Insights
Field notes from production healthcare AI.
Long-form pieces on the work itself: HIPAA-grade workflow design, replacing rules engines with reasoning, EHR integration through Marketplace, evaluation frameworks for clinical outputs, agent governance, and the economics of fractional AI/CTO.
- StrategyMay 22, 20267 min read
Hiring a healthcare AI consultant: a CTO's checklist
The 'healthcare AI consultant' market is loud and uneven. Some practitioners can defend a HIPAA control review and ship a system into production; many cannot. Here is the checklist I use when sourcing AI specialists for digital health companies — what to ask, what to demand to see, and the disqualifiers that should end the conversation.
Read
- ComplianceApr 22, 20265 min read
HIPAA-compliant AI automation: what auditors actually look for
Most teams designing healthcare AI start with the model. The right starting point is the audit. Here's what actually shows up on a HIPAA control review for an LLM-powered workflow — and the design choices that make it pass cleanly.
Read
- ArchitectureApr 8, 20264 min read
Replacing a clinical rules engine with LLM reasoning
Rules engines are brittle, slow to extend, and expensive to test. LLMs are flexible, fast to extend, and impossible to test the same way. Here's the architecture pattern that lets you trade the first set of problems for the second — without losing your safety story.
Read
- IntegrationMar 25, 20264 min read
EHR + Claude: an Athena Marketplace field guide
Athenahealth's Marketplace approval process is a serious gate, especially when Claude or another LLM is in the data path. Here's the workflow that gets you through it without burning weeks on rejected submissions.
Read
- ArchitectureMar 12, 20264 min read
Agent orchestration for clinical decision support
An 'agent' is a fashionable word for something that's been working in production code for decades: a controlled loop that calls tools, reasons about state, and decides what to do next. Here's how to apply that pattern to clinical decision support without the failure modes that have killed lesser CDS systems.
Read
- EvaluationFeb 26, 20264 min read
Evaluation frameworks for clinical AI outputs
An evaluation harness for clinical AI is not a test suite. It's a living artifact — golden cases, judges, drift monitoring, calibration — that lets you ship faster without giving up the safety story. Here's how to build one that survives a clinical review.
Read
- ComplianceFeb 12, 20266 min read
SOC 2 and AI governance: mapping LLM controls to Trust Services Criteria
SOC 2's Trust Services Criteria were written before LLMs existed, but the controls map cleanly onto agent architectures if you know how to translate. Here is the mapping — CC1 through CC9, plus the Confidentiality and Privacy criteria — and the artifacts you should have on file before the auditor walks in.
Read
- ToolsJan 29, 20265 min read
MCP servers for healthcare ops: the safe starter set
Model Context Protocol turns Claude and other model clients into something that can read your systems and act on them. In healthcare, that capability is double-edged. Here is the conservative starter set of MCP servers that delivers 80% of the leverage with bounded risk — and the servers you should not turn on yet.
Read
- StrategyJan 15, 20265 min read
Why clinical AI pilots stall — and the checklist that ships them
The same five things kill digital health AI pilots over and over: undefined success criteria, unbounded scope, no evaluation harness, no named governance owner, and a clinician audience that wasn't in the design room. Here is the checklist that catches all five before week one — and the fix for a pilot that has already stalled.
Read
- StrategyJan 2, 20266 min read
From rules to reasoning: the CTO's playbook for pivoting a clinical product to LLMs
Pivoting a clinical product from a rules-based core to an LLM- and agent-based core is a six-month organizational change, not a six-week refactor. Here is the sequenced playbook — substrate first, scenarios second, threshold pivot, sunset on a schedule — and the org changes that make it work.
Read
- StrategyDec 12, 20256 min read
The economics of a fractional AI/CTO in regulated industries
A fractional AI/CTO retainer at $5K to $15K per month sounds expensive until you compare it to the alternative — a full-time hire that takes four months to find, costs $400K all-in, and may not have the specific AI plus compliance combination you actually need. Here is the math, the shape of engagement that justifies the spend, and the situations where a fractional is the wrong call.
Read