An assistant sits on your site, answers from the pages you published, qualifies the visitor and books the call. An agent works behind the scenes: reads your business every morning, screens applicants, follows up on leads. We build both, with a name, a job description and clear limits.
Visitors expect to ask a question and get an answer, at 9 pm on a Sunday, without filling out a form first. A generic chat widget hands them a canned menu or a stranger in a call center. An AI assistant built on your own content answers the actual question, asks two or three of its own, and books the call. The ones we have built replaced old widgets that nobody used.
Agents are the newer idea, and the one that saves the owner time. An assistant waits to be asked. An agent has a job and does it on a schedule: reads yesterday's numbers and writes the morning brief, calls applicants and asks the screening questions, nudges a lead that went quiet. The models are good enough now, and the cost is low enough, that a small company can have both.
Every assistant and agent gets a name, a job description, a list of what it may say and a hard list of what it must never say.
The audit reads your site, your inbox and your CRM and comes back with two lists: the questions people actually ask, and the tasks that eat your mornings. Those decide what the assistant should know and what the first agent should do.
Then a fixed price build. We write the assistant's job description and its never say list with you (no prices it does not have, no promises, no legal or medical advice), connect it to your content and your CRM, and test it with the questions from the audit before it goes live. Weeks, not months.
On the retainer we review the conversations monthly, add the facts it was missing, publish the content those gaps point to, and give the agents their next task. You own the prompts, the accounts and the conversation history.
It only answers from the pages you published, and it is told to say when it does not know. We write a never say list before launch: no prices it was not given, no promised results, no legal, medical or financial advice, no guesses about availability. We test it against real questions, and we review the transcripts monthly.
An assistant waits for a question and answers it. An agent has a job and does it on a schedule or a trigger, without being asked. Drew on a builder's site is an assistant. The morning brief that reads the CRM and analytics at 6 am and sends a summary is an agent. Most businesses benefit from one of each.
Into your CRM, as leads with their own source. Every chat that reaches a name and a phone number or email becomes a lead with the transcript attached, scored like any other. Chats are also tracked as their own event in Analytics so you can see what the assistant produces compared with the contact form.
The one that fits the job. We build mostly with Claude from Anthropic and choose per task on accuracy, cost and speed. The model is a part we can swap. The assistant's job description, your content and the CRM connection are the parts that matter, and they stay the same whichever model is underneath.
Not if it is honest about what it is and short. The call says it is an automated screening, asks your six or eight questions, records the answers and tells the applicant when a person will call. It handles the first pass at any hour so your manager spends time on the people worth calling back.
Send us your site and we will show you what it could answer today and what it should never say.
What a website AI assistant does all day for a Florida service business: page matched openers, tap qualifying, price callbacks, CRM leads and a morning recap.
AI at workAssistants answer, agents act.
AI at workHow we build a morning brief agent that reads leads, bookings, site conversations and site health overnight and sends one short message an owner actually reads.