AI agents on the levelpartner.ai platform answer your leads and clients on chat, SMS, social messages and the phone, using your own information and your own rules, and they write every conversation into the CRM. Set one up properly and the first reply to a new enquiry stops depending on who happens to be at a desk. This article explains what an AI agent is on the platform, what it can and cannot do, and the order in which to build one so it is useful in week one.
What an AI agent actually is here
An AI agent is a configured assistant with three parts: instructions that tell it who it is and what it is for, a knowledge base it answers from, and a set of actions it is allowed to take. The actions are what separate it from a chat toy. It can book into a calendar, update a contact field, move an opportunity, add a tag or hand the conversation to a person.
Each agent sits inside one account and works on the channels you attach to it. A text agent covers web chat, SMS, WhatsApp and social direct messages. A voice agent answers and makes phone calls. Both read and write the same contact record, so a lead who chats on Monday and calls on Wednesday is one person with one history.
Two ways to run an agent
You choose how much freedom the agent has. In Suggestive mode it drafts a reply and a member of your team approves, edits or discards it in Conversations. In Auto-Pilot it replies on its own and only pulls a person in when a handover scenario applies. Off stops it replying at all.
Most firms should start in Suggestive mode for two weeks. Your team sees exactly what the agent would have said, corrects the knowledge base where it guesses, and you move to Auto-Pilot only once the drafts stop needing edits.
Before you start
- Admin access to your account, and the channels you want covered already connected: web chat widget, a phone number for SMS, and any social pages.
- A written list of the twenty questions your team answers most, with the correct answer to each.
- The calendar the agent should book into, already set up with availability and buffers.
- A decision on when a person must take over: complaints, pricing negotiations, anything regulated.
- One named owner who reviews agent conversations every week.
How to set it up
- Go to AI Agents > Conversation AI and click + Create Bot.
- Select Guided Form Setup.
- Click Create New Bot.
- Choose the bot type: General Q&A or Appointment Booking.
- Select or create a Brand Voice.
- Configure Bot Settings.
- Configure Bot Goals, including the calendar and any handover.
- Save the bot.
- Choose Suggestive as the bot's mode for the first two weeks.
What AI agents are for in a 60-person company
If revenue has been flat for a year, the problem is rarely a lack of enquiries. It is usually what happens to enquiries after they arrive: a slow first response, an inconsistent qualification, a meeting that never gets booked. An AI agent fixes the first hour of that journey. It replies in seconds, asks the same qualifying questions every time and offers a meeting while the lead is still interested.
It replaces three things: the generic "thanks, we will be in touch" autoresponder, the shared inbox someone checks twice a day, and the evening and weekend gap where enquiries simply wait. It does not replace your salespeople. It hands them a booked, qualified conversation instead of a cold form submission. For the wider process behind that first reply, see AI follow up.
Suppose the firm receives 120 website enquiries a month, its median first reply takes five working hours and about a third of enquiries never get a meeting offer. That would be roughly 40 enquiries a month left with no next step. An agent that replies within a minute and offers two times from the partners' calendar would put a meeting offer in front of all 120, and the team would only touch conversations that are already qualified.
How much time it takes to run
Expect a day to build the first version properly, most of it spent writing the knowledge base rather than clicking through settings. After launch, the owner should spend about an hour a week reading a sample of conversations, adding missing answers and tightening the instructions. That weekly hour is the difference between an agent that improves and one that slowly embarrasses you.
Most people give the agent a long, friendly personality and a thin knowledge base. The result is a charming assistant that guesses. Write short instructions and a thick knowledge base. Accuracy matters far more than warmth.
When not to use an AI agent
Do not put an agent in front of conversations where a wrong answer has a legal or financial consequence and no person checks it. Mortgage brokers should not let an agent comment on product suitability. Consultants should not let it quote fees that depend on scope. In those cases keep the agent to booking and information gathering, and route everything else to a person. If you need the agent on the phone rather than in chat, start with the AI receptionist guide.
Troubleshooting
The agent invents answers. It is filling gaps in the knowledge base. Find the question it guessed, add the correct answer as a written FAQ entry and tell it in the instructions to offer a hand-off when it does not know.
It never books anything. Appointment Booking is either not enabled in Bot Goals or the calendar has no availability in the window the agent looks at. Check the calendar's availability and minimum notice settings, then test a booking yourself.
Contacts are duplicated. The agent creates a new contact when it cannot match an email or phone number. Ask for one identifier early in the conversation, and check your duplicate contact settings.
Staff do not see hand-offs. The Human Handover action is assigning to a user whose notifications are off. Assign an active user and switch on the notification for conversations assigned to them in that user's staff settings.
Frequently asked questions
Do AI agents work on WhatsApp and SMS as well as web chat?
Yes. A text agent can answer on web chat, SMS, WhatsApp and social direct messages once each channel is connected to your account. Usage on messaging channels is billed as used. Start with one channel and add the others when the answers are reliable.
Can I have more than one agent?
You can run several, each with its own instructions and knowledge. A common split is one agent for new enquiries and one for existing clients. Keep each agent's job narrow, because a single agent asked to do everything does each job badly.
What happens when the agent cannot answer?
It follows the Human Handover action you configured. Usually that means sending a final message, assigning the conversation to a named user and pausing itself for a set time. Your team picks it up in Conversations with the full history. Our guide to AI customer support covers escalation rules in more depth.
How is this different from a chatbot builder?
A chatbot builder follows fixed branches you draw. An agent reads the question, answers from your knowledge base and takes actions in your CRM. For a side-by-side look at the conversation side, see conversational AI.