Research Report Outline: AI Deployment in Healthcare

The Convergence of MCP, FDE, and Claude Code

Draft Outline - January 26, 2026 Authors: Thomas Startz, Michael Status: OUTLINE - Ready for research expansion

---

Recommended Format

• Style: LaTeX/PDF for distribution to C-suite and investors
• Length: 15-25 pages substantive research (not hot takes)
• Tone: Analytical, data-driven, with practical frameworks
• Distribution: Direct to decision-makers, Substack deep-dive, VC pitch materials

---

Executive Summary (1-2 pages)

Thesis Statement

The bottleneck for enterprise AI has shifted from model capability to deployment expertise. A new convergence is emerging: Model Context Protocol (MCP) as the integration standard, Claude Code as the deployment platform, and Forward Deployed Engineering (FDE) as the delivery model. Healthcare organizations that understand this stack will capture the AI opportunity; those that don't will fall behind.

The Interface Inversion

Traditional software competed on UI/UX—interfaces were the moat. AI inverts this: interfaces become commoditized (agentic, generative, conversational), while custom backend integration becomes the differentiator. The magic is moving to the backend.

Key Claims

1. Claude Code is emerging as THE platform for AI deployment (vs. fragmented alternatives)
2. MCP is becoming the standard for AI-system integration
3. FDE model is the only viable path for complex healthcare AI implementation
4. The talent gap is widening faster than most realize
5. Market signals (commodities, hiring, VC activity) confirm this thesis

Why This Matters Now

• CMS mandates for electronic prior authorization creating urgency
• 95% of AI projects still fail to deliver value
• Healthcare orgs have budget but lack implementation talent
• First movers will establish moats through proprietary implementation

---

Section 1: The State of AI Deployment (3-4 pages)

1.1 The Implementation Crisis

Data to Include:

• MIT 2024: 95% of AI projects fail to create business value
• Only 1% of companies have reached full AI maturity
• 74% struggle to scale AI efforts
• 800% growth in FDE hiring demand

Analysis:

• Why model capability is no longer the constraint
• The "demo to production" gap
• Why standard software implementation approaches fail for AI

1.2 The Process Reality Problem

Key Insight:

"Standard operating procedures are corporate fiction: static, incomplete, and often wildly outdated. AI systems need behavioral fidelity—how the work actually gets done—not the idealized version."

Data to Include:

• Case studies of failed implementations
• The difference between documented vs. actual workflows
• Why AI requires deeper integration than traditional software

1.3 The Talent Gap

Data to Include:

• AI engineer compensation ($300-400K fully loaded)
• Healthcare org ability to attract/evaluate talent
• FDE demand vs. supply metrics
• Skill set requirements: domain expertise + technical depth + deployment capability

---

Section 2: The Emerging Stack (4-5 pages)

2.1 Claude Code as Deployment Platform

Thesis: Claude Code is converging as the dominant platform for AI deployment, similar to how AWS became the default for cloud infrastructure.

Evidence:

• Adoption patterns: Cursor vs. Codex debate "settled"
• Non-technical tasks now running on Claude Code (prior auth, company operations)
• Developer sentiment and adoption curves
• Anthropic's strategic positioning

Market Signals:

• Gold/silver/copper prices as AI infrastructure proxy
• Semiconductor and data center investment patterns
• VC funding flows

2.2 Model Context Protocol (MCP)

What is MCP:

• Technical explanation (accessible to non-technical readers)
• How it enables AI-system integration
• The "USB for AI" analogy

Why It Matters:

• Standard integration patterns reduce implementation friction
• Healthcare-specific MCPs: CMS Coverage, ICD-10, NPI Registry, PubMed
• Enables composable AI systems vs. monolithic solutions

Adoption Evidence:

• Vanta, Gmail integrating MCP servers
• Healthcare MCP ecosystem emerging
• Open standard vs. proprietary integration

2.3 Forward Deployed Engineering Model

Historical Context:

• Palantir's origination of the FDE model
• The "French restaurant" analogy
• Why traditional consulting fails for AI

What FDEs Do:

• Embed on-site with customers (days to months)
• Write production code on customer infrastructure
• Map actual workflows (not documented processes)
• Build trust with end users
• Feed learnings back into core product

Economics:

• Minimum viable contract size ($100K+ for healthcare)
• Revenue model: services drive product adoption
• Warning signs: body shop vs. platform company

---

Section 3: Healthcare-Specific Application (4-5 pages)

3.1 Why Healthcare, Why Now

Regulatory Tailwinds:

