AI, products, and the experience gathered along the way.
06 DRAFTS17 articles
NVIDIA Build gives a free key to DeepSeek V4.1 Flash, Kimi K3, GLM 5.3 and GLM 5.3 Flash: no card, about 1,000 free requests and up to 40 a minute. How to get the key, plug it into Claude Code through the free LiteLLM proxy, and where the limits of this trial access are.
15 min readShort slash commands that change how ChatGPT answers: /human writes like a person, /redteam tears an idea apart, /eli10 explains it like you're a kid. All 100 are here with descriptions, plus a cheat sheet: paste it once and the codes work in every answer.
21 min readOne prompt installs an editor skill in Claude Code. Then you give it links to a creator's reels, it breaks down their editing and turns it into a skill of its own, and edits your video in the same style: pauses, graphics on the words, captions, music and sound.
8 min readOn 23 September Claude Code cloud sessions left beta, and Anthropic gave subscribers a one-time credit: 100 dollars on Pro and 250 on Max. How to claim it by 7 October, start your first task and spend the credit well.
11 min readA free open-source router puts the free tiers of 34 providers behind one address, so Claude Code keeps working after your subscription limit. What is true in the viral video, how to set it all up in 15 minutes, and where the limits are.
10 min readRecurring quality issues, supplier dependency and the limits of safety commitments. Evidence, business implications and practical checks.
6 min readNine use cases across browsers, agent context, search, spreadsheets, interfaces and evaluation. Understand TypeSafe’s technology, examine the evidence and build your first example.
19 min readIn my video I turned into another person mid-spin, then put on a watch and a suit I do not own. Here are the tools, verbatim prompts and commands used, an agent brief, and a protection routine for family and team with a memo and a policy template.
33 min readSlicer Skill + MCP, CT-CHAT and Stanford Merlin: three projects for working with CT scans. Setup routes, hardware requirements, English prompts and lessons from our TotalSegmentator experiment.
11 min readA product model, a branded presentation, a sales report, and a website audit. Four English prompts, starter materials, and concrete checks for the results.
13 min readWhy an agent reads irrelevant files, selects the wrong skills, or keeps checking its work. Inspect the instructions, install two complete skills, and try a reproducible example.
12 min readWhy working well with AI starts long before your first prompt.
2 min readOn prototypes, small launches, and making ideas real.
1 min readHow to record your reasoning when there is no right answer yet.
1 min readWhat to keep after a failed launch, beyond the urge to forget it.
1 min readKnowledge becomes yours when you apply it to a problem of your own.
1 min readA personal journal as a place to make thinking clearer.
1 min readPractical articles carry a Practice label. Other excerpts preview the topics and design of future pieces.
New writing, useful discoveries, and honest lessons. The kind of letters I would want to read.
