At Dellecod, we’ve been building software products for over a decade. Every few years, we’ve seen a new wave of technology promise to change everything. Mobile. Cloud. No code. Some fulfill that promise. Some don’t. Right now, we’re deep in the storm surge of generative AI — and like so many in the industry, we’re trying to figure out which direction the wind is actually blowing.
Benedict Evans recently compared AI to the early internet or smartphones — a potentially seismic platform shift that's just starting to show its contour. That framing stuck with us. Platform shifts in tech don’t just produce new tools. They change the kind of software people expect, how they use their devices, and even what they believe software should do. They turn questions about product features into questions about capability, experience, and value at a fundamental level.
Right now, we’re seeing incredible numbers: ChatGPT reportedly serves 800 to 900 million people per week. That’s not trivial. But scratch the surface and a pattern emerges that’s familiar to anyone who's ever launched a free app — massive curiosity, scattered use, and a smaller core of power users. Only around 5% are paying. Even fewer are integrating it meaningfully into their workflows.
In other words, it’s still early. As early as 1997 felt for the internet. Back then, we knew something big was happening, but couldn’t yet point to the ubiquitous applications that would become indispensable. We had early enthusiasm, speculation, and plenty of false starts. We also had no real idea what business models would survive.
AI feels similar, but with higher stakes. The rate of investment is staggering. Billions are flowing into foundational models and infrastructure. Compute efficiency is improving 20 to 40 times year over year. The scale and ambition are immense, but so is the uncertainty. What is AI really good at, today? Which tasks or industries will it actually transform, and which will remain largely untouched?
So far, the clearest use cases seem to cluster around certain types of white-collar work — coding, marketing content, some forms of analysis or summarization. We’ve seen real productivity gains, especially when tools are embedded into existing workflows. But we’re still missing the breakout product — the Excel or iPhone of this era — that will anchor AI into how non-technical people think and work every day.
That absence is important. A technology becomes mainstream not when it’s ubiquitous in theory, but when people feel they can’t do their job without it. Until AI finds its way into opinionated, domain-specific, well-crafted software, it’s just potential. We’ve always believed in making software that fits closely with how people already work, then nudging that behavior forward. Successful AI adoption will follow the same path.
There’s debate about whether the value in AI will flow to new players like OpenAI or Anthropic, or toward the incumbents — Microsoft, Google, Meta — with trillions of dollars in reach and infrastructure. In practice, we think much of that value will be captured by the companies that don’t just build the model, but manage to translate its strengths into usable, trustworthy interfaces. Interfaces that feel like software — not research demos.
Today, many AI tools still feel like power tools with the safety guard removed: capable, but unimprovised for daily business life. That will change. The most useful generative AI products in five years may be ones you barely recognize as “AI” anymore. They’ll just be part of the UI, the logic, the decision tree.
The conversation about AGI — artificial general intelligence — is another terrain altogether. People often throw around “AI” to describe whatever feels new and surprising. Whether AGI is five years away, already partially here, or a mirage entirely misses the more pragmatic question: what, exactly, are large language models useful for today?
For now, they are fast, fluid engines for pattern recognition and language generation. That’s not trivial. But it’s also not limitless. We think it’s worth remembering the history of platform shifts: they don’t gift people superpowers overnight. They take time to trickle into workflows and back offices. It took spreadsheets years to replace ledgers. Even the iPhone needed the App Store before it became indispensable.
None of this means AI isn’t transformative. It just means transformation doesn’t arrive all at once. It descends through stages — experimentation, infrastructure building, accidental discovery, productization, market consolidation, and eventually, abstraction. By the time we figure out what “AI” really is, we might not call it AI anymore. It might just be how software works.
To us, the most exciting part of this moment isn’t the hype, or even the demos — it’s the open questions. What products will people really need? What new expectations will users have? What definition of “good software” will this wave rewrite?
We don’t have all the answers, and we’re okay with that. But we’re watching closely, building carefully, and betting that the meaningful stuff — the jobs-to-be-done, the workflows, the things people love and return to — those will matter as much in the age of AI as they ever have.