Your EHR Decides Which Healthcare Workflow Automation You Can Build
FHIR read scopes, HL7 v2 feeds, X12 transactions and screens with no API each support a different kind of build. Map access first, then pick workflows.
Writing
Blog
Notes on AI automation, CRM integration and API work. Everything here comes out of systems that are actually running, not from a content calendar.
An AI Voice Agent for Healthcare Earns Its Keep on Four Call Types
Which clinic calls a voice agent can finish, which ones it must hand off, and the BAA, latency and EHR write-path detail that decides whether any of it survives a month.
Six Legal Workflow Automation Builds and the Rules That Bite
Intake, conflict checks, engagement letters, matter opening, deadlines and billing, with the API limits, retention rules and ethics constraints that decide whether the build survives.
Where Document Workflow Automation Quietly Loses Your Files
Document workflow automation is easy to demo and hard to keep running. Here are the OCR confidence thresholds, idempotency keys, retention rules and approval gates that decide whether it survives a year.
Custom API Integration Is 20% Connection and 80% Everything After
What a custom API integration actually involves: field ownership, rate limits, idempotency, backfill reconciliation and the failure modes that kill integrations quietly.
VoIP CRM Integration: The Parts That Break After Go-Live
Click-to-dial and call logging are the easy day. Here is what a VoIP CRM integration actually needs: webhook ordering, number normalisation, recording retention and matching rules.
QuickBooks and HubSpot Integration: What the Native App Won't Do
The native QuickBooks and HubSpot integration syncs contacts and pushes deals to draft invoices. Here is where it stops, and what you build when it does.
QuickBooks and CRM Integration Breaks in Four Predictable Places
The QuickBooks Online API is easy to connect and hard to keep connected. DisplayName collisions, SyncToken conflicts, token expiry and throttling, with the fixes.
Low Code Workflow Automation Tools, Ranked by Where They Break
A build-side look at low code workflow automation tools: execution limits, error handling, self-hosting terms and the point where a visual canvas stops paying for itself.
LangGraph vs n8n: The Choice Is About State, Not Features
A build-side comparison of LangGraph vs n8n: checkpointing, retries, human-in-the-loop, cost at volume, and how to run both together without regretting it.
How to Evaluate AI Agent Development Companies Before You Sign
Listicles rank AI agent development companies by marketing budget. Here are the production questions that separate a shipped agent from a demo: tool logging, spend caps, approval gates and rollback.
Hiring an AI Automation Consultant: A Build-Side Reality Check
What an AI automation consultant does, where these projects break in production, how to compare an agency against an independent builder, and when not to hire one at all.
CRM Workflow Automation: How It Is Built and Where It Fails
What CRM workflow automation covers, six examples with real field names, the rate limits and dedupe rules that break it, and how to choose where it runs.
n8n Alternatives, Judged by What Breaks in Production
A build-side comparison of n8n alternatives: Activepieces, Windmill, Temporal, Zapier, Make and Pipedream, with the licence terms, execution limits and failure modes that matter.
AI Automation Services: What You Actually Get, and What Breaks
What AI automation services cover, where they fail in production, and the questions that separate a system still running next year from a demo that stopped on a Tuesday.
What will be covered
- Zapier to self-hosted n8n: when the migration pays for itself, and what it costs to run
- CRM migrations that do not lose data, and the fields where migrations quietly lose it
- Salesforce, HubSpot, GoHighLevel: what each API actually exposes and where it rate limits
- AI agents in production: guardrails, spend caps, logging, and where a human signs off
- What an automation project costs, and why most quotes are wrong before the first call