Handle more volume without more hires
Custom AI Agents Built for Your Business
AI built around how you work.
When no off-the-shelf tool fits, I build an AI agent around your exact process. It reads, decides within rules you set, takes action in your systems and asks a person when it should.
Starting from $3,500 per agent · Final price after your free audit
Multi-step tasks done in minutes
Pay for exactly what you need instead of bloated software
Starting from $3,500 per agent Paid proof of concept from $1,500. Care plan from $299/mo. Final price after your free audit. All prices ›
Full scope
Exactly what you get.
No surprises: everything that's included, what costs extra and what isn't part of it. Your final scope and price come in writing after the free audit.
The project · Starting from $3,500 per agent
- A paid proof of concept: a working agent on a sample of your real data
- The full agent built on Claude, OpenAI or Gemini, whichever fits
- Connections to your systems through their APIs
- Guardrails, approval steps for sensitive actions and a log of every action
- Measurement against your current process before launch
- A walkthrough for the people who review and approve its work
After launch
- Sold per agent; the care plan after launch is optional
- Care plan: monitoring of the agent's actions and error logs
- Tuning of instructions and rules as cases come up
- Updates when the AI models or your systems change
- A monthly report of tasks handled, time saved and errors caught
Extra costs, billed at cost
- AI model usage from Claude, OpenAI or Gemini, based on volume
- Hosting or a server for the agent, if your setup needs one
- API access fees, if any of your software charges for it
- All billed at cost in your business's name, with no markup
What I need from you
- A sample of real data and examples of good and bad results
- The rules: what the agent may do alone and what needs approval
- API access to your systems, limited to what the agent needs
- A person to review flagged cases and approve sensitive steps
Not included
- Legal or compliance review of your process
- Error-free results: AI can make mistakes, so a person stays in the loop
- Medical, legal, financial or emergency decisions made by the agent
- Rebuilding or cleaning up your existing databases
What you get at launch
- The agent running in your own accounts, with keys in your name
- Documentation of its rules, connections and approval steps
- The action log and the measurement against your old process
Example conversation
See it work
It does the legwork. You approve what matters.
- 1
It reads the request
An email, form, document or record comes in. The agent reads it and pulls what it needs from your systems.
- 2
It works within your rules
Routine steps it does on its own, like looking things up, filling in records and drafting replies, inside the limits you set.
- 3
It asks before anything sensitive
Over your limits, or when it isn't sure, it sends the case to a person for approval. Every action goes in a log.
The problem
Why it matters.
Some of your most time-consuming work is too specific for any app: intake that needs judgment, research across several sources, or a process only your business does. That's usually where the biggest savings are hiding.
What's included
What it does.
- Built on leading models like Claude, GPT and Gemini
- Connected to your systems through their APIs
- Guardrails, approval steps and audit logs
- Runs in your own accounts
- Measured against your current process before launch
- Monitored and improved monthly
How it works
From first call to live.
- 1
Proof of concept
A working agent on a sample of your real data, before the full build.
- 2
Build and integrate
Connected to your systems with approvals where needed.
- 3
Launch and measure
Tracked against time saved and error rates.
Who does what
The agent does the legwork. People keep the final say.
- Reading emails, forms and documents as they arrive
- Looking up and cross-checking details across your systems
- Filling in records and drafting replies or summaries
- Sending approval requests to the right person
- Logging every action it takes
- Setting the rules: what it may do alone and what needs approval
- Reviewing flagged cases and approving sensitive steps
- Any medical, legal, financial or emergency decision
Example
A week of intake work, before and after.
- About 60 requests a week read and entered one by one
- Details checked across two or three systems for each one
- Around 10 minutes per request, about 10 hours a week
- Backlogs on busy days and the occasional missed detail
- Requests read and entered as they arrive
- Cross-checks done by the agent and shown with each case
- About 2 hours a week reviewing flagged cases and approvals
- Every action logged and measured against the old process
Illustrative example for an office handling about 60 intake requests a week, not a client result. Your numbers depend on your business, and the free audit estimates them.
Who it's for
Best fit.
- Mid-size companies
- Professional services
- Logistics
- Insurance
- Legal
- Healthcare administration
Works with your tools
Nothing to move.
I build around what you already use. Common examples:
- Claude
- OpenAI
- Google Gemini
- Your APIs and databases
- Microsoft 365
- Google Workspace
Pairs well with
Related services.
AI Consulting and Strategy
Find the revenue leaks you can't see from inside.
Starting from $750 for a BlueprintLearn more ›Workflow Automation and Integrations
Faster response at every step customers see.
Starting from $750 per workflowLearn more ›AI Knowledge Base and SOP Assistant
Consistent service no matter who's working.
Starting from $199/moLearn more ›Custom AI Agents questions.
What's the difference between an AI agent and an automation?
An automation follows fixed steps. An agent can read, interpret and decide within limits you set, which makes it useful for work that needs judgment.
Is it safe to let AI take actions?
Agents only get the access they need, with approval steps for anything sensitive and a log of every action.
How long does it take to build?
A proof of concept usually takes one to three weeks, depending on the systems involved.
What is a custom AI agent?
Software built on a model like Claude, GPT or Gemini that works through a multi-step task in your business: reading, looking things up, deciding within your rules and taking action in your systems. It's built around your process instead of a generic app.
How much does a custom AI agent cost?
From $3,500 per agent, with a paid proof of concept from $1,500 and an optional care plan from $299/mo. AI usage, hosting and any API fees are billed at cost in your business's name. The final price is in writing after the free audit.
What is a proof of concept?
A working agent on a sample of your real data, built before the full project. You see how it handles your actual cases, and how it compares to your current process, before you commit to the full build.
Can the agent make mistakes?
Yes. AI can get things wrong, so a person stays in the loop: anything outside your rules or that it isn't sure about goes to review, and sensitive actions need approval.
Where does my data go?
The agent runs in your own accounts, with API keys in your name, and only gets access to the systems it needs. You can remove that access at any time.
When do I need an agent instead of a regular automation?
When the work needs judgment or spans several systems in a way no app covers. If an existing tool or a simple automation can do the job, I'll tell you in the audit.
What happens after launch?
The care plan is optional. With it I monitor the agent's actions and errors, tune its rules as new cases come up, update it when models or your systems change and send a monthly report.
Free AI audit
Find out where you're losing customers.
In 30 minutes, we'll find your biggest leaks. You leave with a short written plan: what to fix first, what it costs and what it should pay back. No obligation.