Org scale
20 to 40 direct reports and three engineering managers across the US, Croatia, India and Chile. I own a $10M+ client portfolio with NPS above +70.
Michael Eakins · Engineering Leader
I lead engineering orgs that ship AI into production.
My organization at Bullhorn reached 40+ engineers across the US, Croatia, India and Chile, with three engineering managers reporting to me. I ship RAG pipelines on AWS Bedrock and Claude. I also built Pipemason, a multi‑agent development pipeline that is live and billing.
- engineers in the organization I lead, across four countries
- 40+
- engineering managers reporting to me
- 3
- daily AI tool adoption across my org, behind evaluation gates
- 76%
- daily users at 99.9% uptime on the GM platform I led
- 200M+
01 Leadership
How I run an engineering org
I stay hands-on in architecture and code, and I run the org on outcomes I can measure.
A delivery system that measures
I moved the org to SAFe and started mentorship programs. Velocity went up 40% and technical debt went down 35%.
Quality and compliance gates
Snyk, SonarQube and automated test gates in every CI/CD pipeline, under HIPAA and SOX compliance.
02 AI in production
RAG, LLMs and agents in regulated software
I have shipped AI into software that has to pass HIPAA and SOX obligations. These are production systems, not demos.
RAG on AWS Bedrock and Claude
I shipped AWS Bedrock and Anthropic Claude retrieval pipelines into client applications at Bullhorn, a HIPAA and SOX regulated SaaS. Manual processing time dropped 60%.
LLM tooling inside the SDLC
I built LLM tooling that generates, modifies and reviews code, with prompt templates, evaluation harnesses, regression tests and human sign-off before anything merges.
Adoption across a whole org
My organization uses Claude Code, GitHub Copilot and Cursor daily at 76% adoption. We got there with evaluation gates, not mandates.
Agents and inference in production
MCP-connected agent workflows, tool calling and serverless inference behind AI features. Content delivery time dropped 70%.
- AWS Bedrock
- Anthropic Claude
- Claude Agent SDK
- Cursor SDK
- MCP
- RAG
- Semantic search and ranking
- Evaluation harnesses
- Prompt engineering
- Serverless inference
When agents write a real share of the code, review has to change. You stop reading the diff and start checking the evidence. That means tests the agent had to pass, contracts frozen before parallel work, capped retries, and a human approving the merge. The engineer who ships it still owns it.
03 Built and run by me
Pipemason
Pipemason takes a Jira ticket to a tested, PR-ready branch. A monitor agent walks 14 enforced phases and hands the work to specialist agents.
90 specialist agents Launched with 22. Grown to 90 as new domains shipped.
The phases cover analysis, frozen contracts, architecture, red and green TDD, implementation, end-to-end tests, a screenshot diff against the design, and security gates. It is the delivery system I would want for any team. Contracts come before code. Evidence comes before the merge. Failures escalate instead of looping in silence.
- Code stays local Customer source code never leaves the customer machine. The runner executes locally.
- Bring your provider Runs on the Claude Agent SDK or the Cursor SDK.
- Gated deploys 300+ automated tests gate every deploy.
- Live product Stripe billing, Sentry monitoring, and iOS and Android companion apps.
04 The record
Where I have done it
-
2022 to present
Bullhorn
I lead enterprise mobile, web and AI platforms for staffing and healthcare. I architected the Go HWS React Native app, which drove a 2400% increase in booking conversions for Fortune 500 clients.
-
2018 to 2022
General Motors
I led 12 direct reports on the connected‑vehicle platform behind myChevrolet, myGMC, myBuick and myCadillac. 200M+ daily users at 99.9% uptime, with deployment time down 75%.
-
2014 to 2018
CareSource
HIPAA platforms for 2M+ Medicaid and Medicare members. I introduced automated testing and CI/CD. Defects dropped 60% and releases went from monthly to every two weeks.
-
2007 to 2014
Earlier years
Computer vision at GE before deep learning made it cool, a 4M+ line VB6 to C# and .NET modernization, and enterprise .NET for telecom at Cincinnati Bell.
-
1995 to 2001
US Army
Satellite Systems Team Chief. I led a 5‑person team keeping mission‑critical communications running, deployed from the Middle East to the Balkans.
BS Computer Science, Westwood College · Certified SAFe 5 Practitioner and SAFe 5 Leader · Contentful Certified Professional · US Army veteran
05 Background
I started writing code on a Tandy 1000 HX my father brought home from Radio Shack. I served six years in the Army, then went to school for software right as the dot-com bubble burst. People told me software was over. It was exactly the right time, and the lesson stuck. The hype never matches what ships.
I write about AI in production on LinkedIn every week, and I keep a public ledger of dated predictions graded hits and misses alike.
Let's talk
I am open to discussing engineering leadership opportunities, especially where AI is part of how the org builds and what it ships.