LevelUp Day 2026: The GoHighLevel AI Agent and Voice AI Updates That Matter, and How I'd Use Each One

Status check: every feature below comes from HighLevel's LevelUp Day 2026 roundup and the help article it links for that feature. I give each status exactly as HighLevel writes it: Beta, Labs, or no label at all. "No label" means the roundup does not mark it as Labs or Beta, not that I have confirmed it is switched on in every account. HighLevel says the roundup will be updated with more details, so check the linked help article before you plan client work around any of these. I have not run all of them in my own account yet, so the steps follow HighLevel's docs.

I build AI agents and Voice AI for agencies every week, and most LevelUp lists are full of things that look good on stage and change nothing in a real build. This one is different. Here are the 11 updates I would actually plan around, in the order I would test them, with a short note on how I would use each one.

1. Convert n8n workflows into Managed Agents

Status in the source: Beta. The menu path in the roundup is AI Agents > Agent Studio > Import from > n8n (Beta).

You upload or paste an n8n JSON export and Agent Studio builds a Managed Agent with the same logic. The importer maps triggers to inbound webhooks, turns logic nodes into decision rules and turns app nodes into CRM actions or API calls. You review the mapped steps, connect the apps it needs, then choose Save as draft or Create & publish. The help article says it will flag steps it sees as unsupported or unnecessary, and imported agents land in their own folder in Agent Studio.

How I'd use it: I would start with one small n8n flow, like a lead intake or a follow-up chain, and always save it as a draft first. Then I would open every mapped step and test the inbound webhook with a dummy contact before anything goes live. The lesson: the import moves your logic, not your judgement. Any step the importer flags is where the old flow was hiding a workaround.

Help article: Import n8n Workflows into Managed Agents

2. Voice AI MCP connector: tools live on a call

Status in the source: No Labs or Beta label in the roundup. It is for speech-to-speech Voice AI agents, and the help article says availability depends on the supported speech-to-speech model and feature access.

Speech-to-speech agents can now call tools from connected apps or custom MCP servers while the caller is still on the line. Open the agent, go to MCP Servers, click + Add MCP, pick a catalog app or connect a custom server, choose the tools and write the "When to use" instruction for each one. Tool call statuses show up in the test transcripts.

How I'd use it: This is how a receptionist agent stops saying "someone will call you back". I would give it two or three tools only, for example check availability and create a booking, and write a very plain "When to use" line for each. Then I would read the test transcripts to see every tool call before I publish. Fewer tools means fewer wrong calls, and the HighLevel help article also says to select only the tools the agent needs.

Help article: MCP Integration for Speech-to-Speech Voice AI Agents

3. Conversation AI MCP server support

Status in the source: Labs. Step 1 in the roundup is "Enable MCP support for Conversation AI in Labs". The help article says it works with prompt-based bots, up to five MCP servers per bot and a maximum of 100 tools.

Conversation AI bots can use MCP servers mid-chat, next to the Knowledge Base and Workflows. You can connect 40+ catalog apps or custom MCP URLs, pick which tools the bot can use and write when it should use them. Open a prompt-based bot, go to MCP Servers and click Connect MCP Server.

How I'd use it: Same rule as Voice AI: one job per tool. For a chat bot I would connect the one system the business actually checks by hand today, such as stock, order status or job status, so the bot can answer the question that usually waits for a human. I would keep the Knowledge Base for fixed answers and MCP for live data.

Help article: Connect MCP Servers to Conversation AI Bots

4. Outbound AI Agents

Status in the source: Labs. Turn on "Conversations AI - Outbound" in Labs. The help article says native outbound triggers currently start conversations by SMS, and each agent needs an SMS sending number, compliance confirmation and outbound activation.

Conversation AI agents can now start the conversation when a CRM event happens, such as a form submission, a missed call or a deal stage change. The agent writes a personal opening message from the trigger context and follows up on a set schedule, pausing or changing course based on the reply. You build an Auto-Pilot agent under AI Agents > Conversation AI, then set the sending number, compliance and triggers in the Deploy tab. The help article lists nine outbound triggers and says Working Hours hold first messages until the next sending window.

