ChatGPT Six Weeks In – What I Think Now

ChatGPT hit 1 million users in five days. If you were building software in late 2022, you watched the internet lose its mind. The hype was deafening—existential risk, the end of white-collar jobs, superintelligence arriving next Tuesday. Six weeks in, now at the start of 2023, I'm seeing something different. The frenzy is wearing off, and the real picture is starting to emerge.

Here's what I actually think: ChatGPT is a genuinely significant tool. But what it is and what it's being claimed to be are two completely different things.

Let me be specific about what I've seen work in these six weeks. Customer service teams are using it to draft responses—not replace humans, but cut the time spent on repetitive tickets by half. Content teams are using it as a starting point for blog outlines and email drafts. Engineering teams are using it to explain what code does, or to help think through algorithm problems. These aren't revolutionary. They're incremental. They matter.

What I've also seen: a lot of wishful thinking. Companies are asking if ChatGPT can replace their data analysts, their junior engineers, their copywriters. And the honest answer is no—not yet, maybe not ever. When you actually try to use ChatGPT for these things, you hit real walls. The outputs look good until you inspect them. They're hallucinating. They're making up citations. They're getting domain specifics wrong. They need serious human review.

The Capability Floor Has Risen

What changed isn't that AI suddenly got perfect. What changed is that the floor got higher. If you're doing something routine—explaining a concept, brainstorming ideas, writing a first draft of something straightforward—ChatGPT handles it well enough that you save time. But the ceiling hasn't moved. Judgment-heavy work, specialized domain work, anything requiring recent knowledge or real understanding still needs humans.

This matters for how we think about building products. In 2023, the question isn't "can AI replace this job?" It's "where in this workflow can an AI assistant actually reduce friction?" That's a more boring question. But it's the right one.

I've also noticed something about the real limitations that the media glosses over. ChatGPT is frozen in time—its training data cuts off in September 2021. For anyone building software or running a business, that's a massive problem if you need current information. It doesn't have access to real-time data, APIs, or your company's specific systems. Integrating it into actual workflows requires bridging all those gaps.

The other thing that's obvious when you dig in: this technology is only a few months old at scale. We're treating it like it's mature. It isn't. It breaks in unexpected ways. The safety guardrails are inconsistent. And we're still figuring out what the real use cases are versus what's just shiny because it's new.

For founders and builders, I think the takeaway is this: ChatGPT is useful right now. It's not transformative on its own. What matters is figuring out where it genuinely saves your users time versus where it's just adding complexity. The companies that'll win aren't the ones bolting ChatGPT onto everything. They're the ones thinking carefully about which workflows actually benefit.

Where I'm Actually Paying Attention

The more interesting question to me is what comes next. OpenAI has proven the capability. Now it's a question of speed, integration, and specialized versions. GPT-4 is coming later this year. Open-source models are improving fast. The real competition is just starting.

For Arvension, we're watching what this means for ERP systems and mobile workflows. There's a real opportunity to integrate AI assistance into structured business processes—helping with data entry, flagging anomalies, suggesting approvals. But that's going to take work. It's not plug-and-play.

The hype cycle will continue, which is fine. But if you're building software, you can ignore most of the noise and focus on the boring question: where does this actually save users time or friction? That's where the real value is, and that's where the durable companies will be built in 2023.

The AI moment is real. But what's real isn't magic. It's useful tools in the right context.