I Was Wrong About AI
Eighteen years ago I was at University of Toronto getting my Masters in AI. I was working on the logic and reasoning side of things, but I shared a lab with a number of Geoffrey Hinton’s students, including Ilya Sutskever, who co-founded Open AI.
They were working on deep learning. Some of them were focused on NLP (natural language processing), which to me seemed like a thankless job. I couldn’t comprehend how having machines understand English was going to have a real impact, and frankly it seemed an impossible task.
Luckily, they didn’t feel the same way.
For the next 14 years, I felt like I was justified in my position. Sure, we had voice assistants that could do small tasks for us, if we structured our requests properly, but that was about it. The gap between what AI could do and what I thought was actually useful seemed pretty wide.
Then ChatGPT came out November 2022, and I was like: shit, they actually did it.
But I was still skeptical. It felt like a curiosity. You could have a conversation that seemed like you were talking to a person who understood you. But spend any time with it and you’d see the cracks. It wasn’t really thinking or understanding. It was predicting a likely response.
Still, I was amazed. Reading up on how LLMs actually work, the fact that it could respond with something coherent at all is insane. The amount of information they have encoded just to generate a semi-intelligent response is wild.
The Acceleration
Since ChatGPT, I can’t believe how fast they’ve improved.
In 2022, having AI generate code seemed like a novel idea. Now they excel at it. I still think you need a human in the loop — someone giving it clear instructions and validating the output. But the gap is closing fast.
Multi-agent systems are raising the bar even higher. One agent critiques the code, another improves it, they iterate until they both agree it’s good.
This is the part where people start asking: “Does this mean I should just hand my entire enterprise project to AI?”
Probably not yet.
The Enterprise Problem
There’s a big difference between code and production software.
AI can absolutely get you a working prototype. An MVP. Something that looks complete and polished. And for a lot of use cases, that’s fine. Internal tools, demos, simple content sites—AI can get you there fast.
But enterprise software is different. Enterprise means:
- Real users with real expectations and zero tolerance for downtime
- Real data — often sensitive, regulated, and irreplaceable
- Real money flowing through systems that must be correct
- Real compliance requirements that can’t be retrofitted
- Real consequences when things break — not just a bad review, but a business crisis
You’ve probably felt the urge at some point to ship something AI wrote for you without reviewing it. An email to a boss or colleague, maybe a blog post. But what if your entire business relied on it being correct? You’d probably review it, right?
That’s the issue with enterprise software — your company relies on it, but many teams are abdicating responsibility to AI. When everything is AI-driven, everyone assumes AI is handling it. And when something breaks, no one knows why, or how the code actually works.
What We’re Actually Doing
Will AI be singlehandedly building enterprise software at some point? Honestly, check back with me in six months, and we’ll see!
At Iversoft, we’re being responsible. We use AI heavily—to accelerate development, to support implementation, to move faster. Here’s how we think about it:
Human ownership, always
Someone is always responsible for the outcome. AI accelerates the work — it doesn’t own the result.
Validated output
Every AI-generated component goes through the same review, testing, and QA process as hand-written code.
Speed without shortcuts
AI helps us move faster on the right things — boilerplate, scaffolding, documentation — not on decisions that require judgment.
Team knowledge stays in-house
We ensure the team understands the code that ships. No black boxes, no “the AI wrote it” explanations in post-mortems.
That’s how you ride the wave of AI and stay competitive without betting your business on code no one fully understands. That’s the reality of enterprise software. You need people to build it — and that’s what we bring to the table.