How I'd use it: Missed call text back is the first thing I would move here. Instead of a fixed "sorry we missed you" text, the agent opens with the reason it is reaching out and can book the call. Second, a stage change trigger for quotes that go quiet. I would set Working Hours before anything else, because the help article warns that held messages are not thrown away when more triggers fire, so several can still be delivered.

Help article: How to Set Up Outbound Conversation AI Agents in HighLevel

5. AI Campaigns for database reactivation

Status in the source: No separate label in the roundup, but step 1 is to turn on "Conversations AI - Outbound" in Labs and set up an Auto-Pilot agent with a sending number and compliance confirmation.

You can launch bulk AI Campaigns to old leads, past customers or stalled deals using a Conversation AI agent. Pick a segment or a Smart List, choose the send mode (immediate, scheduled or drip) and the AI handles the first message and the follow-ups. When someone replies, the agent moves into a live conversation to handle objections and book appointments. Start it from Contacts > AI Campaign or select contacts and click Start AI campaign.

How I'd use it: I would always use drip, never immediate, on an old list. Start with a small Smart List, read every reply in the first batch and fix the prompt before the rest go out. The help article makes a point I agree with: a Completed status does not by itself mean the message was delivered, answered or booked, so measure bookings, not sends.

Help article: How to Create and Manage Conversation AI Campaigns

6. Voice AI answers WhatsApp calls

Status in the source: No Labs or Beta label in the roundup.

Voice AI agents can answer incoming WhatsApp calls as the first line or as a fallback. In Settings > WhatsApp, open a number, make sure Allow voice calls is on and pick who answers: Team only, Team then Voice AI, or Voice AI only. With Team then Voice AI, the agent answers after a wait of 10, 15, 20 or 30 seconds, and each number keeps its own setting.

How I'd use it: For most local businesses I would start with Team then Voice AI on a 20 second wait. The team still gets first chance, and nobody hears endless ringing. If the business already gets more WhatsApp calls than phone calls, this is the cheapest way to give them a receptionist on that channel with the agent they already have.

Help article: How to Set Up Voice AI for WhatsApp Calls in HighLevel

7. Clone Voice AI agents to sub-accounts, plus versioning and draft testing

Status in the source: Cloning has no Labs or Beta label in the roundup. Agent Versioning is Labs (step 1 is "Enable Agent Versioning in Labs").

Cloning: agencies can clone a Voice AI agent to many sub-accounts at once and choose to copy linked CRM assets such as calendars, workflows and custom fields. A History tab shows the status of each cloning run. The help article says it is for Agency users, supports up to 10 sub-accounts per cloning run and does not support legacy or flow-based Voice AI agents. After cloning you still set the phone number or channel in each sub-account.

Versioning and draft testing: you edit and test changes in a draft while live callers are not affected, compare prompts and settings between versions, and publish or roll back with one click. The help article says it supports prompt-based Standard and Realtime agents, not flow-based ones, and keeps up to 20 versions per agent.

How I'd use it: This is the pair I am most pleased about. I would build one master agent per niche, test it properly, then clone it with its calendars and workflows to each client. After that, every prompt change goes into a draft first, gets a test call, and only then gets published. If a change goes wrong, roll back instead of rewriting at 9pm while a client waits.

Help articles: Cloning Voice AI Agents to Sub-Accounts and Voice AI Agent Versioning: Draft, Test and Publish

8. Flow-based Conversation AI builder and Build Agents with AI

Status in the source: The flow-based builder has no Labs or Beta label in the roundup, and its help article marks the option to draft a flow with AI as Beta. Build Agents with AI is Labs: enable "Conversation AI: Build with AI" in Settings > Labs.

