Michał Prządka

AI Consultant. Bridging tech, data & business. Lifelong learner, sports addict, father, vegetarian. LI profile:

[Michał Prządka @przadka]

Why the gradient is a list of partial derivatives?

https://t.co/I2vyyPYV98

[Andrej Karpathy @karpathy]

Personal update: I've joined Anthropic. I think the next few years at the frontier of LLMs will be especially formative. I am very excited to join the team here and get back to R&D. I remain deeply passionate about education and plan to resume my work on it in time.

[Michał Prządka @przadka]

Original post:
https://t.co/qUq9NsgVU6

[Michał Prządka @przadka]

Last September I argued that Claude Code was becoming less of a coding assistant and more of an OS for agentic workflows.

I also wrote: “Cursor could do the same if they shipped an SDK.”

Cursor shipped theirs on Apr 29.

Great to see this direction becoming real.

[Michał Prządka @przadka]

Next frontier: deploying these agent harnesses across non-engineering domains.

[Michał Prządka @przadka]

The next developer platform will look like GitHub and Telegram had a baby. And agents will be its primary (only?) users.

[Michał Prządka @przadka]

read Mario Zechner's "slowing the fuck down" post. mostly skill issue dressed as wisdom, partly harness issue, partly model capability. not buying the doom narrative.

[Michał Prządka @przadka]

Bought a @system76 Lemur Pro for the Linux experience and the "right to repair" promise. Tab key broke. Their solution: $175 keyboard assembly + $143 shipping to Europe. For one key. No individual parts, no EU option. The repair-friendly laptop I can’t repair.

[Michał Prządka @przadka]

Bill is a senior engineer. Bill isolates and pins. Be like Bill.

[Daniel Hnyk @hnykda]

LiteLLM HAS BEEN COMPROMISED, DO NOT UPDATE. We just discovered that LiteLLM pypi release 1.82.8. It has been compromised, it contains litellm_init.pth with base64 encoded instructions to send all the credentials it can find to remote server + self-replicate. link below

[Michał Prządka @przadka]

After three weeks using speech-to-text as my main interface to coding agents, I can't imagine going back. @OpenWhispr

[Michał Prządka @przadka]

Two people + agents can now deliver what a 30-person agency did 5 years ago.

[Michał Prządka @przadka]

@gkossakowski lol, maybe, but read this: "The agent could reason over the diff, call tools, and decide where..." - I dont think any human at @cursor_ai would try to make a point that agents with tools are better than agents without tools... not in 2026 :)

[Michał Prządka @przadka]

Just read this: https://t.co/5rjGqI7wqy

Interesting ideas buried under layers of AI-generated filler.

[Michał Prządka @przadka]

@aye_aye_kaplan no you didnt

[Michał Prządka @przadka]

The most important directory on my laptop right now is /home/michal/agents.

[Michał Prządka @przadka]

@HamelHusain I also do this a lot and I wondered who is best reviewer: https://t.co/FYiAg8t6qD

[Michał Prządka @przadka]

Unsupervised eval for coding agents: https://t.co/Vmb8Ty7dmA Agents playing treasure hunt! Challenger agents hide bugs, reviewer agents try to find them, and an LLM matcher scores assignments. No human labeling.

[Michał Prządka @przadka]

@LeszekAI Pełna zgoda - warto korzystać z różnych modeli/cli

[Michał Prządka @przadka]

Unsupervised eval for coding agents: https://t.co/Vmb8Ty7dmA Agents playing treasure hunt! Challenger agents hide bugs, reviewer agents try to find them, and an LLM matcher scores assignments. No human labeling.

[Michał Prządka @przadka]

Hey, Claude Code, please don’t code. New blog post -> https://t.co/qUq9NsgVU6