Cognizant Agentic AI Goes Live on Health Plan Claims

Cognizant Agentic AI is putting AI agents to work on one of the least glamorous and most expensive problems in American healthcare: the pended claim. On September 28, the company announced general availability of Workflow Agentic Processing and Enterprise Model Context Protocol (MCP) Servers for its TriZetto Facets and QNXT core administration platforms — the systems that sit underneath a large share of US health plan operations.

The pitch is straightforward. AI agents carry out a health plan's own standard operating procedures to automate processing of eligible routine pended claims. Denials, exceptions, and complex cases get routed to human reviewers when judgment is needed. Alongside the workflow engine, Cognizant is shipping a library of more than 100 MCP tools, so a health plan's own AI initiatives can plug into TriZetto data context and streamline day-to-day operations — from customer service summaries to faster prior authorization turnarounds.

Agents as first-tier users of the claims stack

What makes the Cognizant Agentic AI launch notable is not just the automation, but the architecture. Cognizant is treating AI agents as a new class of enterprise user — one that needs governed, auditable access to regulated systems. The MCP tool library is the connective tissue: instead of brittle screen-scraping or one-off integrations, agents call standard tools against the same systems of record that human staff use.

Pended claims are the right first target. These are claims that stall mid-process and require manual review before they can be finalized — pure administrative drag. Automation has been chipping away at this burden for years. The 2025 CAQH Index found that US healthcare avoided an estimated $258 billion in administrative costs in 2024 through electronic transactions and improved data exchange. Cognizant's bet is that agents can push that number further by handling the routine tail while keeping humans on the exceptions.

The design keeps clinical and consequential decisions with people. Agents execute the standard playbook; anything that smells like an exception goes to a human. For a regulated industry where a wrongly denied claim is both a compliance event and a human one, that boundary is the whole product.

Why claims, why now

The timing is not accidental. Regulators are forcing the issue. The CMS Interoperability and Prior Authorization Final Rule carries payer compliance obligations beginning in 2026, with electronic prior authorization API mandates taking effect in 2027. Health plans that still run manual-heavy workflows are on a clock.

Prior authorization has become one of the largest sources of friction in US healthcare. The American Medical Association's most recent survey found that 95% of physicians say prior authorization delays access to necessary care, and that physicians and their staff spend an average of 13 hours each week completing the requests. Every hour an agent can safely take off that pile is an hour returned to care.

This release is also the latest step in a deliberate strategy. In May, Cognizant opened TriZetto Unify to AI agents with a headless API model, treating agents as first-tier consumers of the platform, starting with electronic prior authorization. "AI agents are increasingly joining human users in enterprise workflows, and for regulated industries like healthcare, that requires a platform that is policy-governed, auditable and built on industry-standard healthcare interoperability protocols," said Prasad Sankaran, President of Cognizant AI Products and Platforms, at the time. The September launch follows the TriZetto AI Gateway and the broader agent-ready expansion of TriZetto Unify.

According to Cognizant, the release marks the latest milestone in its AI strategy for the TriZetto portfolio, following the TriZetto AI Gateway and the expansion of agent-ready capabilities through TriZetto Unify. The company positions the combination — workflow automation plus a governed tool library — as the template for how health plans move from agent pilots to production: start with the routine, instrument everything, and keep humans where judgment lives.

The pattern: governed agents go mainstream

Step back and the pattern is bigger than one vendor. The Cognizant Agentic AI deployment is one more data point in a year where the enterprise agent winners are not the flashiest demos — they are the governed deployments. Banks are wiring agents into trade lifecycles with audit trails. Payment networks are putting agent spending on live card rails. Now health plans are letting agents touch claims, with humans holding the exceptions queue.

Each of these deployments answers the same trust question differently, but the shape rhymes: scope the agent to routine work, give it standard tools instead of skeleton keys, log everything, and keep a human in the loop where judgment matters. It is the unglamorous plumbing that turns agent pilots into production systems — and it is where the real agent economy is being built, one pended claim at a time.

For health plans evaluating the move, the sensible first step is a narrow pilot: high-volume, low-risk pended claims only, with side-by-side telemetry showing exactly what the agent changed versus what humans touched. Watch the exception queue closely — misclassified edge cases are where agent deployments earn or lose trust. More of my reporting on the agent economy is on the AI News desk.

Sources: Cognizant newsroom; Morningstar / PR Newswire.