There’s something quietly remarkable happening in AI right now. Not just in the progress we're seeing with large models or the velocity of infrastructure investment, but in how we talk about intelligence and creativity more broadly.
We recently listened to a conversation between Marc Andreessen and Ben Horowitz about the present and future of AI — and it’s stuck with us. They brought a level of clarity that's rare: AI, particularly modern language models, is already outperforming most humans in areas we once considered uniquely ours. And that strikes us as a turning point.
When people say LLMs aren’t truly creative, they’re often comparing them to outliers — the Beethovens, the Einsteins, the Van Goghs. But most human creativity isn't expressed like that. In reality, most of what we call innovation is remix. We take old ideas, trace threads across disciplines, and apply things in new places. If AI can already do that — and it increasingly can — then we’re not just inching closer to artificial general intelligence. We’re reshaping what counts as creativity at all.
What feels even more noteworthy is how these systems have started to demonstrate something akin to "theory of mind" — the ability to infer what others are thinking, feeling, or intending. This was long considered uniquely human, or at least deeply tied to lived experience. But now we see chat models simulating conversations that intuit nuanced motivations. False consensus is still a common trap — left to their own devices, they often agree too readily — but under some direction, they can generate and explore diverse viewpoints with depth.
Still, raw intelligence isn’t everything — and may not even be the most important part of success. In leadership, courage and drive often matter more. Emotional awareness, timing, and trust can’t yet be encoded into vectors. This is as true for co-founders and product managers as it is for diplomats. Most meaningful decisions aren’t made by calculating probabilities; they’re made in the presence of doubt, fear, ambition, and instinct. Today’s AIs don’t live emotional lives. Which means there’s still a wide gulf between simulating empathy and actually having skin in the game.
That said, the infrastructure supporting AI is real, and it’s expanding fast. A full 1% of GDP is now going toward AI hardware and energy — a figure that would have sounded implausible even two years ago. If this is a bubble, it's an unusually functional one. Customers are buying, systems are improving, and for once the demand curve appears both steep and rational. Scarcity — especially in talent, chips, and compute — is real, but solvable. Markets tend to solve scarcity, if not perfectly, then at least directionally.
There's also the question of geopolitical leverage. China is catching up quickly — not just in training large models, but in scaling and deploying them across industry. That shouldn’t be surprising. They’ve maintained their industrial base while much of the West chased leaner software margins. In robotics especially, manufacturing know-how becomes as strategic as data or funding. The future of embodied AI might be decided as much by assembly lines as algorithms.
We also appreciated how Andreessen and Horowitz reminded us that we’ve been here before — at least in form, if not in magnitude. Before GUIs, most people couldn’t use computers. Before browsers, the internet was a niche curiosity. What comes after chatbots may not resemble chat at all. Interfaces have a habit of surprising us — especially interfaces that become invisible. So while today’s LLMs are powerful, they might just be the scaffolding for something else: a new interaction paradigm we haven’t seen yet.
For us at Dellecod, this means staying humble and imaginative in equal measure. When assumptions are being rewritten this quickly, the temptation is to double down on what’s worked. But this moment rewards first-principles thinking. It asks us not just to adapt products, but to rethink categories. To consider what happens when the marginal cost of intelligence approaches zero — and what entirely new things become possible in that environment.
We have no illusion that this will be easy. Some of the entrepreneurial muscles we built in earlier chapters of tech may not serve us here. But the upside of uncertainty is possibility, and the best way we’ve found to navigate it is to stay close to the user, close to the truth, and open to wild ideas.
Because the future of AI isn’t predefined. It’s still being built — sometimes by humans, sometimes with their help. And the more we treat that process not as inevitable, but as something participatory and creative in its own right, the better our chances of building things that deeply matter.