The Rise of AI Voice Agents: Your New Digital Assistant

Full video transcript

Welcome back everyone to the Scale with AI Summit. We’re heading into our next session, live with Chris Lanon, covering AI agents and automations — a lot of insights and actionable takeaways over the next 90-minute workshop, with time for questions and answers as well as an over-the-shoulder demonstration. This Summit is part of a partnership with Howdy and Gamma to make the free event possible, and registrants get access to replays, resources, bonus content, and special deals from sponsors, speakers, and vendor partners.

Chris is a member of the AI Profits Accelerator Mastermind, a certified Chief AI Officer, and the founder and CEO of Profit Lab, which he’s building out as a white-label voice agent platform alongside a course. Chris has been part of the community for years and is instrumental in developing the automations behind MarketSauce.

Chris opened with an introduction to AI agents, followed by a practical demonstration: building a voice agent for a follow-up call tied to a MarketSauce product order — showing the workflow behind delivering a report so that an AI voice agent follows up with the customer by phone.

So what is an AI agent? AI agents are versatile digital assistants that help automate tasks — they make decisions and can integrate software tools, typically with a human in the loop approving steps so the agent doesn’t go off the rails. Several SaaS platforms have been announcing AI agent features: HubSpot has specialized agents for content, social media, prospecting, and customer support; GoHighLevel introduced an “AI Employee” feature (about $97/month per sub-account) offering voice, chat, and SMS conversation, plus agents for marketing funnels, websites, content, review management, and workflow automation; Microsoft has Copilot and Copilot Studio for building agents on the PC; Google Vertex offers both code and no-code agent building. Chris also mentioned working with Darby on a “found money bot” using Replit as an AI-assisted coding environment. Relevance AI is a popular no-code agent platform focused on building “workforces” — teams of agents, often starting with sales and support.

What’s the difference between an automation and an agent? Automations handle simpler, deterministic workflows — the same input reliably produces the same output. Agents handle more complex tasks and can make decisions depending on context; you typically want a human in the loop, though the voice agent built in this session runs fully on autopilot. Chris also touched on newer “system 2” reasoning models (like OpenAI’s o1) that are trained on the reasoning process itself rather than just the end result, letting them alter their approach mid-task — visibly “thinking” before answering, unlike earlier models that needed explicit chain-of-thought prompting.

Chris described three tiers of agent setups: a co-pilot version (an assistant working alongside you, like the Replit example), an autopilot version (fully independent, what was being built in the demo), and agent swarms — teams of specialized agents cooperating toward a goal, similar to departments with different roles. He demoed asking ChatGPT’s o1 preview to design an agent swarm for writing a newsletter, which proposed a topic scout, content curator, research agent, writer, editor, and more, along with a communication protocol and workflow for how they’d hand off work.

Common use cases for agents include customer service, lead qualification, candidate pre-qualification for hiring, email management (sorting, prioritizing, personalized responses), research tasks (e.g., automations built with the Perplexity API), content generation, and complex workflow management. Business benefits include time savings, fewer errors, paying only for active time rather than full shifts, 24/7 availability across time zones, and freeing teams for higher-value relationship work. For getting started, Chris recommended identifying repetitive or distracting tasks, automating one process first to prove ROI before scaling, deciding on a supervision level (supervised vs. autopilot), designing the system and choosing tools, testing, and then scaling what works.

The core demo: MarketSauce customers receive an automatically generated report by email, tied to a document ID stored on their CRM contact record. After delivery, an automated SMS asks them to reply “AI” if they’d like to discuss the report. A “yes” reply tags the contact, which activates the voice agent. The agent calls the customer, confirms interest in discussing the report, looks up the specific report using the stored document ID, discusses it, and offers to book a coaching appointment.

Chris walked through building this in GoHighLevel, the CRM and voice agent platform used for the demo. First, he created a knowledge base named “Market Sauce 2” by scraping Darby’s MarketSauce website. Then he created an assistant named Kaylee/Haley, gave it a role description, and let the platform auto-generate an initial prompt — which pulled company information and included style guardrails (be concise, conversational, proactive) and response guidelines (acknowledge when information isn’t available, focus on MarketSauce’s key benefits). Chris then customized the prompt with explicit tasks: retrieve the user’s MarketSauce report, greet them warmly, introduce the assistant, ask if they want to discuss the report, answer questions, and ask about booking a coaching call.