• CMS electronic prior authorization mandates
• Interoperability rules
• Value-based care transition

Pain Points:

• Utilization management backlogs creating patient harm
• Nurse burnout and turnover (40%+ in some systems)
• Manual review processes as bottleneck
• Compliance gaps in audit trails

Automation Potential:

• Revenue Cycle Management (RCM): 90%+ automatable
• Utilization Management (UM): 90%+ automatable
• Care Management: 70%+ automatable

3.2 Utilization Management as Use Case

The Problem:

• Average 7-14 days for manual processing
• Nurse time spent on administrative vs. clinical work
• Inconsistent determinations across reviewers

The AI Solution:

• Chart summarization and evidence extraction
• Criteria matching (MCG, InterQual, payer-specific)
• Documentation support and audit trails
• Human-in-the-loop for complex cases

Implementation Requirements:

• HIPAA compliance (Claude now HIPAA-eligible)
• EHR integration patterns
• Clinical workflow mapping
• Change management

3.3 The Skill Set Barrier

Required Expertise (simultaneously):

1. Clinical domain knowledge (InterQual, MCG, UR workflows)
2. AI/ML fundamentals (retrieval augmented generation, agents)
3. Modern deployment (Vercel, cloud infrastructure)
4. Security/compliance (HIPAA, cybersecurity)
5. Change management (workflow transformation)

Why This Is Rare:

• Traditional healthcare IT knows clinical but not AI
• AI engineers know tech but not healthcare
• Consultants advise but don't implement
• The FDE model bridges all gaps

---

Section 4: Market Positioning & Competitive Landscape (2-3 pages)

4.1 Competitive Analysis

Enterprise Players:

• Palantir AIP: Strong FDE model, less healthcare focus
• Epic/Cerner AI: Incumbent advantage, slower innovation
• Commure: Healthcare AI, FDE approach

Startup Landscape:

• Point solutions (prior auth, clinical documentation)
• Platform plays (healthcare AI infrastructure)
• Services-first vs. product-first models

4.2 Sustainable Differentiation

What Creates Moats:

• Proprietary implementation knowledge (not models)
• Switching costs from EHR integration
• HIPAA compliance barriers
• Long-term customer relationships

What Doesn't:

• Model capability (commoditizing)
• Generic UI/UX (AI replaces interfaces)
• Price competition (race to bottom)

---

Section 5: Strategic Implications (2-3 pages)

5.1 For Healthcare Organizations

Questions to Ask:

• Do we have talent that understands both AI and our clinical workflows?
• Are we building vs. buying vs. partnering?
• What's our 12-month implementation roadmap?

Action Framework:

1. Audit current AI initiatives for implementation risk
2. Evaluate FDE partnership options
3. Prioritize high-ROI automation use cases
4. Build internal capability alongside external support

5.2 For Investors

What to Look For:

• FDE capacity and quality
• Healthcare domain expertise depth
• Customer relationship stickiness
• Product vs. services revenue trajectory

Warning Signs:

• Pure SaaS plays without implementation support
• Generic AI applied to healthcare without domain depth
• No clear path from services to product scale

5.3 For AI Companies

Strategic Choices:

• Build FDE capacity or partner?
• Which healthcare verticals to target?
• How to balance services vs. product revenue?
• When to scale vs. when to focus?

---

Appendix A: Technical Deep Dive on MCP

• Architecture overview
• Healthcare-specific MCP implementations
• Integration patterns and best practices
• Security considerations

Appendix B: FDE Model Implementation Guide

• Hiring and developing FDE talent
• Engagement structures and pricing
• Success metrics and KPIs
• Common failure modes

Appendix C: Healthcare AI Regulatory Landscape

• CMS prior authorization rules
• HIPAA AI implications
• State-by-state variations
• Emerging guidance

Appendix D: Data Sources and Methodology

• MIT AI implementation studies
• Healthcare staffing surveys
• Market sizing methodology
• Interview and research sources

---

Research Required

Primary Research

• Interviews with healthcare CMOs/CNOs on AI adoption barriers
• FDE hiring data from LinkedIn/Indeed
• Customer case study (Premira if closes)
• Developer survey on Claude Code vs. alternatives

Secondary Research

• MIT 2024 AI implementation study (full report)
• Healthcare staffing and burnout statistics
• Prior authorization processing time benchmarks
• Commodities/infrastructure investment correlation to AI buildout

Data Visualization

• AI implementation success rates over time
• Healthcare automation potential by function
• FDE demand growth curve
• Technology stack convergence diagram

---

Timeline

---

Outline generated from Jan 26, 2026 huddle discussion