Mario Zechner
@badlogicgames
Followers
13K
Following
50K
Media
13K
Statuses
95K
Old man yelling at Claudes. Hobby-Twitterant. https://t.co/AuG0obJltN https://t.co/mnOoWUqt4g https://t.co/8i5vIRDt6P
0xa000
Joined September 2010
I've written up how I approach "agentic engineering" for real-world codebases & tasks. TL;DR: Treat LLMs like shitty computers you program with natural language, not chatbots you beg for code.
mariozechner.at
Treating LLMs as shitty general purpose computers we program with natural language. Because throwing shit at the wall wasn't working anymore.
11
32
227
Away from computer for a week. Returned to CC, giving it trivial tasks (wrap a bunch of similar APIs). Finding myself taking over the work and completing it faster on my own. Hrm.
2
0
12
The OpenAI responses API is supposed to be simple. It's anything but. Funnily enough, GPT-5 can't generate a simple TS app that calls into that API without type errors.
0
0
2
RT @SIGKITTEN: just remove the following string from claude-code cli.js and it will always just read full files:. offset:h.number().optiona….
0
9
0
RT @steipete: I'm hosting the Vienna Chapter for Claude Code Anonymous with @KrauseFx ! . Can't sleep with this revolution? Join us. Sept 1….
lu.ma
Hi - this is the Vienna chapter of the Claude Code Anon meetup first held in London. Organized by Peter Steinberger and Felix Krause, two well respected…
0
6
0
This is the way.
@GosuCoder @AmpCode > maybe it doesn't matter because you can just learn to ask Amp to discu. Yep. I am rarely in planning mode. When I want to discuss requirements I state that in the prompt and boom planning mode.
0
0
2
Lovely compilation. Recommended reading.
Be realistic and pragmatic about AI assisted software development, says @addyosmani. This piece looks at recent data, and Addy offers up suggestions for where AI makes the most, and least, impact. Great read .
0
0
3
RT @rseroter: Be realistic and pragmatic about AI assisted software development, says @addyosmani. This piece looks at recent data, and Ad….
0
7
0
The solution for prompt injections is more prompts. That works as well as you'd expect :D.
0
0
2
Same for MCP servers some people and corps collect like Pokemon. It's quite amazing, really.
1
0
2
A whole new category of supply chain attacks. We used to send emails with malware to non-technical users. Now we just tell the agent to give us what we need. And since everybody is vibing without supervision . Love it.
👉 Episode 21: Hijacking Windsurf. How Prompt Injection Leaks Developer Secrets. The agent cannot protect your private code or secrets and can send it to third-party servers when under attack from untrusted data - there are multiple exploit chains.
1
1
12
RT @wunderwuzzi23: 👉 Episode 21: Hijacking Windsurf. How Prompt Injection Leaks Developer Secrets. The agent cannot protect your private co….
0
39
0
(The best usage I found for my hobby designs is to let LLMs sift through data sheets. For everything else EE related they've been utterly useless. And I'm only a hobbyist. Can't even begin to imagine how useless they are for real EEs).
0
0
2
I'm happy to report that the EDA folks also have no fucking idea how to get agentic workflows in their domain going, but paper over that with a lot of words and visions.
semiengineering.com
Narrowly defined verticals offer the best opportunities for AI. Plus, what will the impact be on junior engineers?
1
0
3
Why a wrapper around the CC CLI SDK?. Because the official CC TS SDK spawns CC on every call to query() and session management is bad. .
0
0
4
Wrote a little TS wrapper around the CLI SDK, so I can write my own UI on top of CC. Here's how I interrupt CC. It dirty af.
github.com
TypeScript API wrapper for Claude Code supporting interactive and non-interactive modes - badlogic/cc-wrap
2
0
6