What it takes to get an internal AI app running

November 2026 · 2 min read

Helping teams deploy AI apps at BambooHR meant working through the same setup repeatedly: repositories, CI/CD, secrets, sign-in, access control, data connections, and monitoring. Teams had a working prototype but needed help getting it onto supported infrastructure.

That became a substantial part of my job. I worked with business teams, data engineering, security, and IT to figure out what each app needed and who would own it after launch.

Write down the deployment steps

Some of the longest delays came from small, undocumented dependencies. A setting was disabled by default. Configuration lived in a separate repository. An infrastructure change hadn’t been applied. Once we found the cause, the fix was often straightforward. Finding it took time.

I put the steps into a launch playbook, including the failures I had encountered. It gives teams an experimentation route for testing an idea and a production route for apps that need durable data, sensitive integrations, or wider use.

I also started deploying a minimal service first. Get it building, passing security checks, and answering a health check before adding databases, authentication, and live data. That makes deployment problems easier to isolate.

Move repeated work into the platform

The playbook helped, but each team still had to do much of the setup. After auditing our existing apps, I designed an internal platform to share identity, governed data access, LLM access, and deployment.

The first version is in staging, with app onboarding underway. The goal is for a new app to need a folder and a configuration file, with the shared services already handled.

I still expect teams to need help with their data, application design, and rollout. They should be able to spend more of that time on the work specific to their app.

Want to know more? Ask about my work, or email michael.northrup.data@gmail.com.