Synthesis Series
Starting With Human Experience, Not Technology#
A 45-year-old woman works two part-time retail jobs totaling 65 hours monthly. She cares for her elderly mother with dementia 20+ hours weekly. She has episodic migraines that occasionally prevent work. She lives in a rural county with one bus line. Her phone is a prepaid smartphone with limited data. She has a 10th grade education and limited English proficiency.
Traditional question: “How can AI help her comply with work requirements?”
Right question: “What does her experience need to be for her to maintain healthcare coverage without losing her job, abandoning her mother, or sacrificing her health?”
Work backward from that experience to technology + humans + policy. Not AI alone. Not even technology alone. A coordinated tapestry of data infrastructure, machine learning, agentic AI, user interfaces, human support, and policy flexibility, all designed for someone at the intersection of multiple barriers.
This is intersectional design. She doesn’t have one problem. She has compounding problems that interact: Limited employment documentation capacity + caregiving responsibilities + episodic disability + rural isolation + digital access barriers + language barriers. Systems designed for people with one barrier fail people with multiple intersecting barriers. Technology enabling ecosystem coordination must work for her, not just for people with simpler situations.
The Coordination Problem#
A multiply-burdened member approaches deadline needing: employment verification from two part-time jobs, documentation of caregiving for elderly parent, medical exemption for episodic condition, connection to job training for hour shortfall, transportation to verification appointment.
Each organizational silo has partial information at different times. MCO knows medical condition but not caregiving. Community organization knows job training but not transportation barrier. Provider understands episodic condition but not how to document functional impairment. State sees missed deadline but not coordinated compliance effort.
Each stakeholder acts locally but doesn’t coordinate globally. Member falls through gaps between well-intentioned organizations.
Traditional solutions (more meetings, shared databases, coordination protocols) help at margins but don’t scale. You can coordinate three organizations with phone calls. Not thirty organizations across 18.5 million people.
This coordination failure is the problem AI can help solve. But coordination technology is useless without the ecosystem infrastructure it coordinates: trusted community organizations providing actual support, providers willing to document exemptions, employers cooperating with verification, states building accessible rather than punitive systems, and members having genuine pathways to meeting requirements. AI orchestrates the ecosystem. It doesn’t replace it.
Agentic AI as Coordination Layer#
Agents sit between stakeholders and orchestrate toward shared intent. Not replacing stakeholders. Coordinating them toward outcomes everyone wants: maintained coverage, supported employment, protected health.
Member approaching deadline has agent seeing across organizational boundaries. Agent knows from MCO data about chronic condition requiring exemption. From CHW notes about caregiving for parent. From state system about approaching deadline. From employer APIs that Job A verified but Job B hasn’t submitted. From transit data that member’s route requires two transfers, 90 minutes.
Agent acts without waiting for member to navigate separately:
Initiates medical exemption by sending structured data to provider EHR, pre-populating form with clinical information from claims. Provider receives checkbox attestation, not detailed narrative. The agent assembled supporting documentation.
Contacts Job B payroll system through API, requests verification directly through MCO.
Identifies eight-hour shortfall, searches qualifying volunteer opportunities within one bus transfer, sends personalized list matching schedule constraints.
Notifies MCO care coordinator this member needs proactive outreach. High risk despite compliance intent.
Sends member consolidated message: “Medical exemption processing, Job A verified, Job B confirming tomorrow, three volunteer opportunities near you if needed.”
Every stakeholder gets information when needed in actionable format. Provider gets pre-populated form for quick review, not overwhelming documentation requests. Employer gets automated API request through payroll system, not calls from confused members. Community organization gets referral with specific needs identified. State sees coordinated compliance with partial documentation submitted.
What AI Cannot Replace#
AI cannot build trust. Members experiencing homelessness, mental health crises, or domestic violence need human relationships, not algorithms. The multiply-burdened member who finally accepts help after six months of care coordinator outreach didn’t respond to optimized messaging. They responded to consistent human presence demonstrating genuine care.
