--- title: Things I Learned - 19 Jul 2026 date: 2026-07-19T00:00:00+00:00 categories: - til description: I learned to identify human tasks as depreciating assets if AI can easily verify them. The value of creating AI agent skills is immense, potentially leading to a 'Wikipedia of skills' for adapting to this AI-first world. tags: [personal-update, learning, ai-agents, future-of-work, model-comparison, prompt-engineering, chatgpt, gemini] --- This week, I learned: - Writing is slightly, but only slightly, better than typing (for adult learning.) One factor is that typing is faster, so many people take notes verbatim, summarizing and thinking less. [ChatGPT](https://chatgpt.com/share/6a5c286f-1f08-83ee-9c6a-870c0fb53c91) + [Claude](https://claude.ai/share/1663cb57-5453-42b5-ae4f-099384c76946) - Graphology for personality is pseudoscience. [ChatGPT](https://chatgpt.com/share/6a5c23e8-aa3c-83e8-ac67-32382a87a789) + [Claude](https://claude.ai/share/45d91f44-be1f-43d0-afda-5814f2e7942e) - When I decide to spend time, or someone says "Let's do X", it's worth checking: is this something AI can easily try, and is it clear to verify? If so, reinforcement learning loops could make AI good at it, making it a depreciating asset. - Studying how to live in an AI world is _exhausting_. (Not as bad as my MBA days, but not as easy as my data scientist days, either.) It requires me to make a larger mental shift, i.e. change my perspective, than I have since 2000, and that feels like work. - Both `nl FILE` and `cat -n FILE` add line numbers to files, but `nl` skips blank lines by default, `cat` doesn't. - After using [`rtk`](https://github.com/rtk-ai/rtk) for 2 months, I'm _slightly_ downgrading it. It saves tokens but agents mess up shell commands when using it. It's still probably a net saving, so I've changed my `AGENTS.md` from "Always prefix with `rtk`" to "Prefix supported, high-output commands with `rtk`... skip for bash builtins, pipes, loops, etc." - I find 🔴🟡🟢 convenient status indicators in my notes. Similar ones are: 🟥🟨🟩, ❤️💛💚, 📕📙📗. I'm not fully convinced by: 😄😐😞, █ ▒ ░, ↑ → ↓, ▁▂▃▄▅▆▇, ■ ⬔ □, ● ◐ ○, ⚫ ⚪ 🔘, 🌕 🌗 🌑, etc. though they might have their uses. - Model updates means a SKILL.md and a plugin review / update, e.g. [with GPT 5.6 Sol](https://x.com/keyanzhang/status/2076461227661054015). So, like with any open source repo, use from people who update it regularly and benchmark it and version control it by model. - I asked Gemini 3.5 Flash thinking: "Which of our employees have worked on Microsoft PowerApps? Search @Google Drive and @Gmail". It found one employee and a referral in under a minute. I asked ChatGPT with GPT 5.6 Sol with [gws](https://github.com/googleworkspace/cli) access. It found 3 more, plus 5 possibilities, in 12 minutes. Truly a [rottweiler](https://x.com/sama/status/2075047747250684352). - [Parallel Search Turbo](https://parallel.ai/blog/parallel-search-turbo) seems like a pretty good search API, especially for agents. Low price, high speed, and maybe good quality. #ForNow [ChatGPT](https://chatgpt.com/share/6a58b12c-b5f4-83e8-a325-7349266577c5) - Group chats in ChatGPT will [probably get deprecated](https://x.com/i/status/2076061674306687161) #ForNow. - [What I learned](https://claude.ai/share/1f12542b-6eb3-4b3c-a272-90c82d326b00) from benchmarking my [Ideation Protocol](https://github.com/sanand0/research/tree/main/ideation-protocol-optimization) skill extensively: 1. Once you know the rubric, models can easily create a good prompt to optimize for a known rubric #ForNow. So rubric design matters more. 2. ⭐ Rubric design is really knowing what you want/need. To do this, iterating on output matters. 3. Position bias is real #ForNow. Always check if an (P, Q) comparison matches a (Q, P) comparison. 4. Models are still biased towards longer content, and potentially towards their own output #ForNow. 