Bartek Szczepański

Homeschooling dad. Coding fast @ NightWoods. Building tools for how learning actually works. My homeschool community is the lab.

[Bartek Szczepański @barnski_pl]

@ClaudeDevs The resolution boundary is drawn in the wrong place. This is a stopgap with an intentional gap, not a permanent fix — so
the seam belongs somewhere else.

[Bartek Szczepański @barnski_pl]

@FIBAWC The GOAT of FIBA, the world champion Dennis of course. French intellectual and Serbian horseman don’t belong here.

[Bartek Szczepański @barnski_pl]

@sama Imagine that author of this post is still more likable than Dario. Incredible race

[Bartek Szczepański @barnski_pl]

@TheDunkCentral @ESPNMilwaukee I’m a proponent of remote work as well.

[Bartek Szczepański @barnski_pl]

@TheAndyCortez I guess he has to ask Steph Curry how to handle physicality like a normal person

[Bartek Szczepański @barnski_pl]

@NoTechBen I will miss this content now when the series is over :(

[Bartek Szczepański @barnski_pl]

@TerribleMaps It’s a terrible map

[Bartek Szczepański @barnski_pl]

@NoTechBen What would happen to your hate if Giannis really landed in OKC?

[Bartek Szczepański @barnski_pl]

@mattpocockuk @petermekh I edited grill-with-docs to suggest possible next steps for me. Next session Codex - „great, now I have everything, let me jump straight to next step - implementation”.

[Bartek Szczepański @barnski_pl]

It depends on the use case. Most of the "regular" cases are pretty common and they are customized from "the inside". Hermes' building blocks are on the right level of abstraction - gateway, cronjobs, computer use, browser use, etc. Me and the agent then compose them together achieving custom behavior. Also, Hermes is really happy (too happy IMO) to edit itself.

[Bartek Szczepański @barnski_pl]

I thought chat memory is important. Its responses must get better if they know my story right?

Me everyday: „Tell me about Foo would it be better than Bar?”
Claude: „Given your work on XYZ the interesting question isn’t…”

😭

[Bartek Szczepański @barnski_pl]

@jlongster Look at what?

[Bartek Szczepański @barnski_pl]

@GergelyOrosz I agree. Although the phrase “software that is buggy / has poor UX” is funny in the context of replacing JIRA.

[Bartek Szczepański @barnski_pl]

@ThePrimeagen Supermaven is fantastic. Is there a cursor tab equivalent for neovim?

[Bartek Szczepański @barnski_pl]

@niebezpiecznik Powodem oszustwa jest chciwość, brak edukacji oraz jedzenie majonezu innego niż Kielecki. Facebook co najwyżej jest medium, przez które oszustwo się dokonuje.

[Bartek Szczepański @barnski_pl]

All the “well, actually” comments here about why Anthropic is not blocking anything usually miss the point. Attitude and perception matter. Not just the technical facts. Anthropic is not Apple yet. It cannot afford to casually play with developers’ trust and goodwill. We'll see whether they course-correct.

[Gergely Orosz @GergelyOrosz]

Huh.

So Anthropic doesn’t want an ecosystem built on top of their SDK.

Message received.

Between them banning OpenCode, their hostile stance for OpenClaw, and now this: Anthropic make it they don’t want 3rd parties to build on them.

Bug win for eg Codex, MinMax and others…

[Bartek Szczepański @barnski_pl]

I think it is good enough for most purposes but also it’s great that you are concerned about all those things. Being concerned allows us not to blindly accept nor discard the tool but rather decide if it’s good enough for the project we are currently working on. You being concerned additionally produces articles I really appreciate ;)

[Bartek Szczepański @barnski_pl]

For example, when I made an LLM do TDD and commit every step, I felt like it was easier for me to review the code later and see the whole process in chunks that fit in my head. This skill was actually helpful and anecdotally produces better code. That's why I got inspired by your post to maybe finish feature development with a "test critiquing" LLM. We'll see if this can improve the work or just add more complexity without any value.

[Bartek Szczepański @barnski_pl]

@ploeh I guess in the end I can't trust it 100%. What I can do with all the knowledge I've gained from my experience and posts like this is layer LLM "agent" on top of "agent" and make my final review result in less rework. I usually try different setups and see if it actually helps.