AI cannot create resources that don’t exist. If a rural county has no childcare, no job training programs, and no public transportation, AI can’t coordinate access to services that aren’t there. Technology orchestrates existing ecosystem capacity. It doesn’t manufacture capacity.
AI cannot navigate informal economy. Cash work, family caregiving, informal childcare exchanges don’t generate data trails AI requires. The grandmother caring for three grandchildren while their parents work has no digital documentation. No algorithm fixes that. Human attestation and community validation remain essential.
AI cannot exercise clinical judgment. Assessing functional capacity for work requires nuanced judgment AI can’t replicate. Can someone with episodic bipolar disorder work 80 hours monthly? The answer varies by treatment status, current episode, medication adherence, social support. There’s no algorithmic answer. Providers must make these determinations.
AI cannot address root causes. AI optimizes within broken systems but can’t fix underlying issues: insufficient jobs in rural areas, lack of affordable childcare, inadequate public transportation, housing instability, low wages requiring multiple part-time jobs. Technology doesn’t create jobs or childcare capacity.
AI should not make value judgments. Decisions about who deserves exemptions, how strictly to enforce requirements, how much to accommodate disability are ethical choices that shouldn’t be delegated to algorithms. Michigan’s $47 million automated fraud detection wrongly accused 48,000 unemployment recipients. Algorithmic decision-making in benefits administration has terrible track record.
The ecosystem solution requires: adequate community resources, human relationships built on trust, flexible policies accommodating real-world complexity, governance ensuring fairness, and sustained investment in both infrastructure and people. AI coordinates this ecosystem. It doesn’t substitute for it.
Aligned Intent as Foundation#
This only works if stakeholders share genuine intent. Technology orchestrates toward shared goals; it can’t resolve value conflicts.
States want verification integrity without overwhelming burden. MCOs want coverage stability and population health. Providers want patients maintaining coverage without paperwork drowning. Community organizations want members meeting requirements. Employers want compliance without excessive burden. Members want coverage while managing competing demands.
Not perfectly aligned, but compatible. Everyone loses when documentation-capable people lose coverage due to coordination failures. States don’t save money when people churn off and return sicker. MCOs don’t profit from preventable hospitalizations. Providers don’t benefit when patients lose primary care access. Community organizations don’t succeed when those they serve lose healthcare.
The problem is coordination failure across boundaries, not conflicting values. AI solves coordination problems. It can’t solve value conflicts.
Where interests genuinely conflict, AI becomes dangerous. If states want to minimize enrollment regardless of health impact, AI coordination makes harm more efficient. If MCOs avoid high-cost members regardless of outcomes, AI accelerates cream-skimming.
Foundational requirement: genuine stakeholder commitment to shared positive intent: maintaining coverage for people who should have it, supporting meaningful work, protecting vulnerable populations, reducing unnecessary burden.
With that commitment, AI multiplies effectiveness. Without it, AI multiplies harm.
The State as System Integrator#
States must be system integrators, not just regulators. They control verification systems, exemption processes, compliance timelines. They have regulatory authority over MCOs, oversight of providers, contracts with community organizations. They’re the only entity with formal connection to every stakeholder.
State technology must enable coordination, not just check compliance. Instead of portals passively waiting for documentation, build API-first architectures enabling agents to submit verification on members’ behalf, query exemption status, request deadline extensions, coordinate across stakeholders.
Most state systems are built on enforcement logic: make it hard enough to game that only legitimate compliance gets through. That logic creates barriers for legitimate compliance too.
API-first architecture with AI orchestration flips this. Maintain security through cryptographic verification and audit trails, not through making submission difficult. Enable agents to coordinate on behalf of members, with human oversight for unusual situations.
States building open APIs allowing MCO agents, CBO agents, and provider agents to interact with verification systems create infrastructure for ecosystem coordination. States building closed systems requiring human-mediated submission at every step prevent coordination and ensure dysfunction.