Flow-based builder: a visual canvas for chat bots. Instead of one big prompt, you route the conversation to specialist agents, each with its own instructions and tools such as web search, booking or custom messages. It supports branching, waits and bot transfers. Create a new bot in AI Agents > Conversation AI and choose Flow Based Builder.

Build Agents with AI: describe the agent you want, or start from one of 9 templates, and Build with AI sets up the outbound triggers, inbound channels, working hours, booking actions, hand-off rules and knowledge base. You then refine it by chat before testing and saving.

How I'd use it: I would use Build with AI to get the first draft in minutes, then move anything with more than two jobs into the flow builder. One agent for questions, one for booking, one for hand-off is far easier to fix than one long prompt trying to do all three. Whatever the AI builds, review every route and tool before you switch it on.

Help articles: How to Build Conversation AI Bots with Flow Builder and How to Build Conversation AI Agents with AI in HighLevel

9. A computer for Managed Agents and Ask AI

Status in the source: No Labs or Beta label in the roundup for either. For Managed Agents, the help article says Computer works in chat and manual runs only, not trigger-based runs. For Ask AI, the help article says it is available in sub-accounts with Ask AI Level Up.

Managed Agents: agents can use their own computer to browse websites, manage files and run terminal scripts inside a conversation. They must ask for approval before they submit forms, make payments, publish content or export workspace data. Turn it on in Agent Studio > Managed Agents > Capabilities and set the "Ask Before" permissions.

Ask AI: each user gets a private cloud machine with a browser, terminal and a Files app. Ask AI can work on websites that have no API, run scripts and handle files while you watch a live preview, and you can take control at any time. Logins can be kept in an encrypted store instead of being typed into chat.

How I'd use it: This is for the jobs that have no integration: pulling a report from a supplier portal, checking a directory listing, downloading invoices. I would keep every "Ask Before" approval switched on at the start and only relax one once I have watched it run the same job correctly several times. Because Managed Agent Computer does not run on triggers, plan it as a hands-on assistant, not a background automation.

Help articles: How to Use Computer for Managed Agents in HighLevel and Ask AI Computer: Overview and Getting Started

10. Project Management over MCP

Status in the source: No Labs or Beta label in the roundup.

HighLevel Project Management is now reachable over MCP, so a compatible AI client can create tasks, manage Lists, post comments and update custom statuses with plain language prompts. It uses a hosted endpoint and OAuth sign-in, so there is no local MCP server or VPN to set up.

How I'd use it: I would connect the AI assistant I already work in and use it for the boring part of delivery: turn a client call summary into tasks, set statuses and post the update as a comment. Each person signs in with their own account through OAuth, so keep client spaces tidy before you connect it.

Help article: Connect AI Clients to Project Management MCP

11. Workflows A/B testing

Status in the source: No Labs or Beta label in the roundup. Add it from Automation > Workflows > Add action > A/B Test.

Split contacts evenly across 2 to 10 paths and measure them against a conversion goal. You set the winning criteria (email activity, appointments, tags or payments), the test duration, and whether new contacts go to the winning path when the test ends.

How I'd use it: Put the AI agent inside the test, not just the email copy. Path A: the AI agent sends the first message. Path B: a plain text then a human call. Measure appointments, not opens. Two paths are plenty for most client lists, because ten paths need a lot of contacts before the result means anything.

Help article: How to Configure A/B Tests in HighLevel Workflows

Where I would start this week

1. Agencies with one Voice AI agent copied by hand to many clients: cloning plus versioning first.
2. Local businesses losing missed calls: Outbound AI Agents on the missed call trigger, with Working Hours set.
3. Anyone with an old list: a small drip AI Campaign, measured on bookings.
4. n8n users: import one small flow as a draft and compare it step by step before you move anything else.
5. Everyone: give each agent fewer tools than you think it needs, and read the transcripts.

For today's other updates, see AI Builder now fills in your workflow steps, the Founder Plan and who should upgrade, TikTok Ads and shareable AI agent chat pages and the CRM update with calculated fields and companies.