Meet the person behind it.
Ninety seconds, in my own words: what gets installed, how it stays safe, and why one person delivers all of it.
Short film coming soon. In the meantime, the fastest way to see if this fits is a call.
I install three AI agents inside your business in 30 days: one that handles your inboxes, one that makes sure no lead or invoice slips, one that briefs you every morning. Then I stay on as your AI department, building you a new one every quarter. You see the hours back in a report every month.
Before this practice, I spent eleven years as a senior product designer at Uber, Instacart, and Thumbtack, building software for millions of people. That is the bar I hold your agents to.
You've probably opened ChatGPT or Claude, gotten a few interesting results, and gone back to running your business. Reasonable. Chat tools put all the work on you. An installed agent does the opposite.
You have to know what to ask, remember to ask it, and paste in the context every time. That's a second job, and you already have one.
Out of the box, AI knows nothing about your clients, your vendors, or how you work. Generic answers are why it never stuck.
An installed agent already knows your business and does its job on its own schedule: sorting the inbox, chasing the invoice, writing the brief. You approve the output. Nobody has to become an AI expert.
Every engagement starts the same way. Not a strategy deck. Three working agents, installed inside the tools you already use.
Every inbox you run, triaged by your rules before you open it. Junk moved to a review folder, never deleted. Routine replies drafted and waiting for your yes. You read what matters and approve what goes out.
Every quote, invoice, and lead tracked as an open loop. Anything sitting too long gets a chase drafted in the right tone, ready for your approval. New inquiries get a fast, personal reply while competitors are still checking voicemail.
One short note every morning: what needs you, what is waiting, what is due, what moved overnight. The mental load is worse than the hours. This is the agent that clears it, so you start already knowing where to look.
Every agent runs with you in the loop. Drafts wait for your approval before anything sends. Nothing is ever deleted automatically. The agents do the work; you stay in control of anything that touches a client or a dollar.
A fixed sequence, not an open-ended project. Each week something goes live, and the load drops as it does.
I learn your business. Accounts connected, rules written, your baseline captured: what your hour is worth, where the time goes.
The Inbox Agent goes live across your accounts. The email load drops first.
The Follow-Up Agent goes live. Open loops start closing, chases start drafting.
The Daily Brief goes live, reading from everything installed. Playbooks delivered, and your first numbers: an estimate of the hours coming back.
A full-time admin is a salary, benefits, onboarding, and one more person to manage. The Department gives you the relief without any of that. It starts in the first month, it runs around the clock, and it doesn't take vacation.
AI models change every few weeks. Your business changes too. An agent installed and left alone falls behind, and quietly stops being used. The Department is the fix: I stay on as the person who runs AI across your operation. It runs on a simple rhythm.
We review your numbers and pick the next agent from your roadmap.
A capability you didn't have last quarter: after-hours call capture, client intake, document chasing, whatever is next.
AI moves at breakneck speed. Models and tools change constantly, and I keep every agent updated, tuned, and working as they do.
Not vibes. Counted tasks, estimated hours back, and what that's worth against your own numbers.
A new agent every quarter, everything maintained, a report every month. Most businesses live here.
A new agent every month, for operations with a deep backlog that want to move faster. It carries a premium over quarterly, and availability is limited so delivery stays sharp.
A new agent every quarter, everything maintained, a report every month. Most businesses live here.
A new agent every month, for operations with a deep backlog. Priced at a premium over quarterly, limited availability.
Sized to your operation on a call and stated as a simple monthly number, billed by the quarter. Sized means exactly that: a solo practice with one inbox is a different job than a multi-entity operation with five, and the number reflects the actual scope, transparently, every time.
Want something sooner? Additional agents can be scoped any time as add-on builds. Department members get priority and better pricing. I take a limited number of add-on builds each month, so the queue is real.
One agent gives you hours back. The next one saves more, on top of the first. A quarter at a time, the time back doesn't just add up. It compounds.
For multi-entity operations and firms with a real backlog, I run a paid AI Operations Audit first, credited toward your install. You get a visual map of where your team's time actually goes, week by week, and a prioritized roadmap of what your Department builds first, next, and after that. The audit isn't a separate project. It's how your quarterly roadmap gets written before we install anything.
Ninety seconds, in my own words: what gets installed, how it stays safe, and why one person delivers all of it.
Short film coming soon. In the meantime, the fastest way to see if this fits is a call.
Every agent I install counts its own work: emails triaged, drafts approved, follow-ups sent, loops closed. Once a month you get one page: what the agents did, the estimated hours that came back, and what those hours are worth using your own numbers, captured in week one. Every quarterly conversation starts from that page.