The political challenge: open, coordination-friendly architecture requires trusting stakeholders share aligned intent. Enforcement-oriented architecture assumes adversarial relationships. The choice reveals whether states view work requirements as reciprocal obligations within functioning social contract or barriers to overcome.
MCOs as Ecosystem Participants#
MCOs have data streams from multiple sources: claims showing health status, care coordination showing social complexity, pharmacy showing medication adherence, direct engagement showing communication patterns. Positioned to develop sophisticated AI orchestration.
But MCO agents can’t orchestrate alone. They need to connect to state systems, provider systems, employer systems, community organization systems.
The temptation is proprietary systems optimizing for MCO interests: maximize retention, minimize costs, improve quality metrics, reduce administrative burden. All legitimate. But optimizing within MCO boundaries doesn’t achieve ecosystem coordination.
The alternative: agents designed for cross-boundary orchestration. MCO agents sharing appropriate information with CBO agents, coordinating with provider agents on exemption documentation, interfacing with employer agents on verification, reporting to state agents on member status.
Requires data sharing agreements, privacy protection, authorization frameworks, governance structures. Requires MCO willingness to share information benefiting members even without direct bottom-line benefit.
Business case: ecosystem-level coordination reduces costly churn, prevents catastrophic outcomes, differentiates plans managing complexity from those struggling. But only works if other participants reciprocate.
Providers as Documentation Partners#
Providers control medical exemption documentation. Multiply-burdened members qualifying for exemption can’t maintain coverage without provider attestation.
But providers have no incentive to drown in paperwork. Safety-net clinics are already overwhelmed. Adding work requirement exemption documentation drives burnout and dysfunction.
AI orchestration addresses this through pre-populated exemption requests. Instead of blank forms requiring detailed narrative, providers receive structured requests with clinical information already assembled from claims, previous notes, medication lists. Provider role becomes verification and attestation, not documentation from scratch.
Agent sends structured message to provider EHR: “Patient Jane Smith has documented history of conditions potentially qualifying for exemption. Based on claims, pharmacy, clinical notes: [list]. Do you attest that due to these conditions, this patient cannot consistently meet 80-hour monthly requirements? Yes/No. Optional context [text box].”
Two minutes instead of thirty. Higher documentation quality because comprehensive data, not provider recall. Less burdensome because agent did assembly work.
Requires EHR integration, technically feasible through FHIR but operationally challenging because of vendor relationships, security concerns, workflow integration. Early partnerships with forward-thinking health systems pilot integration and demonstrate value before scale.
Provider benefit: reduced burden. Ecosystem benefit: higher exemption documentation rates, preventing inappropriate coverage loss for medically vulnerable populations.
Community and Faith Organizations as Local Connectors#
CBOs and faith groups have trust relationships and local knowledge formal institutions lack. They know which families care for which elders, who’s in recovery, who faces housing instability, which barriers are most acute.
But they typically lack sophisticated technology infrastructure. Limited budgets, volunteer staff, minimal IT capacity.
Solution: provide agent infrastructure they can configure without requiring technical expertise. National platform CBOs and faith organizations adapt to their communities, with agents learning local patterns while sharing insights across communities.
CBO agents interface with MCO, state, and member-facing agents. MCO agent identifies member needing local support, routes to appropriate CBO agent based on geography and capacity. CBO identifies member with medical exemption need, agent communicates with provider agent to initiate documentation.
Coordination happens across organizations without requiring CBOs to build technical infrastructure. Agents handle routing, status tracking, deadline monitoring, information sharing. CBO staff focus on relationships and support.
Faith organizations have distinct advantage: regular in-person connection through worship, community meals, pastoral care. Agents leverage this by providing actionable information: “Three congregation members have verification deadlines next week,” “Two families need transportation to appointments,” “One member qualifies for caregiver exemption but hasn’t applied.”
