Jonathan Low

Building Mistle - Open source background agents

[Jonathan Low @jonathanlowhy]

@kunchenguid In the long term, this will lead to talent drain.

[Jonathan Low @jonathanlowhy]

@GergelyOrosz That's like $22k a month of usage. Seems hard to rationalize in the long term relative to an engineer's pay.

[Jonathan Low @jonathanlowhy]

@kirso_ @dbsbank This is all good marketing right? 🙃

[Jonathan Low @jonathanlowhy]

@hugorcd @linear Does this entirely replace GitHub for reviews?

[Jonathan Low @jonathanlowhy]

@t3dotchat Was it confirmed how much was donated?

[Jonathan Low @jonathanlowhy]

@kimmonismus That seems way higher than expected. I usually end each week with about 10% left of ChatGPT Pro. And the estimated spend is about 7k. So maybe at most, 9k max. Still, a crazy great deal and this is why I'm going to stick with GPT 5.5 even if Fable is better.

[Jonathan Low @jonathanlowhy]

I❤️Codex

[Jonathan Low @jonathanlowhy]

What we wish vs what actually happens

[Garry Tan @garrytan]

Everyone thinks AI coding tools set founders free.

Watch what people actually build with them: rules, approvals, process, layers. The same cage, assembled faster.

The tool that can scaffold anything in an afternoon will scaffold your bureaucracy in an afternoon too.

Speed of construction is speed of calcification. Build the thing that lets you create new things: experiences that didn’t happen before.

[Jonathan Low @jonathanlowhy]

@garrytan What we wish vs what really happens

[Thomas Jiang @thomasjiangcy]

you can do this and much more with @mistledev 😀(also, you can use your existing subscriptions directly, no need to buy credits!)

[Jonathan Low @jonathanlowhy]

@kevins8 Hmm. Ok. Good food for thought. Thanks.

[Jonathan Low @jonathanlowhy]

@JTLonsdale Apple stands out here.

[Jonathan Low @jonathanlowhy]

@sattyyouneed I have an altar at home where I pray for codex resets. So far, it has worked very well.

[Jonathan Low @jonathanlowhy]

@craigweiss I thought it's all about the launch video.

[Jonathan Low @jonathanlowhy]

It's always good to keep an open mind and sample from both ends of the spectrum. But the sweet spot is more nuanced and lies somewhere in between.

One can absolutely build loops that work and reduce token cost. It just doesn't happen by itself. It takes work and experimentation.

[Alex Finn @AlexFinn]

Everyone on X talking about building AI coding loops

I think for 99% of people it's horrible advice. Mostly being promoted by people who profit off token burn

Unless you are deeply technical, you are going to build flimsy apps and burn outrageous amounts of money

Engineering a loop that produces high quality code requires insane amounts of setup and oversight to make sure it's done the right way

For a vast majority of people the best AI coding strategy is:

1. Work closely with your agent, ask it to build the next smallest people of functionality you can think of
2. After each small chunk, ask the AI what it did. Tell it to describe it to you concisely
3. If anything is too technical, ask it to explain in a non technical way so you deepen your knowledge
4. Have it continuously maintain a current state doc you can feed to another LLM to get 2nd perspectives on crucial decisions
5. Mute influencers who give you vague advice that doesn't really mean anything

This will actually save more time in the long run because you won't have to untangle code you don't understand

[Jonathan Low @jonathanlowhy]

@aryanlabde Bugs. Always bugs.

[Jonathan Low @jonathanlowhy]

@brankopetric00 AWS Solution Architect say it works. So it must be true.

[Jonathan Low @jonathanlowhy]

@akaasten https://t.co/ZYM6jQk1V1