Since the early years of my career I have built custom tools to solve problems. In 2004 I wrote a script that automated malware cleanup on a Windows XP machine and emailed the technician a report with the scan and cleanup results. The next Windows release broke it, and malware removal as a service has largely gone the way of the dinosaur since. Both outcomes were fine. The script did its job for as long as the job existed.
After that I focused on low-to-no-code ways of building tools and websites, because it made sense: the goal was a working solution, not a programming exercise. I have always created things to solve a problem, make something easier, or free someone from a task so they can do something else.
AI has changed what's possible in that space, though not in the way it's usually sold.
AI is not a magic box
There's a persistent idea that you type a request in plain language and software appears. Anyone who has actually shipped something this way knows the gap between a demo and a product. If you don't understand security and design fundamentals, and you don't know the basics of the language you're working in, you end up with something that fails in production — because nobody can support it, including you.
That's the part the marketing leaves out. AI is a force multiplier for someone who already understands the system. It's a faster way to generate problems for someone who doesn't.
That gap shows up after launch too. Code that was generated quickly still has to be read, tested, and reviewed like anything else — which is why I run the Full Monty, a source-level audit that pairs AI-assisted passes with static analysis and returns a prioritized PDF.
What should I automate, and why?
The usual IT advice is that if you're going to do something more than twice, automate it. That's a good starting rule, but it isn't the whole reason. Inside a business, collaboration between departments matters, and most tools aren't designed to work with each other out of the box. Once you start connecting systems, automation gets genuinely complicated — which is exactly where it pays for itself.
Things that come up most often:
- Chatbots for repetitive inbound questions.
- Spreadsheets that should be interlinked apps.
- Inbound and outbound phone work in customer service and sales.
- Social media campaigns and scheduling.
- Anything repetitive and predictable that quietly consumes hours.
The Pareto principle is the useful lens here: roughly 80% of outcomes come from 20% of causes. Find the 20% of tasks producing most of the drag, and start there.
What the work actually involves
As a consultant, the job starts with understanding the situation: where the friction is, what already exists, and what the people doing the work would change if someone asked them. That takes feedback, time, and patience. The automation is the easy part once that's clear.
Used well, AI lets one person or a small team produce far more than they could otherwise. When I rebuilt my own website I used Relume, Webflow, and Antigravity to produce in two days what would have taken most of a month by hand. Those savings are real. But the important thing to remember is that AI is a system you have to learn, and everyone ends up building a slightly different solution for their own needs.
The short version
Automate the work that repeats. Understand the system before you change it. Build something you can support after the demo. That order matters more than which model you use.