How I AI: 3 AI Workflows to Save $140K & Automate Marketing with ElevenLabs' Luke Harries

Welcome back to How I AI! I'm Claire Vo, and my goal is to help you build better and faster with AI. This week, I had a great conversation with Luke Harries, the Head of Growth at ElevenLabs. Luke uses what he calls "vibe marketing," where AI is the engine for growth, automation, and efficiency. He has this great philosophy: everything is a launch. It's what pushes him to systematize and automate every part of his marketing, which keeps a steady flow of high-quality work going out the door.

Luke is a great example of a modern growth leader who isn't just relying on off-the-shelf SaaS tools. He's actually building his own solutions with tools like Cursor and custom AI models. This approach gets him cheaper, faster, and higher-quality results than expensive alternatives. And the proof is in the numbers—he's saved his company over $140,000 in software and agency costs, all while making his team's output faster and better.

Luke shared three of his most useful AI workflows with me. First, we looked at his system for creating polished case studies and social media content in just a few minutes. Then, he explained how he—a marketer—coded a custom translation service to replace an expensive and inefficient vendor. Finally, he showed me what the future of personal AI assistants might look like by connecting his WhatsApp messages to Claude with a Model Context Protocol (MCP). These aren't just cool ideas; they're practical systems you can learn from and use yourself.

Workflow 1: Instant Case Studies with Granola and a Custom GPT

Customer stories are one of the best marketing assets a company can have. But the whole process of interviewing, transcribing, writing, and sharing them is usually slow and manual, which can create a real bottleneck. Luke has found a way around this with a workflow that takes him from a customer chat to a published case study and a tweet thread in less than 15 minutes.

Step 1: The AI-Powered Customer Interview

The process starts with a simple chat with a customer. In our demo, Luke interviewed me about how I use ElevenLabs for prototyping product keynotes at LaunchDarkly. As we talked, he had Granola, an AI note-taking and transcription tool, running in the background.

Key Insight: Granola does more than just transcribe. It intelligently creates a structured summary, pulls out key facts, and even enriches the data with context like the speaker's job title from their email signature.

Step 2: The Custom Copy Editor GPT

Next comes the really clever part. Luke created a custom GPT in ChatGPT to be his dedicated "ElevenLabs Copy Editor." It’s not a simple one-line prompt, but a detailed set of instructions that captures the company's entire tone of voice and content strategy.

He gives the GPT two things:

1. The concise summary generated by Granola.
2. The full, raw transcript to ensure the AI can pull direct, authentic quotes.

Here’s a look at the structure of his prompt:

You are an expert editor and writing assistant specializing in the ElevenLabs communication style. You must enforce American English spelling. Your tone is serious and research-led, similar to Palantir or SpaceX.

The prompt also includes:

• Specific instructions for different content types (e.g., blog posts, tweets).
• Formatting rules, like making headers skimmable summaries of the content below them.
• Positive examples of past blog posts and tweets that perfectly capture the desired style.

Luke’s pro tip is to always edit the underlying prompt rather than the final output. If the AI keeps making a certain mistake or misses a nuance, you can refine the instructions in the custom GPT to fix it for good.

Step 3: Generating the Case Study and Social Content

Once the prompt and context are loaded, the GPT creates a well-structured, on-brand case study in seconds. During our session, it produced a complete blog post titled "How LaunchDarkly Uses ElevenLabs Studio to Prototype Product Keynotes," with clear headers and direct quotes.

And he doesn't stop there. Following his "everything is a launch" philosophy, Luke immediately has the GPT repurpose the content. With a simple prompt like, write a tweet thread for this, the AI generates a whole thread, complete with placeholders for visuals.

[Image: Screenshot of the ElevenLabs Studio interface]

This system turns what was once a time-consuming task into a repeatable, scalable way to create content. Luke even connected this flow to their CRM (Salesforce) using Zapier. Now he can automatically send an email with a Calendly link to new customers, basically putting the entire case study pipeline on autopilot.

Workflow 2: Replacing a $140K SaaS Tool with a Custom Translation Engine

This next workflow really shows what's possible with AI-assisted coding and a builder's mindset. Luke had a common problem: he needed to localize the ElevenLabs website for a global audience. The usual solution was a localization SaaS tool for $40,000 a year, plus translation agencies that cost over $100,000. The whole thing was slow, expensive, and the quality just wasn't very good.

