Szymon Błażyński

AI Native Engineer

[Adam @_overment]

so stoked about this.

[Szymon Błażyński @szymonblazynski]

@miroburn Informacyjnie, w najbliższy wtorek w Łodzi współorganizuje podobne wydarzenie o AI w firmach i sprzedaży https://t.co/qtlGbIjDQt

Jak ktos jest w okolicy to zapraszam!

[Szymon Błażyński @szymonblazynski]

@theo Thats the reason I believe autocorrect can make more trouble than benefits

[Szymon Błażyński @szymonblazynski]

@bzagrodzki 🫡

[Szymon Błażyński @szymonblazynski]

@_overment It never resets, and it worked hands-on instead of routing: 242 reads, 98 edits, started dev servers 12 times. Plus my own policy of a fresh reviewer per visible slice."

[Szymon Błażyński @szymonblazynski]

@_overment Summarized issue by coordinator:

"Pulled the logs, correction: 10 workers + 9 reviewers, not 14/14. Jobs were fine, reviewers ~$0.50 each. The cost was one coordinator session: 593 turns, 814 tool calls, 255k peak context, 88M input tokens. More than all 33 jobs combined.

[Szymon Błażyński @szymonblazynski]

@_overment Love it, I will do

[Szymon Błażyński @szymonblazynski]

@_overment And as much as I love this approach, in my current state I have the feeling that it is time- and token-inefficient. Do you have any tips, how I can approach this inefficiency?

[Szymon Błażyński @szymonblazynski]

@_overment I have strongly used Limen for a couple of days and I get the feeling that it's much more token consuming.

I use Hermess instead of Grok bot, where I do specs. For instance, a pretty easy change on frontend required 14 agents for implementation and the same amount of reviewers.

[Szymon Błażyński @szymonblazynski]

It keeps going...

[Nicolas Camara @nickscamara_]

introducing anydoc

now your agents get 100x faster local parsing for pdf, docx, pptx & 10 more formats

- sub-5ms md conversion
- 500 docx files in 1.7s
- top quality across all 13 formats
- rust-based
- open source

already powering @firecrawl /parse

https://t.co/fVsRsYFNyC

[Pi @pidotdev]

Read the full blog post here

https://t.co/AEbmCZI1fE

[OpenAI @OpenAI]

We quietly released the open-source Codex Security CLI, but Hacker News found it before we had a chance to share it here...

You can now use it to scan repositories, track findings across runs, verify fixes, and add security checks to CI/CD.

This is an early release, and we're listening to your feedback as we continue improving it.

[Adam @_overment]

https://t.co/sJtkgaxYXQ

[Szymon Błażyński @szymonblazynski]

@pivovarit I think the main reason for that is, for whole years we have tools and mental models which help us to do so. It's only the matter of time when we create these tools and processes and it will be much more deterministic

[Szymon Błażyński @szymonblazynski]

@mattpocockuk Trying to create an entire application at once. “Build me a SaaS that makes money for me.” 😜
Also, many people (myself included) create cognitive debt during development. The more they use it, the less they understand the processes behind it.

[Szymon Błażyński @szymonblazynski]

@mattpocockuk Look what you have done @mattpocockuk 😅

[Cloudflare @Cloudflare]

Time to consider not just human visitors, but to treat agents as first-class citizens. Cloudflare’s network now supports real-time content conversion to Markdown at the source using content negotiation headers.
https://t.co/B7wYH4PtA8

[Szymon Błażyński @szymonblazynski]

-19% skuteczności dzięki AI. Badanie Harvardu pokazuje, że AI w złym miejscu pogarsza wyniki. Gdzie przebiega granica opłacalności?

https://t.co/c8pROWyHvU

[Szymon Błażyński @szymonblazynski]

@CharliLaFerrari @Hesamation One the best modes, is to create new session as fast as possible. Effective context window is much smaller than providers announced. If you are in the middle of long session, just summarise what you and assistant said, and start a new session.