Life MCP
A gateway MCP server fronting per-domain children with a unified DuckDB lake. Used in production as a personal agent platform.
MCPagentsDuckDBgateway
Our Work
Engagements you can point to. Code we maintain in the open.
Anonymized case studies with real outcomes, alongside the production MCP and agent work AI-Dynamics runs for itself.
Network operations
AIOps for a Mid-Market Network Team
A national managed-services provider
Problem. A 50-engineer NOC was drowning in repetitive triage on a 5,000-device estate; mean time to acknowledge had climbed past their SLO.
What we built. We deployed a triage agent grounded in live config snapshots and ServiceNow change history, with a closed-loop runbook the team co-authored.
Outcome. MTTA cut from 14 minutes to 4
71%
Reduction in mean time to acknowledge
“They moved from "AI" being a slide on our roadmap to a thing that pages our L2 with a draft of the answer.”
SMB operations
Agentic Operations for a Fitness Franchise
A multi-state fitness franchise
Problem. Each studio location ran its own copy of an email and SMS campaign workflow; corporate had no visibility into who had sent what, and personalization defaulted to broken merge tags.
What we built. We replaced the per-location flows with a single corporate agent stack that templates campaigns from a central brand library and reports delivery and engagement back to corporate marketing.
Outcome. One campaign system across every studio
24
Hours per location per month freed up
“We stopped having to chase locations about which campaign they sent. Now we just look at the dashboard.”
Strategy / advisory
30-Day AI Readiness Assessment for a Services Firm
A regulated-data professional services firm
Problem. An executive team was being pitched AI initiatives by every vendor in their stack and had no way to evaluate them on the merits.
What we built. A 30-day engagement — technical interviews, data-room review, three vendor pitches scored against a rubric we authored with them, and a written memo with a defensible recommendation.
Outcome. A defensible AI decision in one month
$0
Spent on AI tooling that didn't pass the rubric
“For the first time we had a way to say no to AI vendors without sounding like luddites.”
Engineering Showcase
Some of what we build for ourselves.
Production MCP servers and agent stacks AI-Dynamics builds and operates. Useful as evidence of how we think about agent design, evals, and ownership.
Home Assistant MCP Server
Production MCP server exposing Home Assistant state and services to LLM clients, with a tested type-safe schema.
MCPHome Assistantsmart home
Samsung TV MCP
MCP server controlling Samsung smart TVs over the local network, with safety guards around volume and input switching.
MCPIoTSamsung
Outlook MCP
MCP server reading Microsoft Outlook accounts and folders on macOS through the legacy Outlook bridge.
MCPOutlookmacOS