🔮 Sunday edition #532: The speed trap in AI, Grok’s ecosystem play, and truth by citation++

An insider’s guide to AI and exponential technologies

Hi it’s Azeem,

This week, I’m thinking about structural advantages, the hidden forces that determine who wins when technology shifts.

In climate, a paradox: could the anti-green agenda actually accelerate innovation? In AI, we explore why speed, not just smarts, could shape the next phase and how xAI may be quietly building an unmatchable platform advantage. Plus, a parable from the age of horses that every CEO should revisit.

---

When going backward drives progress

Over the past decade, the big lie of the energy transition was that environmental virtue would drive change. In reality, power – and thus the ability to create change – flows to whoever can produce energy and critical materials at the lowest global price. The Trump administration may have sidelined environmental ideals, but it can’t undo that underlying economic fact.

VC investor and EV member Vinod Khosla suggests something counterintuitive, that America’s rollback of green subsidies might unlock the next climate breakthrough. If mature sectors like wind and solar no longer need state support, that money could be freed up for riskier bets: fusion, super-hot geothermal, or low-carbon steel. Once a technology becomes economically viable, continued subsidies risk distorting the market and propping up winners that no longer need support.

The bet only works, of course, if funds don’t end up propping up coal. But the fundamentals have shifted. In most places, clean energy is already cheaper than fossil fuels. If the US wants to outcompete China, it has no choice but to lean in.

---

Does AI actually slow you down?

AI was supposed to save time. In software development, it might be doing the opposite.

A new METR study found that experienced open-source developers using early-2025 AI tools (Cursor Pro with Claude 3.5 and 3.7) actually took 19% longer to complete tasks. Developers had expected a 24% speedup. Experts had predicted even more. In reality, AI added overhead when used in complex projects.

The main drag on performance came from the time spent prompting, waiting, and verifying outputs, often compounded by over-reliance on flawed suggestions.

Anecdotally, I’ve seen the same issue elsewhere.