I built myself an iOS app for calorie/sodium/blood pressure tracking, meal prep planning. The main interface is a chatbot backed by an LLM.
I tell it "tonight i'm eating 150g of rice" etc, it computes calories based on the CIQUAL database (or it estimates), logs it. I set a daily goal and it helps me keep on track.
Over time, I've added the sodium and blood pressure tracking after my doctor mentioned I might have hypertension, so now it helps me keep my sodium intake in check.
And finally I've added meal prep planning: I can scan the barcodes of everything I buy and it can prepare meals that comply with my daily goals. To be honest that one is good in theory but right now I don't have the discipline to scan everything.
I’m using Nono in my own coding agent, spawning a new confined process per run of my agent loop. It works fine but I had to use nono wrap because the SDK do not ship the builtin profiles which was a little annoying.
I'm not looking forward to playing vibe coded games, but yes I feel like models are good enough today to write small scoped games. I used Sol and GLM 5.3 Flash to produce a prototype of a tower defense game for myself, and it works just fine.
I _am_ looking forward however to game developers using AI to debug and optimize their games. Let's squeeze all the performance we can get from existing hardware.
Well said. I wonder whether AI games from low-effort prompts will play like most AI written blog posts read, superficially great but on some level not connecting with the user.
This is an ad for their product, to solve a problem they created themselves that anyone could have seen coming from a mile away. Not sure this is a wise choice.
Anyone using this for their own personal email ? I'm also somewhat confused about the enterprise version, it's only a license and you're still responsible for self hosting it ?
I largely have the same feeling: I used to love it but today it's hard to justify coding by hand still. Even for my personal projects I no longer code by hand.
But it's not all bad because with an agent I'm able to achieve much more. I loved debugging an issue and diving deep but an agent with access to observability data and the code is so so much faster that it makes no sense to _not_ use one. I always wanted to have sophisticated debugging/introspection tooling at work but it nevers get prioritized; now it's doable by an agent almost autonomously. And the same goes for performance and profiling: point an agent at pprof for example and you can test all kinds of stuff without lifting a finger.
However I agree about "the zone", working with agents doesn't feel as good as working by hand.
Thanks for engaging with the post! Totally agree about using agents for observability and performance. It's fun to think of new abstractions to build so that agents can interrogate this and other data more easily and cheaply.
The "usual fairground airbrushed abominations" at least have a distinctive style and mean that someone cares enough to do more than type into a computer a few times and say "eh, good enough."
Why imagine the mental state of the person using the tool? You have no idea one way or the other how much a hypothetical person cared. Also, what does it mean to "care enough"? Can we measure and quantify caring? What if it turns out the AI user cared more under whatever metric you provide, would that change your mind about the quality of the output?
Yes, by observing the quality of the output and means of production. If it's the standard "AI house style" one gets when prompting image models, by which I am immediately and viscerally repelled, then I am confident assuming it's a low effort "typed in a prompt, got a thing, called it a day" result, rather than actually applying some taste and judgement and trying a bunch of different things in different styles and adjusting things until it's something nice.
You know how there are movies that are so bad that they become entertaining to watch just so you can see how much of a trainwreck they are and how many rules they manage to break - and then are movies that are bad in the sense of sub-mediocrity that is drawn out, repetitive and completely uninteresting to anyone?
The art you posted is like the first example. See, someone even made a blog all about it, they enjoy picking that stuff apart. Repetitive AI art created by typing out four words or pasting a Facebook post into the prompt text box is the second. People don't even entertain themselves by looking at how bad it is, they just plain hate it.
Also, how much power does that consume at full tilt ?
reply