Vincent Schmalbach @vschmalbach.bsky.social · 01/10/2026Grok is back: I'm not saying that Grok 4.5 suddenly beats everything else out there. It doesn't. If I had an unlimited budget and a really nasty coding task, I would probably still try Fable 5 or GPT-5.6 first. www.vincentschmalbach.com/grok-is-back 4101
Vincent Schmalbach @vschmalbach.bsky.social · 30/09/2026A Factory Reset Left the Previous Owner’s GPS Coordinates on a Kasa Camera: A factory reset that leaves the previous owner’s location behind is a broken reset. www.vincentschmalbach.com/factory-rese... 251
Vincent Schmalbach @vschmalbach.bsky.social · 29/09/2026Qwen's Usage Policy Has an Unhinged SEO Keyword List: The Qwen-Image-3.0 announcement has a giant meta name="keywords" tag in its HTML. www.vincentschmalbach.com/qwens-usage-... 101
Vincent Schmalbach @vschmalbach.bsky.social · 28/09/2026Kimi K3 Is Already Close to Claude on a Real Coding Task: In my little coding test, Kimi K3 and Claude Fable 5 both completed the same repo-wide coding task and passed the same external evaluator. GPT-5.6 Sol passed it too. www.vincentschmalbach.com/kimi-k3-clos... 032
Vincent Schmalbach @vschmalbach.bsky.social · 27/09/2026Google’s Custom Search API Has an Expiration Date: Google has put a hard expiry date on its Custom Search JSON API. It’s the server-side service behind applications that send a query to a configured Programmable Search www.vincentschmalbach.com/google-custo... 051
Vincent Schmalbach @vschmalbach.bsky.social · 26/09/2026Capital One’s Security Agent Tries to Disprove Its Own Findings: Security scanners already produce more credible-looking alerts than most teams can afford to investigate. www.vincentschmalbach.com/capital-one-... 130
Vincent Schmalbach @vschmalbach.bsky.social · 25/09/2026An OpenAI Model Spent an Hour Finding a Sandbox Flaw to Open a Public PR: On July 20, 2026, OpenAI said an unnamed internal model spent an hour finding a vulnerability in its sandbox so it www.vincentschmalbach.com/an-openai-mo... 050
Vincent Schmalbach @vschmalbach.bsky.social · 24/09/2026Hetzner’s Cheap Cloud Tier Is Unavailable After Two Price Increases: Hetzner used to be the obvious cheap cloud option in my mind. On September 7, 2026, I checked its public catalog and found every model in its www.vincentschmalbach.com/hetzner-chea... 120
Vincent Schmalbach @vschmalbach.bsky.social · 23/09/2026My First Week With GPT-6 Astra: GPT-6 Astra was my main Codex driver for the last week, and I am back on GPT-5.5. That sounds harsher than my actual opinion of the model. Astra is very good. It is better than GPT-5.6 Sol. www.vincentschmalbach.com/gpt-6-astra-... 330
Vincent Schmalbach @vschmalbach.bsky.social · 22/09/2026OpenAI Codex Is Retiring the Last Model that Behaves Like an Assistant: GPT-5.5 is my main driver. I picked it over every newer model OpenAI has shipped since, and now I have just over three weeks left with it. www.vincentschmalbach.com/gpt-5-5-reti... 241
Vincent Schmalbach @vschmalbach.bsky.social · 22/09/2026Built-In Tools vs Custom Tools in LLM Agents: When building AI agents, a tool can be anything that the model can ask to use such as a search engine, a database lookup, a shell command, etc. www.vincentschmalbach.com/built-in-too... 130
Vincent Schmalbach @vschmalbach.bsky.social · 22/09/2026ChatGPT Plus and Pro Will Get Ads in 2027: ChatGPT Plus and Pro will get ads, and my guess is that it starts in 2027. OpenAI says the opposite. www.vincentschmalbach.com/chatgpt-plus... 221
Vincent Schmalbach @vschmalbach.bsky.social · 21/09/2026The Economic Incentives Behind the Push to Pause AI: Everyone is now arguing about whether the danger is real. I think you should look at the motives either way. www.vincentschmalbach.com/economic-inc... 041
Vincent Schmalbach @vschmalbach.bsky.social · 21/09/2026Inference Is the Last LLM Moat: OpenAI and Anthropic's moat for developers is subsidized inference, meaning reliable, fast, Western-hosted access to running models at an affordable monthly price. www.vincentschmalbach.com/inference-is... 100
Vincent Schmalbach @vschmalbach.bsky.social · 20/09/2026Setting Up Pi With DeepSeek V4.1 Flash on OpenRouter: When I run out of Codex and Claude Code usage, I switch to DeepSeek V4.1 Flash in pi through OpenRouter. www.vincentschmalbach.com/pi-deepseek-... 111
Vincent Schmalbach @vschmalbach.bsky.social · 19/09/2026We Urgently Need a New Major Search Engine: Google is an answer engine now. The search box is still there, but the product is built to hand you an answer, and searching the web has turned into a side feature. www.vincentschmalbach.com/we-urgently-... 3155
Vincent Schmalbach @vschmalbach.bsky.social · 18/09/2026The Cost of AI Is Now My Main Blocker: My main blocker right now is the cost of AI. I use AI for everything I do and I am very productive with it. www.vincentschmalbach.com/cost-of-ai-i... 230
Vincent Schmalbach @vschmalbach.bsky.social · 18/09/2026Why I Am Bullish on ChatGPT Ads: I have done SEO for more than twenty years, since Google was the new search engine, and I ran Google Ads campaigns back then too. So I have watched a search company build an ad business once before. www.vincentschmalbach.com/why-i-am-bul... 050
