EMR data and proprietary guidelines into the agent. The economic model mirrors other enterprise AI deployments: variable cloud spend tied to clinical volume, not per-seat pricing.

The clinical risk calculus changes. Pathway Assistant moves decision support from shadow tools into a governed platform with hospital-owned pathways, reducing liability exposure from clinicians using consumer chatbots. But it increases dependence on a single cloud provider for mission-critical clinical workflows.

IBM Acquires Hakkoda to Build Healthcare Data Platforms for AI

IBM Consulting bought Hakkoda, a New York consultancy with over 100 data engineers specializing in Snowflake-centric healthcare data platforms. The acquisition positions IBM as a full-stack AI implementation partner, combining watsonx models with data modernization services for payers and providers.

Large health systems considering enterprise clinical AI—care management, utilization review, risk adjustment—now see IBM as a one-vendor option for data platform and AI deployment. In RFPs, IBM can field a Snowflake-native data build plus AI implementation package that competes directly with Accenture and Deloitte.

The budget impact is higher services line items. IBM's expanded bench means more projects pitched as multi-year, multi-million-dollar data-plus-AI programs rather than point tools. Buyers gain implementation capacity but increase dependence on a single global systems integrator. Vendor lock-in becomes a board-level concern when the same firm architecting your data platform also provides the AI models.

Mayo Clinic and Microsoft Build Multimodal Healthcare Foundation Model

Mayo Clinic and Microsoft announced a healthcare foundation model trained on multimodal clinical data—imaging, text, waveforms—from Mayo's petabyte-scale corpus covering 1.3 million unique patients annually. Microsoft is committing Azure AI compute, including its latest GPU clusters, as the infrastructure backbone.

This is pre-product but forces other cloud providers to show credible clinical foundation model strategies. It also pressures EHR vendors to decide whether to align with Microsoft's stack or remain cloud-agnostic. Google's Pathway Assistant deployment and partnerships like Every Cure drug repurposing represent Google's counter-positioning.

For enterprise buyers, the Mayo-Microsoft collaboration signals that foundation models trained on health system data are becoming a competitive differentiator. Health systems with large, clean data sets may partner with cloud providers to co-develop proprietary models, while smaller systems will license pre-trained models. The strategic question shifts from "do we need clinical AI" to "do we own our training data and model weights, or do we rent them."

What to Watch

Negotiate data ownership and model governance terms in cloud contracts before deploying clinical AI agents. Hospitals that treat these as standard IT procurements will discover too late that their clinical pathways are training vendor models.

Watch how Epic responds to Google's clinical agent deployments. If Epic integrates deeply with Microsoft Copilot while remaining neutral to Google and AWS, health systems locked into Epic face a de facto cloud platform decision.

Track which payers and ACOs deploy IBM's new Snowflake-plus-watsonx packages. If IBM wins large care management AI deals in the next two quarters, expect Accenture and Deloitte to accelerate similar acquisitions.