Everyone thinks AI is hype. I thought so too - until I accidentally became a developer again after 20+ years away from code. Can you be a developer if you don't know how to write code? Don't know how to read and debug code? That question would have seemed absurd two years ago. Today, it's the wrong question.
I just published a mobile app to the App Store. Two years ago, I hadn't written a line of code since 2002. In between, I was what I would call a "GTM Executive" - leading pre-sales teams, doing direct sales and partner sales, building customer success teams, driving transformational engagements at large clients. When I was at Tableau, Christian Chabot would say, "you are either building the product or you are selling the product" and I was definitely on the selling side. I was never good at building.
Then I started my own consulting company focused on GenAI and it's been a weird, wild two years.
The Toy Phase
In early 2024, I was like everyone else - playing around with ChatGPT, writing haikus and poems and feeling good about myself. This is where the "hype" narrative gets its fuel. Party tricks and novelty demos.
But then something shifted. The skills started stacking, one on top of the other.
From Prompts to Production
First, I became good at prompt engineering. Then I quickly realized the limitations of ChatGPT - lack of automation, no integrations, no way to operationalize anything. So I became an "Agent Builder." My initial agents were primarily prompts chained together that would leverage web search, web crawl, and documents you fed the LLM.
Suddenly, I was automating tasks that would take humans days or weeks to complete. For a B2B SaaS client, I built an agent that would research target accounts, analyze executive backgrounds, and pull relevant snippets from their case studies that matched the prospect's industry and use case. What used to take a salesperson hours of prep work before each call now happened automatically.
But I still had my training wheels on.
The Technical Threshold
Then I got exposed to the mysterious world of REST APIs.
That sounded technical and scary, so I stayed away from it - until one day I learned how to use an agent builder to make an API call as part of one of its activities. I discovered Postman as a way to test and debug REST API calls. Now I was building more sophisticated agents that didn't just web search and crawl documents, but could integrate with actual systems via REST APIs.
I was recording a lot of YouTube videos and realized I was terrible at writing titles, descriptions, and tags. So I built an agent that would use API calls to automatically generate all of that for me. Suddenly, REST APIs didn't seem scary anymore - they were just a way to get things done.
I was still staying within the safe confines of agent builders that hid complexity from me. It didn't feel technical. It still felt like something any business executive could figure out.
The Swiss Army Knife Problem
Then I started realizing that LLMs were great at everything but they were like the proverbial swiss army knife - good at a lot of stuff but not always the best for everything. I discovered Serverless functions and custom function calling. I could now call Python code from my agents and this unlocked more speed, efficiency, and the ability to do some things in a deterministic way. I had no idea how to write Python code, but I was using AI to write the code and use them in my agents.
But automating workflows with one agent builder platform was limiting. To get salespeople to adopt my solutions, I had to solve the last mile problem. They wanted to live in their tools. I realized that to solve real business workflows, you needed multi-agentic systems.
Now I was building workflows that would span applications - Warmly to Clay to Agent.ai to Clay to Agent.ai to Webflow to Clay to SendSpark to Clay to SalesLoft. For a client's website visitors, I built a system that would identify anonymous visitors, research their company and role, generate personalized landing pages with relevant case studies, create custom video messages, and trigger automated follow-up sequences - all without a human touching it.
This is real revenue flowing through systems. Not hype.
Crossing the Developer Line
Then MCP came out and I had to set up Claude Desktop to connect to an MCP server. That required installing node.js on my Mac, which needed homebrew. Now I felt I was tipping over from being a business executive to someone doing "technical stuff." But it was still lightweight - more a means to an end.
MCP unlocked vibe agent building. I could use Claude and MCP servers to build out my agents faster. I'd tell Claude what the agent did and it would list out each step in gory detail. All I had to do was copy each step, each prompt, and I was building agents at 10x the speed.
But I'm Not a Developer
Vibe coding was becoming popular, but that wasn't for me.
I wasn't a developer. I just built agents.
Then I got the chance to observe an intern use vibe coding to build an app. He wasn't writing any code, but he was building apps that met the business need. Anytime we needed changes, he knew how to vibe code them. He kept building more and more complicated stuff and his vibe coding never hit a wall.
Then a client wanted me to build an app. Now I was using VS Code, Replit, Claude Code - and I was building apps. My job turned out to be the product manager defining what the app did, and the QA tester finding bugs and new features, adding them back through vibe coding. But AI did all the development for me.
Then the client wanted a mobile app and I needed to build a "monorepo" app (yes, I had no idea what that meant either). Now I was not only using VS Code and Claude Code, but I was committing code to GitHub and using Railway to deploy apps. I was using Expo Go to test mobile apps. I got an Apple Developer license because I was going to publish an app to the App Store.
Now, I was feeling emboldened, so I decided that I wanted to see what was involved in installing a private LLM on my own hardware. So I took an old Apple laptop and set up ollama and Phi3 which did not work (hardware was not strong enough), so I installed Gemma. I quickly realized that these small models are really toys and nothing like the powerful models we have access to.
So I wondered what I would need to install larger models. I found a gaming laptop my son wasn't using that had a built in NVDIA card, but the Windows OS was not supported. So I installed Linux on it and then installed a private LLM. Now I need to go find more RAM to beef it up, so I can see if it can actually do anything useful.
I did this as a curiosity project before going to bed Friday night!
The Adage: AI Cannot Do This... YET
When I started this journey, the AI world was nascent. All we had was ChatGPT. It took six months before agents became a thing, then they were everywhere. A year in, AI code generation tools were not that great. Vibe coding was great for demos but invariably went around in circles, leaving you with a mess of code you had no idea how to fix. Now they are amazing. The tools caught up to the promise and we are still in the early innings.
If I look at my growth by leveraging AI, the sky is the limit.
If someone had told me that I would be developing web apps, mobile apps, using Visual Studio, using GitHub, deploying apps to Railway, and publishing apps to the App Store, I would have said they were crazy. If someone had told me that I would be automating complicated human workflows that drive real revenue for real companies, I would have said they were crazy.
But that is where I am now.
The Hype vs. The Reality
The hype said AI would change everything someday.
The reality? It already has.
I'm not a developer. I'm a GTM executive who learned to build because AI made it possible - and because I was curious.
The gap between "AI-curious executive playing with ChatGPT" and "building production systems that drive business outcomes" isn't as wide as you think. But you have to dive in headfirst.
The skills stack faster than you can imagine. The capabilities compound monthly. The inflection points sneak up on you - one day you're chaining prompts together, the next you're deploying apps to production.
If you're still on the sidelines waiting to see if this is real, you're already behind.
Not because of the hype, but because the people diving in are learning, building, and solving real problems while you're watching.
Pick one workflow this week. Automate it. See where it takes you.
The journey from haikus to production apps is shorter than you think.