AI governance and architecture review
Boards and executives get a clear call on risk, architecture, data boundaries, operating controls, and what should actually move forward.
Start the review
The long view on enterprise AI.
I do advisory and venture studio work grounded in thirty years of building software, with the data to prove what works.
Grand Marais, Minnesota, 2007. Lake Superior rewards the long view.
Current field research
How consultants turn expertise into AI software
I am interviewing independent consultants and fractional leaders about recurring client workflows, failed automation attempts, and what makes AI software client-ready.
Evidence, not only enthusiasm
$100M+
Enterprise value driven across Fortune 500 transformations
30
Years of experience building and operating software
4
Public repositories for agent audits, loop controls, and infrastructure examples
The engagement
Move a serious AI initiative from ambiguity to accountable execution.
I work with leaders who already know AI matters. The problem is deciding what should ship, how it should be governed, and who is accountable when it reaches real workflows.
Map the real workflow, not the demo path.
Put controls, costing, tracing, and escalation into the design.
Ship the smallest governed system that proves the operating model.
Three ways to engage
Pick the path that matches the decision in front of you.
Each path is scoped around a concrete operating decision.
Production AI workflow build
Product and AI operators turn a selected use case into a working, traced, cost-aware workflow that can survive real usage.
Build the workflowVenture and product acceleration
Founders and internal venture teams get operator judgment, buyer framing, and shipping discipline around an AI-native product.
Discuss the ventureHow an engagement runs
01
Decision and risk intake
Name the workflow, buyer or reviewer, data boundaries, current artifacts, and decision you need to make.
02
Architecture and control map
Pressure-test the system shape, evidence trail, governance duties, cost model, and failure handling.
03
Governed delivery plan
Turn the recommendation into a scoped build, pilot, operating cadence, or executive decision package.
Agentic coding research
My research is proof of the operating model.
Governed AI systems need evidence, instrumentation, and shipping discipline. My public repositories record how I test those methods.
~245K
tool events studied
They span shell, edits, search, planning, delegation, MCP, and web traces.
4
public repositories
Audit methods, loop controls, and infrastructure examples share one collection.
MIT
licensed methods and code
Each repository can be inspected, adapted, tested, and cited independently.
Evidence before automation
I look at real execution traces before recommending where agents belong in a workflow.
Failure modes made visible
I turn retries, missing checks, tool churn, and verification gaps into operating controls.
Public methods, private data
I keep the raw work private. The repeatable audit patterns become usable field material.
Open-source research
Audits, loop controls, and infrastructure examples.
This collection is discovered from the GitHub public-portfolio topic and refreshed every six hours. New tagged repositories appear here without a site release.
Validated companion examples for AI agent gateways, observability, and evaluation pipelines.
A tested starter system for students and solo builders using coding agents, BB, Obsidian, and a skill-driven commitment planner.
Tested reference patterns for verifiable Claude Code loops: structured tools, evidence gates, Stop hooks, and prompt caching.
Auditable DuckDB methodology and tooling for analyzing Claude Code session logs and verification behavior.
Proof.
The ventures and projects below run the methods I advise on in production.
Projects
AI Phone Receptionist
Live
AI phone receptionist for busy front desks. Receptn answers missed, busy, and after-hours calls, handles approved questions, captures appointment or callback requests, and sends structured details to the team. Live and accepting paid users.
macOS Menu Bar Manager
Live
Menu bar manager for macOS. Tucked hides menu bar icons behind a thin divider: one click to collapse, hover or hotkeys to reveal, an always-hidden zone for the rest. One-time purchase on the Mac App Store, with no permission prompts, no subscription, and no data collected.
Kept
Private Meeting Intelligence
Active
Private meeting recorder for Mac. Kept records system audio and your microphone without inviting a bot, transcribes and labels speakers on your Mac, and saves the note, transcript, and audio directly to your Obsidian vault. No Kept account or app server.
Claude Model Switcher
Multi-Provider Model Routing
Active
Minimal, auditable bash tool for switching AI providers in Claude Code. One command swaps between Anthropic, DeepSeek, Gemini, KIMI, or local models. Configuration-driven. No hidden network calls. For developers who want control over which model answers.
Bash
JSON Config
OpenRouter
Multi-provider
Linear Claude Bridge
Run your Claude Code context as a Linear agent
Active
Runs your Claude Code setup as an assignable Linear agent. Delegate an issue or message it in an agent session, and a Claude Agent SDK session runs on your machine in the working directory you choose. It carries that directory's CLAUDE.md instructions, MCP servers, and skills, then posts replies in the issue's agent-session thread. The bridge is for talking to an agent that knows your context. Issue-to-PR automation is outside its scope. The tested implementation stays under 1,000 lines and uses no framework. It runs on Claude Code subscription authentication without an Anthropic API key.
TypeScript
Claude Agent SDK
Linear
MCP
Past Projects
Insurance AI
2024-2026
Boring AI for real businesses. Turns AI tasks into governed, auditable business processes. Same outcome every time. No hallucinations where it matters. First vertical: insurance. Submission triage. Claims intelligence. Bordereaux validation.
Successful Exits
Healthcare Analytics
Acquired 2024
Conclusn is a healthcare analytics platform transforming EHR/PM data into actionable business intelligence for medical practices. Specialized in ophthalmology and specialty practices with custom dashboards, predictive insights, and revenue optimization tools.
XBA Group, Inc.
Enterprise Business Intelligence
Acquired 2005
Microsoft Certified Partner specializing in Business Intelligence and Information Worker competencies. Built enterprise data warehouses, OLAP solutions, and integrated Microsoft platform implementations. Delivered training partnerships with Microsoft and achieved significant ROI across multiple verticals through advanced SQL Server Analysis Services and SharePoint solutions.
Advisory Engagements
Digital Workplace Operating System
Completed
Independent consulting engagement for a global, fully remote consumer app company of roughly 150 people. Client name withheld under NDA. The tool stack grew organically and the source of truth eroded with it. Scope covers the current-state assessment, peer benchmark, future-state architecture, 12-month roadmap, and a governance model that works without top-down enforcement.
View Michael Isaac on LinkedInThe operator behind the work
Michael Isaac.
Former Chief Product Officer, SynapseDx (2024-2026)
I have spent three decades across enterprise data, product leadership, venture building, and AI infrastructure, and I have built, bought, and sold software companies along the way.
I started coding at 12 on a Commodore 64, built my first company at 25, led thousand-person teams at Avanade by 40, and exited my latest venture at 55.
Through MPIsaac Ventures I take select advisory engagements and back AI-native ventures. Small by design: every engagement gets me, not a bench.
Away from work I am an instrument-rated pilot, a Scotch enthusiast, and a grandpa to 5 girls. I still wish I held the 1983 world record for Atari Seaquest.
Bring me a real AI decision.
A governance question, a workflow that has to ship, or a venture worth backing: tell me what you’re deciding and when. I’ll reply with a point of view.
Your inquiry goes to a person, not a pipeline. No intake data is sent to analytics.
Email: [email protected]
Location: Excelsior, MN