You're giving agents access to your inboxes and your clients. That only works if the guardrails are real, not a promise. Here is how the work stays safe.
Nothing sends and nothing deletes without your yes. Every action that touches a client or a dollar has an approval gate by design, not as an option you switch on.
I connect to the tools you already run, with least-access permissions, and nothing confidential moves anywhere we haven't agreed to. You can see and revoke access at any time.
The models and platforms underneath, Anthropic and the systems you already use, carry their own SOC 2 and enterprise compliance. I build on top of that foundation, and choose tools with their security posture in mind.
Anything involving money, contracts, or client records is reviewed by a person before it goes out. The agents draft and prepare; you decide.
Straight about what this is: I secure the work through approval gates, least-access, and vendor-certified infrastructure. My independent practice is not itself SOC 2 audited, and I will not claim a certification I do not hold. If your operation needs a formal compliance program, we scope that honestly on the call.
Established, owner-led companies, typically $1M+ in revenue, where the owner or a small team is drowning in email, follow-up, and coordination. The kind of business where giving people their time back is worth investing in.
Missed calls and slow bid follow-up are lost jobs. The Follow-Up Agent chases estimates and subs, the Daily Brief keeps every job's open items in front of you, and after-hours lead capture is a popular first expansion.
Client intake, document collection, and status update emails eat billable hours. The Foundation agents clear the inbox load, and expansion agents handle intake and document chasing.
Agencies assign you a technician from a bench. Here, the person on the sales call, the install, and the quarterly review is the same person. I learn your business once and it stays learned.
I build AI systems inside real businesses. Not slide decks about AI, and not random automations. The actual operating layer: the agents, the playbooks, and the human-in-the-loop guardrails that keep it safe.
I've spent 11 years and counting shipping software products as a senior designer at companies like Uber, Instacart, and Thumbtack, and I work at the frontier of today's AI tools: Claude, Claude Code, and the agents being built on top of them. Every working pattern goes into a documented playbook, so what gets installed in your business is a proven pattern, not an experiment.
Being clear about this protects the quality of the work and your trust in it.
A person reviews and approves anything that carries risk. Nothing auto-sends, nothing auto-deletes. Leverage with control, not automation for its own sake.
I modernize how the work gets done. Your licensed professionals own every regulated decision and review.
I don't build something and disappear. Every install comes with coverage: standalone builds include their first two months of care, and on the Department everything stays covered as part of the relationship. An abandoned system falls behind, and I put my name on my work.
I keep the client list small enough that delivery stays sharp. If capacity is full, I'll say so.
That's the Department's whole job. Models update every few weeks; on the Department, everything installed is maintained, tuned, and fixed as part of the relationship. Standalone builds include their first two months of care, with the option to keep coverage going after. If you're ever between coverage, fixes are available at standard on-demand rates.
Because that's how real progress compounds without waste. Each quarter: review the numbers, install the next agent, keep everything sharp. You're never paying for a maintenance fee you can't see. You're paying for a working relationship with a rhythm.
Your accounts, your data, tools chosen with you. Nothing confidential goes anywhere we haven't agreed to, and vendor privacy policies are part of the tool selection conversation. Anything sensitive gets a human review step by design.
I build primarily on Claude, which I consider the best model family for this work, plus the systems you already run on (your email, your calendar, your CRM). I name my stack because I'd rather you know exactly what you're getting.
No. These are working agents that do defined jobs inside your operation with human approval gates. The chatbot you've tried is why the gap section above exists.
It's a sequence. The Inbox Agent goes live in week two and the email load drops first. The Follow-Up Agent lands in week three, and the Daily Brief caps it in week four, reading from everything installed. Real change inside the first month, by design.
Every price is scoped to your operation, because a solo practice with one inbox and a multi-entity operation with five are different jobs. On the call I'll learn your scope and give you a simple monthly number, billed quarterly. No hourly billing, no surprise invoices, and if we're not a fit, I'll tell you on the call.
Working patterns and frameworks from building AI operating systems inside real firms.
Most owners don't think "I need AI." They think "why does this keep landing on my desk." Here is how to tell if yours is one of them.
The busywork nobody prices out is often the most expensive thing in the building. Here is the math worth doing before you decide it isn't.
The questions that separate someone who ships working systems from someone selling a demo. Bring them to every call, mine included.
If your business has real admin pain and someone inside willing to work with me for 30 days, book a call, tell me where the time is going, and we'll see if it's a fit.