I mostly agree with you, especially on software design being underappreciated. A lot of what slows teams down today isn’t typing code, it’s reasoning about systems that have accreted over time. I am thinking about implicit contracts, historical decisions, and constraints that live more in people’s heads than in the code itself.
Where I’d push back slightly is on framing this primarily as an LLM limitation. I don’t expect models to reason from first principles about entire systems, and I don’t think that’s what’s missing right now. The bigger gap I see is that we haven’t externalised design knowledge in a way that’s actionable.
We still rely on humans to reconstruct intent, boundaries, and "how work flows" every time they touch a part of the system. That reconstruction cost dominates, regardless of whether a human or an AI is writing the code.
I also don’t think small teams move faster because they’re shipping lower-quality or more experimental software (though that can be true). They move faster because the design surface is smaller and the work routing is clear. In large systems, the problem isn’t that AI can’t design; it’s that neither humans nor AI are given the right abstractions to work with.
Until we fix that, AI will mostly amplify what already exists: good flow in small systems, and friction in large ones.
I've been thinking about why AI seems to accelerate some teams dramatically while leaving others mostly unchanged. This post is an attempt to articulate what I think is missing: not better tools, but better routing of work, context, and ownership. Curious how this resonates (or doesn't) with others.
This news brought joy back to my life. I live in Berlin and often rent these Miles car share cars. The ID3 is an absolute joy to drive, way better than the Tesla's out there. But for me, those touch buttons are an actual nuisance.
Thank you for your insightful response! I completely agree that effective human-AI collaboration hinges on a robust interface that enables meaningful two-way communication. Whether through sensory-motor interactions or more abstract cognitive interfaces, achieving a truly symbiotic relationship is crucial.
The question of what humans can offer to AI—especially in the context of autonomous, self-aware AGI—is an important one. While today’s AI doesn’t yet necessitate human input in the way a biological symbiont might, there’s still immense value in human intuition, ethical reasoning, creativity, and the ability to frame problems in ways AI cannot. Of course, the world I envision is one where humans are still our priority and the world is shaped to maximise for human happiness. It definitely is not the current state.
Even a superior AI (in some domains) could still benefit from human collaboration, not just as an input provider but as a co-evolving counterpart.
Your point about AI interaction tools is particularly compelling. Historically, technologies that seamlessly integrate into human habits—touch, vision, natural language—have seen widespread adoption. AI needs interfaces that lower friction and increase accessibility, much like how smartphones and browsers revolutionized digital interactions.
AI-human collaboration is at a critical juncture. As AI agents increasingly integrate into our digital workflows, I’ve been thinking about a fundamental challenge they face: the disconnect between how humans interact with software and how AI systems access it. Most application APIs are designed for data exchange, not for mimicking human interaction patterns. This gap is preventing AI from truly enhancing our productivity in the ways we’ve imagined. The solution might lie in an unexpected place: accessibility APIs, which could revolutionize how AI understands and navigates human-centered interfaces.
Where I’d push back slightly is on framing this primarily as an LLM limitation. I don’t expect models to reason from first principles about entire systems, and I don’t think that’s what’s missing right now. The bigger gap I see is that we haven’t externalised design knowledge in a way that’s actionable.
We still rely on humans to reconstruct intent, boundaries, and "how work flows" every time they touch a part of the system. That reconstruction cost dominates, regardless of whether a human or an AI is writing the code.
I also don’t think small teams move faster because they’re shipping lower-quality or more experimental software (though that can be true). They move faster because the design surface is smaller and the work routing is clear. In large systems, the problem isn’t that AI can’t design; it’s that neither humans nor AI are given the right abstractions to work with.
Until we fix that, AI will mostly amplify what already exists: good flow in small systems, and friction in large ones.