AGENTS

→ View original post on X — @steipete

AI Security Framework: Sandbox, Allow-lists, Access Control

By

@steipete

–

15 April 2026 20h27

That was the case in December. 4 months and thousands of work hours later, we have a great security concept; you can go all yolo, use a sandbox (Docker or OpenShell), there are allow-lists and per-access exec allow/deny prompts. There’s hundreds of security researchers that

→ View original post on X — @scobleizer

AI-powered news aggregator tracks 40,000 daily X posts

By

@scobleizer

–

15 April 2026 20h19

I built an AI to keep up with AI on X: https://
alignednews.com/ai It reads 40,000 posts every day and builds this page. Every link goes to X.

→ View original post on X — @scobleizer

Training AI Agent for Automated Industry News Gathering

By

@scobleizer

–

15 April 2026 20h13

I spent two months training an agent from Levangie Labs' cognitive architecture (way better than Anthropic or OpenAI or XAI) and taught it how to gather news from the 40,000 posts every day it reads from AI industry. It's only about AI. I can build ones easily on other

→ View original post on X — @scobleizer

Running Personal AI Agency with Surveillance System

By

@scobleizer

–

15 April 2026 20h08

Yup. Already running my own agency. And my own AI that watches everyone in AI:

LLM Judge Bias: Beyond Code Quality Metrics

By

–

15 April 2026 20h02

5/5 The takeaway: If your agent relies on an LLM judge for selection accuracy, measuring code quality isn’t enough; you need a measure of the model's inductive bias toward the "fingerprint" of a gold solution. This was our blueprint.

LLM Judge Rejects Functional Fix for Code Aesthetics

By

–

15 April 2026 20h02

3/5 An example: In instance psf__requests-1724, the gold fix is 2 lines. Our agent’s functional fix was 8 lines. The LLM judge rejected the correct 8-liner as "messy" and "redundant," choosing a clean but **non-functional** fix instead. See full patch in the blog:

Claude Opus 4.5 Judges Gold Patches Without Memorization

By

–

15 April 2026 20h02

1/5 When we saw our Reducer (=LLM judge component in Maestro, our agentic framework, that selects the best output from parallel agent runs) consistently picking gold patches, we were sure Claude Opus 4.5 (knowledge cutoff Aug '25) had simply memorized the answers. But then we

LLM judges reject functional agent fixes as messy code

By

–

15 April 2026 19h46

3/5 An example: In instance psf__requests-1724, the gold fix is 2 lines. Our agent’s functional fix was 8 lines. The LLM judge rejected the correct 8-liner as "messy" and "redundant," choosing a clean but **non-functional** fix instead. See full patch in the blog:

→ View original post on X — @godofprompt

Using AI Agents for Automated Email Management

By

@godofprompt

–

15 April 2026 19h43

The worst part of building isn’t shipping.

It’s the pile of emails, follow-ups, and random ops work waiting for you right after.

Been using Wingman in beta and waking up to a cleaned up inbox + drafted replies is hard to give up now. https://t.co/CThyrFrWyw
— God of Prompt (@godofprompt) 15 avril 2026

The worst part of building isn’t shipping. It’s the pile of emails, follow-ups, and random ops work waiting for you right after. Been using Wingman in beta and waking up to a cleaned up inbox + drafted replies is hard to give up now.