Jakub Misiło

i have been helping startups & small teams as a product engineer for over 4 years

[Jakub Misiło @misilo_jakub]

coding with AI is like being an early-stage founder with a small team: you architect the solution, tell people what to do, and your team executes it.

the difference is that your team consists of juniors. they copy what you’ve done, reuse general knowledge from the web, or invent their own solutions. unfortunately, they can make a lot of mistakes.

on the other hand, juniors learn quickly - especially when you provide resources such as examples, style guides, design systems, skills, and a clear definition of what you want to achieve. they want to learn, but they need long-term support and guidance.

in my view, that’s the stage AI is currently in. it behaves as I did a couple of years ago, especially when building something: it gets things done quickly, but the resulting solution can quickly become very difficult to work with. sometimes it feels like only the AI that created the project can understand what it has done.

however, with good guidance, templates, and examples in the codebase, the output keeps getting better. even though you still might review its work, AI has sped up development significantly.

p.s. I wrote this take ~3-4 weeks ago, it feels a little bit outdated

[Jakub Misiło @misilo_jakub]

@GoogleDeepMind jokes coming to the end soon

[Jakub Misiło @misilo_jakub]

@typesafeai's Jev has been trending for a while, so I couldn't resist testing it out. it was a perfect opportunity to finish my ai-themed quarter.

so there it is- the model-router package. this utility helps you pick the best model for the job, based on several criteria. Jev is leveraged for making a choice and suits the job perfectly!

check it https://t.co/qHZsX3PuB0

[Jakub Misiło @misilo_jakub]

@wojteklesinski github 👀

[Jakub Misiło @misilo_jakub]

i was thinking a lot about (personal) development lately.

on one hand, learning has never been easier - there are so many resources, and it's so easy to get information.

on the other hand, it's hard to figure out what to learn to stay relevant, and even harder to stick to what you are learning.

i've struggled with that, yet i think i found something that works for me.

i want to pick quarterly themes - a broad topic i explore during the period. i don't plan themes far in advance, i pick them based on where industry is heading.

i already did it - last quarter i picked agents & harnesses. i dug deeper into internals, explored packages like AI SDK and Mastra, harnesses like Pi, software factories, and built my own projects - see my personal agent.

this finally gives me direction for a certain period and guarantees no need to think about what to explore everyday. yet i am still flexible and won't hesitate to explore something else quickly, potentially combining it then with the main theme.

this quarter is coming to an end, but there is still time for one more project, which i am working on now 🤞

also i had to pick up my next theme. choice was quite simple, as i have a pretty obvious gap in my skill set - infra. i will try to explore it in the upcoming months 👀

p.s. attaching some work related to the previous period below

[Jakub Misiło @misilo_jakub]

- https://t.co/R3Zl0RZpte
- https://t.co/iYuB4yDJ4x
- https://t.co/jc3mz5VLxh
- and more... 🤭

[Jakub Misiło @misilo_jakub]

i was thinking a lot about (personal) development lately.

on one hand, learning has never been easier - there are so many resources, and it's so easy to get information.

on the other hand, it's hard to figure out what to learn to stay relevant, and even harder to stick to what you are learning.

i've struggled with that, yet i think i found something that works for me.

i want to pick quarterly themes - a broad topic i explore during the period. i don't plan themes far in advance, i pick them based on where industry is heading.

i already did it - last quarter i picked agents & harnesses. i dug deeper into internals, explored packages like AI SDK and Mastra, harnesses like Pi, software factories, and built my own projects - see my personal agent.

this finally gives me direction for a certain period and guarantees no need to think about what to explore everyday. yet i am still flexible and won't hesitate to explore something else quickly, potentially combining it then with the main theme.

this quarter is coming to an end, but there is still time for one more project, which i am working on now 🤞

also i had to pick up my next theme. choice was quite simple, as i have a pretty obvious gap in my skill set - infra. i will try to explore it in the upcoming months 👀

p.s. attaching some work related to the previous period below

[Jakub Misiło @misilo_jakub]

@EHurts95 17 was good only in the first month or two. the. it got extremely nerfed

[Jakub Misiło @misilo_jakub]

@sekachov silver is 🔥 as well

[Jakub Misiło @misilo_jakub]

what’s next with this bad boy? 🍎

[Jakub Misiło @misilo_jakub]

@Karmedge software might be cheap now but 99% of it is being lost anyway

[Jakub Misiło @misilo_jakub]

@gabedenys 🙋🏻‍♂️

[Jakub Misiło @misilo_jakub]

tip:

take a quad and go for a ride

i know some spots 🤭

[Jakub Misiło @misilo_jakub]

@konnd00 i think it will be

[Jakub Misiło @misilo_jakub]

we get 1% of the model quality they have internally

[Jacob Coxon @hilbertspaess]

I resigned from Anthropic today. I spent the last three years doing pretraining research at both OpenAI and Anthropic. Neither company is acting responsibly. They are racing straight to self-improving superintelligence and gambling with our lives. More thoughts below.

[Jakub Misiło @misilo_jakub]

@OpenAI i am curious how advanced are closed, in-house only models. we get just small % of the power for sure

[Jakub Misiło @misilo_jakub]

my favorite auth abstraction

saves so much boilerplate and makes protected pages so much cleaner 🫧🧏🏻‍♂️

[Jakub Misiło @misilo_jakub]

back after week out. did i miss anything?