I feel like there's a huge political/economic opportunity in the US simply from getting government to shift subsidies from corn/ethanol to solar. Of course you also have to get buy-in from the farmers, but.
If you have a blog post on the topic I'd be interested. My experience with AI coding isn't really the same, but my experience with typing discipline definitely is, and I've thought about writing about it myself but don't really know where to start.
hard to believe anyone who writes something this overwhelmingly positive.
I use a ton of AI, it writes a shit load of code for me, and I eventually get good results.
Its also goes off on ridiculous tangents, implements things i dont want, over engineers almost everything. Makes large code bases convoluted and unreadable. Yes humans also do that, LLMs do it a lot faster.
> Now, with AI coding, I can implement any decent sized feature, exactly how I want, in a few hours and it's all polished and ready, all edge cases handled.
No way a human wrote that, sounds like a late night TV advertisement.
You obviously haven't worked on a good codebase. I've worked on both so I know the difference.
Same feature on a good codebase vs bad codebase is like 1k lines of code vs 20k lines of code and is worse in the latter case. Not an exaggeration.
Clean code is real, it's just not what's been peddled to us.
If your code is full of bugs and you ask Claude or Codex to build more features, it will take 10x to 20x the effort because of workarounds and hacks which keep compounding.
If your code has very few to no bugs and no hacks/workarounds, it will also implement hacks/workarounds and no bugs. The difference is pretty insane actually.
You can try it yourself; create a really low quality codebase, full of hacks and try to build features with AI based on its recommedations.
> Enki (Sumerian: 𒀭𒂗𒆠 dEN-KI), also known as Ea (Akkadian: 𒀭𒂍𒀀 dE₂-A), was the Mesopotamian god of wisdom, crafts, fresh subterranean waters, magic, and incantations.[1][11] He was believed to rule the Abzû. In Mesopotamian astronomy, he was associated with the stars of the southern band of the sky.[17] Enki's wife was Damgalnuna, and their children included Nanshe, Asalluhi, Marduk and Enbilulu. His sukkal (attendant deity) was Isimud. Servants of the god included lahmu, kulullû, and the Seven Sages.
I'm in a similar position. I can easily imagine the kinds of tasks where "agentic workflow" seems useful. I just don't have any of them personally right at the moment, and think the world of software has more of them than the needs of software actually justify (due to previous bad engineering and planning).
> The other thing I don't like about agents is their ability to run any command. You can set the option to review every command it runs, but you're then just hand-holding the agent.
The thing about this is that you're treating "agent" as if it refers exclusively to the products offered for that purpose by the major LLM companies. Nothing stops you from designing and implementing your own, with a security model that you think is more sensible. The models won't have been trained against your specific tools, but you can still give them descriptions of what those tools do and then prompt them to make appropriate use of them. All that's really happening is that the LLM outputs some JSON, and your client detects that the output was a tool call (instead of a direct response), parses JSON, invokes whatever other code with whatever parameters and perhaps writes back a response.
Maybe it’s a personal thing but I found that at some size it would be like having someone chuck a big foreign codebase in your lap and saying, “Ok, I did most of it, now you make it work.”
I kinda have to stay engaged the whole time the best I can.
What that boils down to is, “go ahead and yolo small things I can fully review” or “really wrangle it close through a larger effort” or “go full vibecode and prayer.”
> As a historical U.S. benchmark, an average 18-hole facility had about 80.7 irrigated turf acres and used approximately 152.5 acre-feet/year—about 49.7 million gallons (188 million litres) annually.
So about 220.
But it's still, ahem, a drop in the bucket compared to the California almond industry.
A while back I satirically suggested https://simonwillison.net/2026/Jul/17/spot-birds-not-golf/ that if hyperscalers wanted to relieve the pressure over data center water use they could buy up a few dozen country clubs and turn them into public parks.
I'm surprised that the AI didn't point this out to you.
Also regarding the blog post:
> So how many times would the age of the universe fit into hearing every possible shuffle playthrough of MiniDisc?
The number you've cited is the age of the universe in years, and it takes much less than a year per playthrough, so an adjustment is necessary to those calculations.
Yes, a good analogy. Except the cat actually follows the football rules and can beat some humans. And has no physical limitations to play other kinds of sport that you might imply.
More importantly, beginner human players don't exhibit that tendency. The history of the position doesn't bother a human (except as required for castling and en passant rules), and the analysis becomes generally easier as pieces come off the board.
Humans do make these errors when playing blindfolded. If you even the playing field and give the LLM the position at each turn, it does not make mistakes.
reply