Current Projects
Current, running, in revenue motion.
Two agents, not ten. I built the larger multi-agent version first, then deliberately settled on two — one for the for-profit portfolio, one for the nonprofit work — each isolated by credentials, workspace, mailbox and scope. Neither reaches into the other's domain. Fewer agents is the point: more of them meant more drift between them, and two keeps one synchronized working environment instead of several that quietly diverge.
Each agent is designed to be a PM — a Project Manager, not an assistant waiting for instructions. It runs the list, keeps me on task, and holds the detail that technical work demands. It is a second brain with the technical skills sitting alongside it. Since my brain injury, that is not a convenience — it is the part that makes the work possible at all.
What Makes It Unusual
The mechanics, not the marketing.
- Skills-as-procedure, not docs-as-prose. Startup, session close, ethics check, credential handling, archive trimming — all event-triggered skills the agents invoke themselves.
- A 6-point ethical framework (Judeo-Christian) wired into a pre-decision skill that gates persuasion, finance, and content work before it ships.
- Pre-commit hook that scans PDFs via pdftotext, blocking credential leaks even from generated session reports.
- Browser isolation by config — fresh Firefox profile per launch, zero persistent state. Closed an OAuth-prefill leak in one pass.
- Cross-profile shared memory via symlinks, so the PM agent keeps continuity across two separate Anthropic accounts.
- A solution-shape heuristic the agents apply before proposing anything: skill? extension? API-vs-GUI?
- Real revenue motion: leads flow from LinkedIn Sales Navigator into GoHighLevel pipelines via agent-built workflows; SMS reminders fire via cron + GHL API; the booking calendar conflict-checks Google Calendar.
How I Coach Agents
A team, not a tool. I call each agent a PM — a Project Manager — and I manage them the way I'd manage people.
Verbatim line for revenue drift — agents repeat it when conversations wander off income work.
Verbatim line for security stops. Codified after a single incident; not repeated since.
- Direct human direction in-session overrides any standing instruction. The PM updates the standing rule after the fact, so the correction sticks instead of being re-made next week.
- The PM is expected to push back. If I drift off revenue work, or reach for a fix whose blast radius is too wide, saying so is part of the job — not a breach of it.
- No fabrication about me. If the PM doesn't know my history, ask — don't invent.
Background
The short version.
10+ years thinking about AI before the tools were widely available. I used to fear the rise of AI — before I needed it. I suffered a Traumatic Brain Injury a couple of years ago that took away so much of my tech and sales ability, specifically in the "attention to detail" department. The agent environment is partly memory prosthesis, partly force multiplier, partly an experiment in whether AI can carry real operational weight for a small operator running a portfolio.
Working with me
Open to short contracts.
I'm not a coach and I don't sell a program. I ran an independent IT consulting business for nineteen years, supporting over a hundred small businesses single-handedly. What I do now isn't the same work — it's much more of the work I actually want to do, with AI-enhanced skills and assistance behind it. I set up AI systems that keep working after I log off.
The problem most owners run into isn't that AI can't do the work. It's that every session starts over. You explain your business again. You re-establish the rules again. You re-teach it what it already learned last week. Another tutorial doesn't fix that, and you're not big enough to hire someone full-time to sit on it.
That's the gap I build across: persistent context, documented procedure, real guardrails, and recovery when something breaks — so the system remembers your business instead of meeting it fresh every morning.
$150/hour. Short contracts, worked live over Zoom. You keep your own AI account; I build what sits inside it.
What I sell is the method, not a shelf of finished products.
My own book needed a reader app that didn't exist — illustrated pages with audiobook narration synced to them, page by page. So I had one built, using skills I did not possess, in weeks of part-time effort instead of the months and the budget that road normally costs. It's live on Google Play, two more apps are behind it, and I hired no developers. That same build is now being turned into something other publishers can use for their own titles.
If you've ever wondered how someone like you gets their own app built, that's exactly the kind of question I take the mystery out of.
We build it together, in your environment, so you understand what it does and why. When we're done you own it — including the parts that let you extend it yourself.
Book a time to talk, or email cn@christiannaef.biz.