Faith organization doesn’t need to understand administrative complexity. Agent provides actionable information in context of relationships they already maintain.
Community Inclusive Social Enterprises as Service Coordinators#
CISEs, organizations employing vulnerable populations while building community capacity, are unique ecosystem participants. Simultaneously employers (providing work), community organizations (offering support), and service coordinators (connecting to resources).
CISE agents coordinate across multiple domains. When CISE employs someone subject to work requirements, agent verifies employment hours directly to state systems, coordinates with MCO agents on health needs affecting work capacity, connects to other employer agents for multiple jobs, interfaces with CBO agents for additional support.
More powerfully, CISE agents identify opportunities for peer support earning qualifying hours. When one member successfully navigates requirements and maintains coverage, agent matches them with others facing similar challenges. Hours spent providing peer navigation count toward navigator’s work requirements, creating positive feedback loops.
Ecosystem benefit: converting administrative burden into community capacity-building. Instead of every vulnerable person struggling alone, successful navigators support others while meeting their own requirements. AI orchestration makes this matching and coordination possible at scale.
Employers, Caregivers, and Individuals#
Employers as Verification Partners: Large employers using standard payroll systems (ADP, Paychex, Gusto) integrate with agent ecosystem through API connections, automatically verifying hours to state systems via MCO or CBO intermediaries. Small employers lacking IT infrastructure use simplified interfaces like web forms or text confirmation. “Was Jane Smith employed by your business in April? Reply YES/NO.” Agent handles submission to verification systems. For gig platforms, coordination requires bulk data sharing agreements with major platforms. Platform agents provide aggregated earnings without exposing customer information, respecting privacy while enabling verification. Employer benefit: reduced administrative burden. Automated agent verification reduces costs while improving data accuracy.
Caregivers as Information Sources: Family caregivers for children, elderly parents, or disabled family members often qualify for exemptions but struggle to document caregiving. AI agents enable validated self-attestation networks. Individuals attest to caregiving responsibilities with confirmation from care recipients, family members, or community witnesses. Agent manages attestation workflow with cryptographic verification. Care recipient’s provider agent confirms functional limitations requiring assistance without detailed care logs. Validates caregiving need without invading family privacy. Ecosystem benefit: recognizing caregiving as legitimate work incompatible with 80-hour monthly formal employment.
Individuals as Ecosystem Participants: Members are not passive recipients. They are active participants whose agents represent their interests across boundaries. Member-facing agents translate complex requirements into actionable steps, coordinate on member’s behalf across organizations, maintain relationship history and preferences, and critically, maintain member control and consent. All actions require authorization. Members can revoke authority anytime. Agents increase agency rather than replacing it, handling coordination complexity so members focus on actions requiring human judgment. Member experience: having dedicated assistant who knows their situation, coordinates across all helping organizations, translates administrative complexity, advocates on their behalf, but follows their directions and respects autonomy.
Privacy, Consent, and Governance#
For ecosystem orchestration, information must flow across organizational boundaries. This creates substantial privacy and security concerns.
Solution: federated agent systems where organizations maintain control over their data while enabling coordination through secure interfaces. MCO agents do not get direct EHR access. They request specific information through provider agents, which validate requests and release only authorized data.
Member consent is foundational. Members authorize agents to coordinate on their behalf and specify which information shares with which organizations. Consent is granular: authorize medical sharing with MCO and primary care but not CBOs, employment verification with state systems but not other employers.
Blockchain or distributed ledger provides tamper-proof audit trails. Every data exchange logged with cryptographic verification. If member questions information sharing, audit trail provides complete history.
Zero-knowledge verification enables coordination without information sharing. Employer agent verifies to state agent that individual worked qualifying hours without revealing specific hours, earnings, or employment details. Verification is cryptographically sound without exposing underlying data.
Governance question: who sets rules for agent behavior, information sharing, coordination protocols? Requires multi-stakeholder governance where states, MCOs, providers, community organizations, and member representatives collectively establish guardrails.