He reviewed the assistant’s configuration options: whether the contact or the AI initiates the conversation on pickup, pre-call and post-call webhooks (for sending contact data out or saving call transcripts to an external automation), voicemail detection with a custom message, voice selection (11 Labs voices, with Emily noted as a common smooth-sounding choice), language options, ambient background noise settings (coffee shop, call center, etc.), speech normalization for numbers/currency/dates, “back channel” acknowledgments (verbal cues like “yeah” while the customer talks), HIPAA-style data opt-out, call recording, responsiveness, interruption handling, and voice speed/temperature (emotional expressiveness).

The assistant was attached to the knowledge base and to a calendar for booking. Tools were added, including built-in booking tools (check availability, book/reschedule/cancel appointments), CRM task creation, contact notes, self-scheduling follow-ups, email response, and a call-the-user tool. Chris built a custom tool called “Get Market Sauce” with a description (“retrieve the market sauce report for a user”) and a webhook pointing to a Make.com automation. That automation takes the contact ID passed in by the agent, looks up the contact in the CRM, retrieves the stored MarketSauce document ID, pulls the report content from Google Docs, and returns the text back to the agent. He also showed a find-and-replace feature for word substitutions (e.g., swapping “delve” for “dive into”) and for customizing greetings.

To trigger the agent, Chris created a CRM tag (“Kaylee active”) — adding the tag to a contact fires the call via a separate GoHighLevel automation, and the tag is removed after the call completes. He then bought a dedicated phone number (about $2.50/month) for the assistant to use, noting that was the last step required.

Audience questions during setup covered: GoHighLevel’s built-in “AI Employee” voice feature, which at the time only handled inbound calls and bookings, not outbound calls or a custom knowledge base; alternative voice agent platforms like Vapi and a competing product called CloseBot, with Chris noting that sourcing and importing a phone number from Twilio was historically a pain point that this platform’s built-in number purchasing avoided; and typical costs, noting the platform (built partly on Bubble, according to Chris) starts around $225/month for three GoHighLevel sub-accounts up to $1,000/month for unlimited sub-accounts, plus setup complexity that scales with the automation.

Chris then ran the live demo: adding the trigger tag to his own contact record (which had a real MarketSauce document ID) and receiving a call from “Haley.” The first attempt only had general MarketSauce company context because the report-lookup automation wasn’t running yet. Chris re-triggered the call, and on the second attempt the agent successfully pulled his actual report, discussed its psychographic and geographic sections, generated an audience avatar named “Alex Coachman,” role-played as that persona to explore pain points, and — after Chris said he wasn’t feeling well — offered and booked a coaching appointment for the next morning at 8:30 a.m., confirmed on his calendar in real time, before wrapping the call.

Following the demo, Chris showed the call notes summarizing the conversation, the full call transcript (exportable to another automation), and the recording, noting all of this could be piped into further automations — for example, generating action items or texting a summary to a coach.

In Q&A, topics included: pause/latency in the AI’s responses (attributed to lookup latency, which Chris expects to keep improving, and he recommends disclosing to callers that they’re speaking with an AI); recommendations on GoHighLevel for solopreneurs (praised for bundling email marketing, websites, funnels, SMS, and reputation management in one platform, with plans to offer sub-accounts as part of Chris’s upcoming course for about $100/month); the dedicated phone number cost (~$2.50/month per number); and scaling concerns around concurrent calls, which are billed per minute of talk time (about 15 cents/minute) with the ability to add more numbers as needed.

The conversation also touched on GoHighLevel’s newly announced “AI Employee” promo (free usage between November 15 and the end of 2024), the fact that the voice knowledge base is powered by an external platform integrated via API (appearing as an iframe inside GoHighLevel), and a broader discussion about using Gamma for building simple personal-brand and landing pages (Gamma’s Pro plan starts around $20/month), plus ideas raised by attendees for other voice agent applications, such as course-progress follow-up calls to encourage students to keep pace.