The agency's human translations were so poor that his team kept using ChatGPT to check and correct them. This sparked a realization: if the AI was the source of truth for quality, why not use it for the entire process?

Step 1: Building the Solution in Cursor

Over a weekend when he was supposed to be on vacation, Luke decided to solve the problem himself. He used Cursor, an AI-native code editor, to build the core of a new translation system. This is a big deal for non-engineers. It means marketers can build the first 90% of a solution themselves, then just work with engineering to polish and deploy it.

I think human-in-the-loop SaaS, where your job is about putting low-skilled workers in some sort of flow, which translation is, I think that's very risky.

Step 2: The Automated Workflow

The system he built is simple and effective:

1. Trigger: Whenever a developer changes a text string in the website's codebase on GitHub, a GitHub Action is triggered.
2. AI Translation: The action sends the string to an LLM via a small, custom server.
3. Contextual Prompts: This is key—each language has its own dedicated prompt file. This file contains brand guidelines, a glossary of terms, and stylistic nuances for that specific language (like German, for example).
4. Commit: The translated string is automatically sent back to the codebase as a commit.

For content stored in their CMS, they simply added a "Translate" button that hooks into the same service. This replaced a multi-vendor, multi-day process with an instant, high-quality, and virtually free alternative. The maintenance cost? Zero.

This story says a lot about the future of the SaaS industry. As it gets cheaper and easier to build custom AI solutions, more companies will choose to build instead of buy, especially when they want more quality and control than off-the-shelf products provide. And good news for developers—Luke mentioned that ElevenLabs plans to open-source this solution!

Workflow 3: Connecting Your Brain to AI with a WhatsApp MCP

Luke's final workflow gives us a peek into the future of personal AI assistants. He believes that for an AI to be really useful, it needs access to three things: your email, your calendar, and your WhatsApp. He decided to tackle the third one himself by building a Model Context Protocol (MCP) for WhatsApp.

An MCP is a standardized way to give AI models like Claude access to external tools and data sources. Think of it as a secure API that lets your AI see and interact with your personal apps.

How the WhatsApp MCP Works

Luke has open-sourced this project on his GitHub. Here’s the breakdown:

1. Local Setup: You run a script locally in your terminal. It uses the What's Meow library to pretend to be WhatsApp Web.
2. Scan & Sync: You scan a QR code, just like you would to log into WhatsApp on your computer. The script then securely downloads all your messages to a local SQLite database on your machine. It does a one-time sync for your message history and then listens for new messages as they come in.
3. Expose as a Tool: A local MCP server runs, exposing functions to the AI like getRecentMessages or sendMessage.
4. Connect to Claude: You connect your local MCP server to an AI model like Claude. Now, you can simply talk to Claude in natural language and have it query your WhatsApp data or even send messages on your behalf.

Use Cases and Possibilities

This creates a lot of new possibilities. For instance, Luke showed how he could ask Claude:

What are some recent messages I've received on WhatsApp?

He can also use it to stay on top of industry trends discussed in busy WhatsApp groups. For example, he could ask it to summarize the thoughts on ElevenLabs from my messages over the last week. He could then take that summary, feed it into his custom GPT, and instantly generate a tweet thread based on real-time community sentiment.

You can even chain MCPs together. Luke showed how you could use the WhatsApp MCP to get a summary, send that text to an ElevenLabs MCP to generate a voice note, and then send that voice note back through the WhatsApp MCP. This is how you start to build autonomous agents that can handle complex tasks across different tools, all guided by simple conversation.

Final Thoughts: Everything is a Launch

Luke's workflows are great examples of a new way of working. By adopting an AI-native, builder-first approach, he's saved his company a lot of money and built systems that are faster, more scalable, and produce better work. His philosophy of treating everything as a launch makes sure that every single thing he works on—from a case study to a new feature—gets the attention and distribution it needs to succeed.

These examples show that some of the best uses of AI come from connecting different tools and models to create something new. Whether it’s combining Granola with a custom GPT or linking WhatsApp to Claude with a custom MCP, anyone has the opportunity to build their own custom tools for getting things done. I hope these workflows give you some ideas for what you can automate and systemize in your own work. What will you launch next?

Thanks to Our Sponsors

Thank you to our amazing sponsors for supporting the podcast:

• Orkes—The enterprise platform for reliable applications and agentic workflows
• Retool—AI that's designed for developers, and built for the enterprise

Episode Links

• Watch on YouTube
• Listen on Spotify
• Listen on Apple Podcasts