A relative asked me over the holidays what classes someone should take to get into AI. I’m also introducing myself in an online course this week. This post covers both.
The Novelty Phase
I started using ChatGPT within the first week of its public launch on November 30, 2022. For most of 2023, I was casually aware of it and interacted with it periodically, but mostly for the novelty. I had it write fictitious stories, songs, a poem or two. As far as business value, I may have used it to write a few emails. But writing has been something I’ve enjoyed my entire adult life, so I didn’t lean on it for that purpose.
Then, on December 19, 2023, a physician at the organization where I work demonstrated some of the things he’d accomplished with AI that year.
The presentation immediately got my attention. As an IT leader, I’d fallen behind and needed to catch up quickly.
The Cramming Phase
On January 1, 2024, I began working through online courses. In the three months that followed, I completed all of the following:
- A Crash Course in Data Science (Johns Hopkins University)
- AI Foundations for Everyone Specialization (IBM)
- Amazon Bedrock Getting Started (AWS)
- AWS Cloud Technical Essentials (Amazon Web Services)
- Building AI Powered Chatbots Without Programming (IBM)
- Career Essentials in Cybersecurity by Microsoft and LinkedIn
- Career Essentials in Generative AI by Microsoft and LinkedIn
- Digital Transformation Using AI/ML with Google Cloud Specialization
- Digital Transformation with Google Cloud
- Generative AI Essentials (Coursera)
- Generative AI: Introduction and Applications (IBM)
- Introduction to AI in the Data Center (NVIDIA)
- Introduction to Artificial Intelligence (AI) with Honors (IBM)
- Introduction to Generative AI (Google Cloud)
- Introduction to Large Language Models (Google Cloud)
- Introduction to Responsible AI (Google Cloud)
- Managing Machine Learning Projects with Google Cloud
- Modernize Infrastructure and Applications with Google Cloud
The online courses one can take are virtually limitless. But it was time to pivot toward doing.
Building Things
FreeContactQR — The first website I built using AI was freecontactqr.com . I’d been paying $119.88 per year for a service that I only used to generate the occasional contact QR code. I built the replacement by copying and pasting between ChatGPT and Windows Notepad. At the time, ChatGPT could only handle around 250 lines of code output, so I had to split the site into five modules, each under the limit. I probably spent the better part of three weekends on it—which seemed impressive at the time but now feels excessively long. Later, I consolidated everything into a single page using Claude and haven’t touched it much since. It looks it.
Thanks But Nope — Next, I built a site that uses AI in its core functionality: thanksbutnope.com . The genesis was simple—I was getting bombarded by vendor outreach at work. If I ignored a message, sometimes they’d follow up four or more times, likely using automated tools. Thanks But Nope lets you paste in a vendor message, and it uses Claude Haiku running on Amazon Bedrock to generate a context-sensitive decline response. I still use it every day. I built it between breakfast and lunch on a Saturday. I was still using AI in a browser—Claude this time—but with far less effort than the first project.
parks.tips — The first site I built using a command-line AI coder, specifically Claude Code
, was parks.tips
. I’d been paying $32 per month for a commercial URL shortening service. Claude Code built the replacement in a weekend. URL shortening lets me send out a short link (like parks.tips/timex
) that redirects to a longer URL—useful when sharing links in technical contexts. Claude Code didn’t just write the code. It deployed everything to AWS: Route 53 for DNS, CloudFront for global caching, S3 for static hosting, Lambda for the redirect logic, API Gateway for the admin interface, DynamoDB for storing the links, Cognito for authentication, WAF for security, and CloudWatch for logging. Eleven AWS services, configured and deployed by an AI in a weekend.
Doing Things at the Office
When OpenAI released GPT-5, their demo showed building applications from a single prompt—including dashboards. So I built one. I created a dashboard that consumes data from the API of our IT ticketing system and updates itself every two hours with metrics on ticket completion. I used OpenAI’s Codex CLI to build the front end and Claude Code to deploy it on AWS. It was a one-day project, and my team references it regularly. But there’s no particular ROI to speak of—it’s a convenience, not a transformation.
The transformation came a few weeks ago.
A manager asked if I could find a technical way to transcribe three recorded phone calls into text. The recordings totaled close to half an hour, and she needed to document them. Typing them out by hand would mean playing the audio a few words at a time, consuming more than an hour of her day. I sat down with Claude Code and AWS to see what we could build. By the next morning, I was demoing a working, secure, HIPAA-compliant voice-to-text transcription solution to colleagues. It now saves the company several hours per week and costs $2 to $5 per month to run.
On January 1, 2026, I launched this blog with Claude Code’s assistance. Following a conversation about creating a globally fast static site, Claude recommended Hugo and helped me create a custom theme. It deployed everything to S3, cached globally on CloudFront. Claude continues to help with editing, sometimes generating entire articles, and keeps a GitHub repository up to date for me. Blogging used to be a lot of work. Now I can convert an idea into a published post in around 15 minutes.
These days I use Claude Code for everything—including organizing meeting notes as markdown files in GitHub. It’s helped me stay organized as I’ve had to become my own project manager following the retirement of a long-time colleague from my team. Anthropic has noticed similar non-coding use cases for Claude Code and recently introduced Claude Cowork on January 12, 2026 (currently Mac only) to provide a more streamlined interface for productivity tasks.
The Economics
I get as much value from a $200-per-month Claude Max subscription as I would from unfettered access to a junior software engineer—who, according to recent salary data, averages somewhere between $80,000 and $95,000 per year in the United States. Except this one has wide knowledge across most technical domains and can do research in minutes that would take me hours.
These days, instead of asking what AI can do, I think more in terms of what it can’t.
Software products that used to be expensive to buy—products that took teams years to create—can now be replicated or replaced with a fraction of the effort. There will come a time when even the non-tech-savvy can create custom software for whatever they need, as Shopify CEO Tobi Lütke demonstrated this week. His annual MRI came on a USB stick with Windows-only viewer software. So he prompted Claude to build him an HTML viewer on his Mac. One prompt. It worked. I wrote about why that matters a few days ago.
What we can already do at a consumer level will become reality in the enterprise over time. That doesn’t negate the need for domain expertise in the problems you’re trying to solve. But the need to buy every business application may erode in favor of creating your own.
My Advice
For anyone who feels like they’re just getting started: take some classes. And then start building.
Make AI a daily, hourly, minute-by-minute part of your life. The intuition you develop—knowing when to reach for it, how to prompt it, what it’s good at and what it isn’t—only comes from practice. Courses give you vocabulary and context. Building gives you instinct.
That’s the journey so far. It’s accelerating.