Implementation Timeline#
Building ecosystem orchestration infrastructure is 2-3 year project, not 10-month sprint. December 2026 deadline is too soon for sophisticated cross-stakeholder AI orchestration. But foundations can be built now.
2025-2026: Basic Agent Infrastructure. Individual organizations deploy agents within their boundaries. MCOs build agents for member engagement and care coordination. States develop API-first verification systems. Providers pilot EHR-integrated exemption documentation. CBOs adopt configurable agent platforms. Each tests agent functionality internally, learning what works before attempting cross-boundary coordination. Goal: building capacity within organizations while designing for future integration.
2026-2027: Bilateral Coordination. Connect agents across organizational pairs. MCO agents coordinate with state agents on verification submission. Provider agents integrate with MCO agents on exemption documentation. CBO agents connect with MCO agents on referral coordination. Employer agents link with state agents on verification. Each bilateral connection establishes protocols, privacy frameworks, governance structures. Learn from early partnerships before complex multi-party coordination.
2027-2028: Ecosystem Orchestration. With bilateral connections established, build orchestration layer enabling agents to coordinate across multiple organizations simultaneously. Member-facing agents orchestrate across MCO, provider, CBO, employer, and state agents. Cross-organization workflows adapt based on member needs and real-time circumstances. Full vision of ecosystem coordination becomes operational once foundational bilateral relationships are stable and governance frameworks proven.
The Choice#
Without ecosystem orchestration, work requirements generate persistent dysfunction. Members navigate complex requirements alone, falling through gaps between well-intentioned organizations. Administrative burden overwhelms safety-net providers and community organizations. States build enforcement-oriented systems penalizing coordination failures. MCOs optimize within boundaries, unable to address system-level problems. Coverage loss concentrates among documentation-challenged populations rather than people lacking work capacity.
With ecosystem orchestration, human and technological, the same policy framework becomes navigable. Coordination happens across organizational boundaries. Members receive orchestrated support rather than fragmented interventions. Administrative burden decreases because agents handle routine coordination, freeing humans for relationships and judgment. Coverage stability improves because documentation failures decrease. The system bends toward supporting people rather than excluding them.
But this requires building the full ecosystem, not just deploying AI:
Human infrastructure: Care coordinators, community health workers, peer navigators, trusted community organizations, faith-based support networks. These are the relationships that build trust, provide emotional support, and address needs algorithms can’t see.
Resource infrastructure: Actual job opportunities, childcare capacity, transportation options, volunteer placements, training programs. AI can connect people to resources. It can’t create resources that don’t exist.
Policy infrastructure: Reasonable exemption categories, accessible documentation requirements, grace periods during transitions, graduated compliance pathways. Technology can’t fix fundamentally inaccessible policy.
Governance infrastructure: Multi-stakeholder decision-making, member voice and control, transparent audit trails, accountability mechanisms, protection against algorithmic bias. AI without governance becomes surveillance and control.
Financial infrastructure: Adequate funding for community organizations, sustainable reimbursement for providers, appropriate capitation rates for MCOs, investment in social determinants of health. Coordination technology is worthless if organizations lack resources to deliver coordinated services.
AI amplifies ecosystem capacity. It doesn’t substitute for it. The investment decision isn’t AI versus humans. It’s building comprehensive ecosystem infrastructure where AI enables coordination across well-resourced, well-governed, genuinely committed stakeholders.
The question for health insurance executives isn’t whether to invest in AI. It’s whether to invest in AI as part of comprehensive ecosystem development or AI as cheap substitute for actual infrastructure. The former transforms systems. The latter automates dysfunction.
Choose comprehensive ecosystem development with AI as coordination layer. The return on investment is measured not just in operational efficiency but in a functioning healthcare system that maintains coverage for vulnerable populations while meeting legitimate work participation expectations.
That’s the future worth building. And it requires much more than algorithms.