Beyond the Agent Loop
Building an AI agent used to mean building half a platform around it. Persistent sessions, context management, tool orchestration, recovery, subagents. The plumbing could be more work than the intelligence. OpenAI's Agents API changes that equation. And it raises an interesting question: what will developers build when they can finally stop building the machinery?
AI Decision Models: From Reasoning to Action
AI is learning to think. But can it make the right call? A new generation of decision models promises to turn AI from a generator of words into a component that makes structured, actionable judgments. From autonomous agents to automotive voice systems, this could fundamentally change how we architect software. The real opportunity may be closer than you think.
A Million Apps, Not a Single New Idea
AI can build almost anything in a weekend. But where are the new ideas? We are witnessing an explosion of software, celebrated as an unprecedented wave of innovation. Yet much of it is simply the reproduction of things that already exist. Have we entered an age of invention, or just industrialized imitation?
Your Car Can Talk. It Still Takes Orders.
Modern LLMs have made automotive voice control far more flexible, allowing drivers to speak naturally, correct themselves, and maintain context. Yet linguistic flexibility does not necessarily create conversational interaction. The next step is moving beyond smarter commands toward real-time, agentic systems that understand goals, adapt dynamically, and participate in genuine dialogue.
Prompts Are Programs
Andrej Karpathy's remark that "the hottest new programming language is English" has become one of the defining observations of the LLM era. I believe it is only half the story. The more interesting question is not whether English is becoming a programming language, but what kind of programming language it is.
The Last Expensive Resource
Every technological revolution makes a once-scarce resource abundant. Steam reduced the cost of muscle, computers the cost of calculation, and the internet the cost of communication. AI may be doing the same for practical reasoning: as the computational cost of machine reasoning falls, the organizational cost of human reasoning may fall with it.
How to Get Rich with AI (Fast)
Search for AI business ideas and you’ll drown in the same recycled suggestions everyone else gets. That’s not where fortunes are made. The real opportunity isn’t finding another application for AI. It’s discovering problems that nobody even realized were AI problems in the first place.
“AI does not create value”
On the surface, this statement appears to be indefensible. Around the world, companies are injecting billions of dollars into AI precisely because they expect to see value creation—increased efficiency, smarter decisions, innovative products. Yet, the claim holds an important truth. AI does not create value merely by existing.
If AI Isn’t Intelligent, Are We?
Dismissing AI as non-intelligent may feel safe, but it carries a hidden implication: much of what humans do might not qualify either. The debate exposes a deeper tension—not about machines, but about how fragile, flexible, and perhaps self‑serving our definition of intelligence really is in practice, especially when it threatens our sense of uniqueness.
Do We Still Need Software Architects?
If AI can generate working software in minutes, why does software architecture still matter? As prototyping becomes cheap and iteration nearly instant, traditional upfront design starts to look increasingly obsolete. Yet this shift raises a critical tension: while speed and experimentation improve, we may be trading away long-term structure, maintainability, and scalability in the process.
