For PE-backed companies
Your people build the AI. Your company keeps it.
An eight-week program where three to five of your people each build three working agents on systems you already own.
Most AI projects leave a company with a policy. This one leaves working systems your own people built and know how to maintain.
Nine weeks, door to door
Two 60-minute sessions a week and about an hour of homework. It starts with an audit, which is included. Your team uses Claude or ChatGPT, connected to the systems you already pay for.
Audit
Week 0
- Map what you haveSystems and data flows mapped, hours baselined, a ranked build slate and a scope of work.
Copilot
Weeks 1 and 2
- FoundationsPolicy translated with IT. One read-only connection each, and a first real question answered.
- ScopesA second system connected, and the first automation running without anyone touching it.
Delegate
Weeks 3 to 5
- Spec and skillsFive agent specs each, three chosen, two skill files written.
- Agent oneBuilt from its spec and running on a schedule.
- Agents two and threeRegistered and running on real work for a full week.
Teammate
Weeks 6 to 8
- Let it runTest sets, a monitor agent, and the cost of every system known.
- Hub and brainA personal hub that shows what ran and what broke, and a second brain that holds their standards.
- MeasureThe register signed with IT, six-minute demos to the sponsor, and the time saved written in their own words.
The four modes follow Allie K. Miller's arc from microtasker to teammate.
What stays when the program ends
Each person
- Three working agents, each with an output that lands with a named person
- A personal hub that opens on their own numbers
- A second brain that holds how they work
- Two skill files and a copilot set up for their job
The company
- 9 to 15 agents its own people can maintain
- A System Register signed by IT
- A standing approval path for the next build
- The cost of each system and a measured time delta
- A 90-day roadmap
Proof
Already running at a multi-site healthcare group
Built by one person on systems the company already owned, using the method the program teaches.
AI agents in production
patient calls handled by voice agents, each one reviewed by a second agent
custom EMR plugins, including a HIPAA-compliant AI scribe
What people get each day without asking
A morning text with every appointment change, refill request and inbound call, then a second one covering every open chat.
A daily report with cash-pay and billing projections, current KPIs and appointment volume by type.
The day's dose sheets, printed and waiting at 7:00 am.
I don't hold your data.
I train your people to build on infrastructure you already own and control.
Built in your tenant
Everything runs in your accounts, under enterprise terms that rule out training on your data.
No logins, no credentials
I never sign in to your systems, and nothing is deployed on mine.
Read-only by default
For all eight weeks, unless IT signs off on something more.
IT approves first
Every agent gets a register entry before it runs. IT signs the register in week 8.
A person on every output
Each agent reports to someone by name. Nothing sends itself.
Under NDA
Signed before the audit starts.
Max Birnberg
I build AI systems that people use every day. At a multi-site healthcare group I built the voice, chat, security and reporting agents above, custom plugins and a patient app on a HIPAA-grade EMR, and a two-way link between the CRM and the EMR.
The hardest build, that CRM and EMR link, failed for a week until I stepped back, planned it properly and prompted in plain language. That is the method I teach.