Skip to content
MedPAC Technologies
Client Sign In
All insights AI Build

We build software with AI agents. Humans orchestrate.

How MedPAC actually ships: AI agents do the typing, human staff direct the work, and every change is traceable from prompt to production.

Christian Armstrong, MD, MHSA Chief Executive Officer, MedPAC Technologies 9 min read

This is not a prediction about how software will be built. It is a description of how our team shipped this quarter.

What changed and what did not

What changed: nobody on our team spends the day typing implementations. What did not change: somebody is accountable for every line that reaches a facility running real patients. The accountability did not move to the model. It stayed with a person whose name is on the approval.

Approval gates are the whole safety story

An agent can open a pull request. An agent cannot merge one. That single rule does more for safety than any amount of prompt engineering, and it is the rule we built iBeam around.

Context is an asset, not a chat log

The expensive part of agent-driven work is re-establishing context — the schema, the conventions, the decisions made in March that nobody wrote down. Teams that treat context as disposable pay that cost on every task. That observation is what Bindry exists to fix.

What this means for a regulated business

Auditors do not ask whether a human typed the code. They ask who reviewed it, against what requirement, and whether you can show the trail. Agent-driven development is compatible with that — arguably better at it, since the trail is captured by default instead of reconstructed later.

Questions from this piece

Can AI agents build production healthcare software safely?

Yes, when human approval gates sit before every merge, every change is traceable from the originating prompt to production, and the accumulated context is versioned rather than recreated per session. The risk is not the agent writing code; it is shipping without a human who owns the decision.

What does a human do in an agent-driven development team?

Humans define the problem and constraints, review work at approval gates, decide what ships, and own the outcome. The work shifts from typing implementations to specifying and judging them.

Keep reading