AI agents that do the work and know when to hand it back.
An AI agent doesn't just answer questions; it takes steps. It answers the call, checks the calendar, books the appointment and updates the CRM. I design, build and monitor custom AI agents for owner-led and mid-size businesses, the same way I built my own.
What is an AI agent?
An AI agent is software powered by a large language model that pursues a goal by planning steps, using tools (a calendar, CRM, database or phone line) and acting on the results, rather than only replying to a message. A well-built business agent works within clear permissions, keeps a record of what it did, and hands off to a person when a situation is uncertain or high-stakes.
A digital team member that can take actions, not just chat, and knows when to call you.
Why agents are different from the chatbots you've already tried
Most businesses have met a website chatbot that could only repeat the FAQ. Agents are a different category of software, and the difference is action.
Chatbots answer
They match a question to a scripted reply. Useful for hours and directions; useless the moment the conversation goes off-script.
Automations execute
They run a fixed sequence reliably, but can't handle anything they weren't built to expect.
Agents decide and act
They understand the request, choose which tool to use, take the step and check the result, inside limits you set.
Good agents escalate
The mark of a well-built agent is knowing what it shouldn't handle, and passing that to a person cleanly.
Six parts every agent I build has
The model is only the middle. What makes an agent safe and useful is everything around it.
Goal & instructions
The agent's job description: what it's for and how it should behave
Tools
Calendar, CRM, phone, email, database, the actions it's allowed to take
Knowledge
Your FAQs, policies, price lists and history, cited when used
Reasoning loop
Plan → act → check the result → repeat until the goal is met or it's time to hand off
Guardrails
Permissions, spending and promise limits, and topics it must refuse
Human handoff
Clear rules for when a person takes over, with full context passed along
Logs & evaluation
Every conversation and action recorded, reviewed and scored
The AI agents businesses ask me for most
Each starts from a real workflow in your business, not from a template.
Voice receptionist & intake
Answers every call 24/7, qualifies the caller, books the appointment and routes the urgent ones.
Sales follow-up agent
Follows up every lead on time, answers questions and moves serious buyers to a meeting.
Customer support agent
Resolves the common questions from your own knowledge base and escalates the rest.
Operations agent
Estimates, scheduling, job costing and status updates, the back-office work that never ends.
Research & analysis agent
Gathers data from many sources and turns it into a decision-ready report.
Expert advisor agent
Your expertise, available around the clock to prospects and clients.
Chatbot, automation or AI agent?
Choosing the simplest tool that does the job is cheaper and safer. Sometimes that isn't an agent.
| Chatbot | Automation | AI agent | |
|---|---|---|---|
| How it works | Scripted replies to expected questions | A fixed sequence of steps | Plans and chooses steps toward a goal |
| Handles the unexpected | No | No | Yes, within its guardrails |
| Takes actions | Rarely | Yes, the same ones every time | Yes, chosen to fit the situation |
| Best for | Hours, directions, simple FAQs | Predictable, repetitive tasks | Conversations and multi-step work that varies |
| Needs | Little upkeep | Occasional maintenance | Testing, monitoring and review |
Hire it like a person, test it like software
Write the job description
Tasks, tools, limits and escalation rules, written the way you'd brief a new hire.
Prototype on real examples
Your actual calls, emails and questions, not invented demos.
Evaluate before launch
A test set of real and edge-case conversations, scored for accuracy and tone until it passes.
Launch with guardrails
Limited scope first, human review on sensitive actions and full logging from day one.
Monitor & improve
Weekly review of transcripts and failure cases, with the agent's instructions tightened as we learn.
What my agents never do
These limits are built in, not bolted on.
Discounts, refunds, timelines and guarantees stay inside limits you define.
Legal, medical and financial questions are captured and routed to a qualified person.
Each agent sees only the data and tools its role requires.
I recommend clear disclosure, and California law requires it in certain sales and election contexts.
A good fit, and not a fit
I'd rather tell you now than six months into an engagement.
✓Good fit
- High call or message volume, especially after hours
- Repetitive conversations that still need some judgment
- Businesses serving customers in more than one language
- Teams that will review the agent's work during its first weeks
✕Not a fit
- A single simple FAQ; a chatbot or a better web page may be enough
- Fully autonomous agents making high-stakes decisions
- Projects with no one to review transcripts after launch
- Use cases that depend on the agent pretending to be human
Agents I built and run myself
The AI Advisor
My own advisor agent, trained on 40 years of work, talking with visitors in 20 languages by voice or video avatar.
View →Legal intakeYour Legal Brain
Answers law-firm calls 24/7 in English, Spanish, Farsi and Armenian, qualifies the matter and files the intake.
View →ResearchK2 Parcel Finder
Turns any California address into an investor-grade land feasibility report in under two minutes.
View →Consumer AINapa Wino
An AI sommelier that plans a tasting day across 50+ wineries from a plain-English description.
View →AI Agents: frequently asked questions
What's the difference between an AI agent and a chatbot?
A chatbot replies with scripted answers. An AI agent understands the request, decides what to do, uses tools such as your calendar or CRM to do it, and checks the result. It can complete a task, like booking an appointment, rather than just talking about it.
Are AI agents safe to use with customers?
They can be, when they're built with narrow permissions, clear limits on what they may promise, human handoff rules, full logging and testing on real conversations before launch. Those safeguards are part of every agent I build.
What can an AI agent connect to?
Most business systems with an API or integration: phone lines, email, calendars, CRMs, help desks, databases, practice-management systems such as Clio, Filevine and MyCase, and your own documents.
How much does it cost to build an AI agent?
It depends on the number of tools it connects to and how much testing the use case needs. After discovery you get a fixed build price plus an estimate of monthly running costs (AI model usage, telephony and hosting), so the all-in number is clear before you commit.
Will customers know they're talking to AI?
I recommend telling them, and in some situations it's required. California's bot-disclosure law, for example, covers bots used on large online platforms to sell goods or services or to influence a vote, and similar rules are spreading. Clear disclosure also builds trust; customers mostly care that they get a fast, correct answer.
How long does it take to build an AI agent?
A focused first agent typically moves from job description to supervised launch in weeks. The timeline depends mostly on integrations and on how much real conversation data is available for testing.
Put an agent on the work that never stops
Book a free 30-minute call. We'll identify the one conversation or task where an agent would pay back fastest.