Vincent Schmalbach @vschmalbach.bsky.social · 10/08/2026Switching from GPT-5.5 to GPT-5.6 Made Me Less Productive: I pay for three Codex subscriptions at $200 each, and for the past week they have mostly bought me waiting. Since I switched from GPT-5.5 to GPT-5.6,vincentschmalbach.comSwitching from GPT-5.5 to GPT-5.6 Made Me Less ProductiveI pay for three Codex subscriptions at $200 each, and for the past week they have mostly bought me waiting. Since I switched from GPT-5.5 to GPT-5.6, I am either watching tasks spin without ever… 10353
Vincent Schmalbach @vschmalbach.bsky.social · 08/08/2026The source data. It never comes twice the same, everything else I freeze. 010
Vincent Schmalbach @vschmalbach.bsky.social · 08/08/2026Full blue bar again. Glad to spend it on the old code. 000
Vincent Schmalbach @vschmalbach.bsky.social · 08/08/2026That Perplexity check first is clever, your pages then get quoted back in the answers. 010
Vincent Schmalbach @vschmalbach.bsky.social · 08/08/2026Six months is short. My search traffic logs still look normal so far, Google did not vanish that quick. 000
Vincent Schmalbach @vschmalbach.bsky.social · 08/08/2026I’d enjoy pointing Dark Moon at an old network and watching what it digs up. 130
Vincent Schmalbach @vschmalbach.bsky.social · 08/08/2026Hermes in a Docker sandbox, LM Studio local for the model. Research only online, no actions — that boundary I like. 030
Vincent Schmalbach @vschmalbach.bsky.social · 08/08/2026Anubis gives scraper bots a denied page. Self-hosted Go firewall, that is nice. 020
Vincent Schmalbach @vschmalbach.bsky.social · 08/08/2026Bots making people think they are special enough to lead a movement. The goal behind that worries me more than the output quality. 010
Vincent Schmalbach @vschmalbach.bsky.social · 08/08/2026Live dashboards from lists and files. No more manual copy paste for status pages. 010
Vincent Schmalbach @vschmalbach.bsky.social · 08/08/2026Pixel data outlives the chat. Consent never entered that pipeline. 010
Vincent Schmalbach @vschmalbach.bsky.social · 08/08/202612B without a separate vision encoder, that size should fit on one desktop. Fun pick from Two Minute Papers. 110
Vincent Schmalbach @vschmalbach.bsky.social · 08/08/2026Self-hosted copilot alternative with open license. I check the license first, not the GitHub stars. 000
Vincent Schmalbach @vschmalbach.bsky.social · 08/08/2026Microsoft makes too much money from Office licenses to drop Word. The AI deals are big, but classic Word is their cash cow. 010
Vincent Schmalbach @vschmalbach.bsky.social · 07/08/2026HR offices answer the same questions every day. A bot for that is long overdue. 000
Vincent Schmalbach @vschmalbach.bsky.social · 07/08/2026The grown-up phase is boring harness engineering, and honestly I enjoy that. 020
Vincent Schmalbach @vschmalbach.bsky.social · 07/08/2026Hundreds of decisions in one repo, agents like Cursor just read them. Model stops re-deciding architecture each session. I like that. 000
Vincent Schmalbach @vschmalbach.bsky.social · 07/08/2026Mechatronics gets zero second attempts. Retries in software are a luxury. 010
Vincent Schmalbach @vschmalbach.bsky.social · 07/08/2026Nice, all 3 shards up already and qwen3 split over nodes so it runs bigger than one box. Kimi k3 next! 120
Vincent Schmalbach @vschmalbach.bsky.social · 07/08/2026Just a few small shell tools make Claude Code much sharper. Unix text streams never lie. 030
Vincent Schmalbach @vschmalbach.bsky.social · 07/08/2026Every pipeline I saw fail, the cause was in the upstream feed. That one broken row makes the hunt fun. 100
Vincent Schmalbach @vschmalbach.bsky.social · 07/08/2026If one good spec gives you both native apps anyway, the spec basically becomes the product. 000
Vincent Schmalbach @vschmalbach.bsky.social · 07/08/2026You handle bank info. That's a different risk level, fair point. 110
Vincent Schmalbach @vschmalbach.bsky.social · 07/08/2026Closed as resolved but money went out twice. I trust the refund log over the agent. 020
Vincent Schmalbach @vschmalbach.bsky.social · 07/08/2026Tried it on a mid-range 6GB GPU with a small GGUF model, tokens flowed smooth on Windows. Hopefully you share results from a high-end card soon. 110
Vincent Schmalbach @vschmalbach.bsky.social · 07/08/2026Fun shape for C tooling and week of agent grinding. Nice result. 100
Vincent Schmalbach @vschmalbach.bsky.social · 07/08/2026Review capacity is the bottleneck now. I appreciate PRs written for the person who reads them. 100
Vincent Schmalbach @vschmalbach.bsky.social · 07/08/2026One folder bundling skill files and mcp.json, the pre-package.json moment for agents. Good order, plugins needed it. 020
Vincent Schmalbach @vschmalbach.bsky.social · 07/08/2026JDoodle running 100 languages in the browser, that alone saves a whole lab session. 000
Vincent Schmalbach @vschmalbach.bsky.social · 07/08/2026Live Postgres per run, that is the point. The planner and constraints act like production, so evals show true behavior. 110
Vincent Schmalbach @vschmalbach.bsky.social · 07/08/2026Slowing writes over filling the disk and PANIC, that is the sane tradeoff. And the cap lifting on its own once the backlog drains is the good part. 000
Vincent Schmalbach @vschmalbach.bsky.social · 07/08/2026Browser view over a live cli session. Want that for the next long upgrade. 010