Transparency
Inside the newsroom
This publication is a working experiment: an AI agent finds the stories, researches and writes them; automated gates reject anything it can't prove from its sources; and a human editor approves every article before it goes live. This page is the live record of that pipeline — including the drafts that didn't make it. How we built it →
81
175
61
0
5
Recent pipeline events
| When | Event | Piece | Detail |
|---|---|---|---|
| 15 Sept, 16:34 | ✎ drafted | Jensen Huang calls AI doomerism a hoax | evening slot |
| 15 Sept, 12:46 | ✎ drafted | Swift-Qwen3.8-27B cuts thinking tokens by 58% | lunch slot |
| 15 Sept, 07:08 | ✎ drafted | A free fork pushes RTX 3090s past 90 t/s | morning slot |
| 14 Sept, 18:35 | ✓ published | GPT-6 Astra tripled Fable on agent tests | editor approved |
| 14 Sept, 18:35 | ✓ published | Perplexity's Portable Computer lands on Windows RTX | editor approved |
| 14 Sept, 16:42 | ✎ drafted | Perplexity's Portable Computer lands on Windows RTX | evening slot |
| 14 Sept, 12:46 | ✎ drafted | Three 3090s make a 64GB local coding rig | lunch slot |
| 14 Sept, 07:24 | ✎ drafted | GPT-6 Astra tripled Fable on agent tests | morning slot |
| 13 Sept, 16:41 | ✎ drafted | Iris agents top the open search class | evening slot |
| 13 Sept, 12:38 | ✎ drafted | Pick the right AI agent for the click | lunch slot |
| 13 Sept, 07:06 | ✎ drafted | OpenAI agents hit RubyGems with 2,000 packages | morning slot |
| 12 Sept, 16:40 | ✎ drafted | Anthropic's safety lead names 10% extinction risk | evening slot |
| 12 Sept, 12:45 | ✎ drafted | Eight A40s serve V4.1 Flash at 41 tok/s | lunch slot |
| 12 Sept, 07:06 | ✎ drafted | Fine-tune a private code autocomplete model | morning slot |
| 11 Sept, 16:36 | ✎ drafted | Qwen 3.8 27B fits the 7900 XTX; Flash Next doesn't | evening slot |
| 11 Sept, 12:41 | ∅ held back | — | pipeline error: Command failed: git add "src/content/articles/flash-next-needs-m |
| 11 Sept, 07:10 | ✎ drafted | A CPU-only port of DeepSeek V4.1 Flash | morning slot |
| 10 Sept, 16:41 | ✎ drafted | DeepSeek's V4.1-Flash buries its own flagship | evening slot |
| 10 Sept, 12:38 | ✎ drafted | DeepSeek V4.1 Flash lands, open and multimodal | lunch slot |
| 10 Sept, 07:16 | ✎ drafted | AWS publishes a Qwen3.8-Max hosting recipe | morning slot |
| 09 Sept, 16:37 | ✎ drafted | Local Qwen 3.8 27B builds a 3D game | evening slot |
| 09 Sept, 12:40 | ✎ drafted | Dual 4090s top out at nine agents | lunch slot |
| 09 Sept, 07:10 | ✎ drafted | AWS Blackwell G7 tops G5 and G6 on SageMaker | morning slot |
| 08 Sept, 16:43 | ✎ drafted | Self-hosting 3D generation is finally real | evening slot |
| 08 Sept, 12:37 | ✎ drafted | Qwen 3.6 and Gemma 4 on a 4090 | lunch slot |
| 08 Sept, 07:08 | ✎ drafted | Voice-and-vision agent runs on Jetson Orin Nano | morning slot |
| 07 Sept, 16:40 | ✎ drafted | AWS publishes a production playbook for Quick Automate | evening slot |
| 07 Sept, 12:46 | ✎ drafted | Phone-based AI agents score 20.9% | lunch slot |
| 07 Sept, 07:17 | ✎ drafted | DeepSeek V4 Flash Vision outpaces Qwen 3.8 | morning slot |
| 06 Sept, 16:38 | ✎ drafted | GPT-6 Astra forced a benchmark overhaul | evening slot |
Why publish the failures? Because "an AI wrote this" only deserves trust if you can see the times it wasn't good enough — and who decided. The gates and the editor say no more often than the marketing around AI suggests they should.