Request, Review, Accept, Repeat: My New Life Inside the Loop
Coding with agents turned my job into one infinite loop of asking, reviewing and approving. I lost the fun part of solving problems. Here's what I got back.
Tutorials
My AI security scanner reported three vulnerabilities in code it never read
The run said success. The dashboard showed three findings. The report read like a careful security...
Slashing AI Agent LLM Bills by 75%: Sub-Millisecond Context Compression and Semantic Prompt Optimization in Python
Slashing AI Agent LLM Bills by 75%: Sub-Millisecond Context Compression and Semantic Prompt...
I gave my open-source project an AI crew that cannot touch GitHub
How I built Steward Crew on Switch: three AI agents in Discord and Slack, one plain-Python Ledger,...
Real-time AI dubbing in the browser: five timing bugs and how we measured them
I work on YTVT (Your Voice Translated), a Chrome and Edge extension that translates the speech in a...
Goodbye, ChatGPT - Why I Cancelled My ChatGPT Subscription
OpenAI has built its brand on productivity. ChatGPT is sold as a tool that helps you think, write,...
The Agent Made the Test Green by Weakening It
When an agent resolves a red test by weakening its assertion rather than fixing the behavior, the suite goes green while the bug ships. Treat the test file as off-limits, require t...
Home-made Evals Just Taste Better
Note This is a shortened version of the original writing on kig.re — my personal blog. Here is...
How do I run untrusted LLM-generated code in an isolated environment?
Run untrusted LLM-generated code in a hardware-isolated sandbox with the network locked to an...
Nothing was broken. Nothing was watching.
I built a pipeline that finds open PRs in OSS repos, runs an AI review on them, and drafts a cold...
La allowlist que igual deja salir el .env
Dos fallas en mi firewall para agentes: lo que la allowlist permite termina siendo la salida del secreto, y el agente puede reescribir la política que lo juzga.
The right answer was the odd one out 67 % of the time
I build practice exams for Dutch certification exams: the VCA safety certificate, the hospitality...
My Claude Code router cost more than Opus. Here's what the benchmark taught me
I built AutoMode because most of my Claude Code sessions look the same. I ask a quick question, then...
Claude Code Ignoring CLAUDE.md? See What's in Its Context
When Claude Code ignores a rule, the usual move is to reword it. Start with the verb, use backticks...
I let an AI agent read one document. It tried to make a payment. Try it yourself in 60 seconds.
AI agents are getting real tools: email, payments, code repos, customer data. At the same time, they...
I Built an AI Software Engineering Team Inside Telegram. They Have Jobs, Responsibilities, and Apparently Trust Issues
Codex is the manager. Cursor writes the code. OpenCode tries to destroy it. And I get to decide...
The Carmack Pipeline: How DOOM's 1993 Hacks Predict Modern LLM Architecture
"Computers are incredibly fast, accurate, and stupid. Human beings are incredibly slow, inaccurate,...
Ubuntu 24.04 Restricts User Namespaces. That's Why Your Agent's Sandbox Won't Start
Ubuntu 24.04 denies capabilities in unprivileged user namespaces, so bubblewrap can't build Claude Code's sandbox. Reproduce it, then pick a fix.
Two companies, one outage, zero shared logs: building Airlock on Switch
#HowISwitch · Flip the Switch hackathon Here's an incident every Nigerian developer who has touched...
We Stopped Installing LLM SDKs in Every Microservice
How we built QuickStart AI so model calls and paid web APIs go through two internal services, with one policy for keys, retries, streams, and cost.
Log Doctor: Letting AI Take the First Look at My Kubernetes Errors Using AWS Bedrock and AWS Opensearch
Description: "Every 15 minutes, a small CronJob reads the error logs, asks an LLM on Amazon Bedrock...
Run Budget: Every Agent Run Needs a Spend Limit
An agent that retries forever can burn a month of API spend overnight. Give every run a budget for tokens, tool calls, time, and dollars, and make it stop cleanly with a report whe...
Title: The Memory That Knows When It Is Wrong
Description: Perfect recall is dangerous without discrimination. How one agent stores the signal that...
Land the Interface First: Split One Feature Across 3 Agents Without Conflicts
Three agents, one feature, three made-up Refund types. Land the TypeScript contract and tests first, then branch, then dry-run the merge with merge-tree.