← cd ../insights

· ai · 4 min read · by Steven Haws

Sitting with AI

A gas streetlamp with a lamplighter's ladder beside a modern electric streetlight, joined by glowing wiring under the street

AI should make your thinking stronger, not do it for you. Here’s what changed my mind, and what didn’t.

This one is a change of pace. It’s less of a how-to and more of a project retrospective and a few personal lessons learned. I started this year as a self-proclaimed AI skeptic, and in many ways I still am. I believe most people use AI as a brain replacement instead of an effort augmentation tool, and I think society will feel that for years to come. But I’ve spent most of this year on a technical project that changed how I think about what AI is here to do, and what it does best. Let me start with the backstory.

The Hype I Waited Out

November 2022 was a landmark month in tech. OpenAI released ChatGPT, the first large language model (LLM) that an everyday person could talk to and actually get value from. It’s a name that may forever be tied to LLMs the way Kleenex is tied to tissue.

With a decent amount of tech experience under my belt, I waited for the hype to die down. As the months rolled on, it looked more and more like the machine learning and automation I’d been seeing in SaaS products for years. So I decided it wasn’t worth my time.

Around the start of 2025, GitHub Copilot became a free add-on to VS Code, pitched as a side-by-side coding companion. I used to code, but not often enough to remember syntax and libraries quickly, and I hadn’t kept up with the newest languages. So I gave it a whirl. It was a fun diversion for a couple of weekends. But it wasn’t good at writing complete code, and I spent more time figuring out what it had broken than it saved me. Worse, babysitting it took the fun out of building. So once again, I set it aside.

Through 2025 I worked with several companies on AI strategy and rollouts. Most of them felt like appeasing executives who knew there was a shiny new tool they weren’t using yet, but didn’t really understand why they needed it.

When It Finally Clicked

In early 2026 I started working with a company that was determined to find the real use case for AI, and was willing to go through the growing pains of changing its processes and structure to get there. IT leadership pushed everyone to experiment and gave them the resources to try many kinds of agents and formats, even side by side. That push lined up with some major leaps in what LLMs could do.

Slowly, a pattern emerged. The first problem to solve was governance of data and code. Once standards were set and dedicated agents were built to act as guardrails, the acceleration was impressive. By our internal metrics, the speed at which solutions could be designed, built, tested and deployed jumped 10 to 20 times over what was possible before.

That changed the conversation. With development resources no longer the bottleneck, product management stopped debating who got development time and started asking what we should solve with the AI-powered machine that IT had become. The sky was the limit, and teams across the business were eager to tackle problems that had been holding them back or weighing them down administratively.

Lamplighters and Electricians

In one particularly inspiring IT meeting, a member of the leadership team walked through the history of lamplighters in the early 1900s, as streetlights moved from gas to electric. Lamplighters gave way to electricians. The work didn’t disappear, it changed. That’s when it dawned on me what kind of transition we’re in. If I were a better student of history, I would have recognized this pivot point sooner.

Each major shift in how we interact with technology has changed the skills we need, and has also increased how much meaningful work we can get done:

  • 1940s–50s, Programmable Computing: Rewiring hardware gave way to writing code.
  • 1970s–80s, the GUI and the PC: Command lines gave way to the mouse, windows and icons, and everyday people could use computers.
  • 1990s, the Commercial Internet: Standalone computers became portals for communication, commerce and information.
  • Late 2000s, Mobile: The iPhone made technology something we carry everywhere instead of a desk we sit at.
  • 2010s, Cloud and IoT: Data followed us across devices, and technology started responding to our voices and habits.
  • 2020s, Generative AI: Rigid commands gave way to plain-language conversation with a tool that can reason and create.

What I Got Right, and What I Got Wrong

My original concern still stands. Too many people use AI to replace their thinking instead of augmenting it. But I was wrong about what the future should look like.

As business leaders, we have a chance to step into a future where the skills that matter are fundamentally different. I remember when the command line was the way to get things done on a computer, and later when Googling became the best way to find information. Now we get to rewrite our habits again, this time to elevate our thinking.

Humans aren’t going anywhere. If we learn to put AI to work on the lower-level technical tasks, we free ourselves to focus on creating and strategizing. That doesn’t mean everyone should forget the command line, binary or compilers. Those specialists will always be needed, and if that’s you or your people, there will be a place for that skill. That market will just get more competitive as the work becomes more niche.

The lamplighters didn’t lose the light. They learned to wire it.

If any of this sounds like what you’re seeing in your business, let’s schedule a time to chat about whether a shift in expectations and strategy makes sense for your team. It might just change the way you do business.