The boring math behind a 150-tool AI suite: how I pick models by quality, latency, cost and capability
Dev.to Article #2 Draft（2026-11-09） Title: The boring math behind a 150-tool AI suite: how...
Tutorials
Your Prometheus Is Hoarding. Time for a Funeral Policy.
Every dead metric is a bill you’re still paying. I know, you’re tired. Tired of dashboards that...
29% of Your Cloud Bill Is Waste. Start With the Graveyard.
You’re paying rent on servers everyone forgot. Cloud waste just did something it hasn’t done in...
Why Your Airflow Tasks Fail at 3 AM (And Why Re-Running Them Is a Trap)
One outage, four suspects — interrogate in order. Pull up a chair, grab a cup of chai, and let us...
Your LLM Is Not a Security Boundary: Auditing Agent Tools for eval() and Path Sinks
In May 2026, Microsoft's security team published two remote-code-execution issues in its own agent...
Your Prettiest Dashboard Is Lying to You (And Grafana Will Never Tell You)
Pull up a chair, grab a cup of chai, and let’s talk about the prettiest graph on your big screen....
Your Agent Reads What Attackers Write
Everything the agent reads is text. Some of that text was written by an attacker. The on-call...
Thirty seconds in one job: what Seedance 2.5's controls actually do
Answer: Seedance 2.5 renders clips from 1 to 30 seconds at 480P, 720P and 1080P, billed per output...
What gets refused on Seedance, and how the two review layers differ
Answer: Seedance reviews a job twice. Layer 1 screens the prompt before any task exists — a clear...
What unshare -n leaves behind: building a boundary for an AI recon agent
Why the boundary has to hold My AI agent's sandbox passed all 13 of its isolation tests on...
Seven models, five runs, 18 books: a test of AI recommendations
A list of five plausible books can look like a successful recommendation. Repeating the request tells...
AI Audit Evidence, Part 6: A Record Nobody Read Is Only Stored
A record sat in the database for a year before anyone opened it. That is not negligence. There is no...
I gave my social media scheduler an MCP server. Here's what I learned building it.
Social media scheduling is the kind of task AI agents should be perfect at. It's structured (a...
Same model, same chip, 27 apart: the runtime path is the performance
YOLOv11-N on a Qualcomm Dragonwing RB3 Gen 2 (QCS6490), measured on Qualcomm AI Hub's real-device...
Your AI agent's token bill is mostly input. Here is the field guide to cutting it.
Your AI agent's token bill is mostly input. Here is the field guide to cutting...
Whisper on the RK3588 NPU: the encoder offload works — everything around it is the real lesson
Status: encoder offload merged into a working whisper.cpp path on an RK3588 board, transcription...
2비트라는 이름의 GGUF가 3비트 이름의 파일보다 컸습니다: 공개 Darwin-180B 양자화 파일 6개를 헤더까지 열어 보니
TQ2_0 이름의 파일은 92.0GB, Q3_K_S는 90.0GB였습니다. 파라미터의 28.9%가 조회용 표이기 때문입니다. GGUF 헤더를 읽어 파일 구성을 계산했습니다.
Training MoE Models Is Complicated. Hugging Face Just Wrote the Manual.
I've spent a fair amount of time recently experimenting with Mixture of Experts (MoE) models on my...
PrivaSearch: private AI search with SearXNG and a local LLM
I wanted AI-synthesized search answers without sending my queries to anyone. PrivaSearch is the...
Making browser perception cheap for LLM agents: snapshots, caching and diffs
Most "give the model a browser" setups hand it one of two things: a screenshot, or a dump of the full...
We are all excited about bringing new agents online. I am thinking about retiring one, how do you go about it?
Nobody Writes About Retiring an Agent. Here's the Checklist. ...
Nobody Writes About Retiring an Agent. Here's the Checklist.
Nobody Writes About Retiring an Agent. Here's the Checklist. We have onboarding docs for...
Frontier LLM prices sat flat all October. Then two cheaper flagships took seats, and one doubles on December 1
The LLM Price Index didn't move once in October. Same ten flagships, same ten list prices, from...
Production Problems in RAG and How to Solve Them
Indexing a document is easy. Keeping millions of vectors in sync with documents that keep changing is the real job. Nine production failures, each replayed live in the browser with...