5. **How to optimize a prompt or skill**: Research and figure out what you _really_ want, first. Then, ask a smart model for a prompt that optimizes for it. Benchmark only if you'll use it a lot - it's still a lot of work, and meta-prompting does a good job #ForNow. [gbrain skillopt](https://github.com/garrytan/gbrain/blob/master/docs/guides/skillopt.md) might be premature optimization. - [You can use GPT 5.6 Sol in Claude Code](https://x.com/i/status/2076119366647894371) #ForNow. (But what's the point? Harnesses seem to be working better with their own models #ForNow.) - Our clients keep saying "We need to build a data lake" or "We need an enterprise data strategy." I keep telling them, "No, agents can do it for you." What I missed is: _technology_ is the smaller part of the problem. Finding who has what data, getting access to it, and sorting out permissions ("governance") is the bigger part. - Giving agents expert **task-specific, testable procedures** seems better than expert **roles** or **mental models** #ForNow. But benchmark in any case. [ChatGPT](https://chatgpt.com/share/6a563182-4860-83e8-974e-90ec5c8ec2ac) - Python 3.3 introduced `str.casefold()`. It performs more comprehensive Unicode caseless matching than `lower()`; `'Straẞe'.casefold()` becomes `'strasse'`. (🟢 Unicode case-folding is standardized.) `contextlib.closing(x)` calls `x.close()` when its context exits. (⚪) In a dataclass, use `x: list = dataclasses.field(default_factory=list)`, not a mutable literal default. (⚪) I learnt these while reviewing Codex-generated Python—illustrating, rather than proving, that reviewing AI-generated code can teach and catch errors. (🟡 Review remains useful across tooling. Review 2029.) “Do not discriminate against intelligence—artificial or otherwise” is a rhetorical value judgment, not an empirical conclusion. (⚫ Rhetorical value judgment, not testable. Review now.) - Here's a nice idea from ChatGPT. "When itching to correct or clarify, FIRST restate their position to their satisfaction. 'Did I get you right, fully?'" This emerged from the prompt suffix: Based on your research, and my past conversations, what are the top areas where and how (specifically) I can apply this principle on myself and others to maximize impact? - Automated evals can catch stuff humans miss. And vice versa. And given how many evals we create, we need automated evals to be written in an easy-to-review way. [Do Automated Evals Work?](https://parlance-labs.com/blog/posts/auto-evals/) - The [BINEVAL](https://arxiv.org/abs/2606.27226) paper reiterates that a bunch of Yes/No binary questions beats scales or ratings for many benchmarks. You know exactly how to grade and WHY you got a certain score. This is more reproducible and easier to learn from / act on. - When asked "How long will this software take?" models typically provide estimates assuming human speed #ForNow. Maybe they haven't been trained enough on agentic timelines. So, when my colleague got a 2-4 week estimate which he was able to solve in hours, it was a surprise. (But, of course, it's best to verify before promising speed.) - SKILL.md dramatically lowers the cost of learning a skill (since you don't learn it - the agent does). That means that the value of creating skills is much higher - hundreds can use what you create (giving you recognition, if not money). I think I've underestimated the number of skills people will have available (I thought dozens - but it may be thousands #ForNow) and the number of skills people will create (I thought tens of thousands - but it may be millions #ForNow.) A Wikipedia (community curated, verified, high quality catalog) of skills might emerge #ForNow, if it hasn't already. - Tacit knowledge is often just un-measured knowledge. Once I put a sensor on the bellboy's hands at The Curzon Court, AI can figure out how he opens the door with the key and why I can't do the same. The subset of tacit knowledge that's AI-resistant is where attempts are expensive ("How to negotiate a merger" rather than "How to open a door") and feedback is slow/vague ("Does the client trust me" rather than "Did the door open"). - The fact that Composio has ~20,000 tools is a market signal that connectors are commoditizing, and are a depreciating asset #ForNow. - A weak model needs a forgiving harness - which ends up slowing down model learning. Stricter, accurate verification environments are better for fastest model learning. - [ChatGPT Work](https://learn.chatgpt.com/docs/get-started-with-work) lets you run for longer, faster, install plugins and skills, host a website, etc #ForNow. It's somewhere between Chat and Codex. It consumes Codex limits - something to watch for (since chat limits are quite generous). - Codex temporarily removed the 5-hour usage limit. [Tibo](https://x.com/thsottiaux/status/2076365965915467978). So, since I have 3 banked rate-limit resets #ForNow, I can, in theory, use 4 full weeks of Codex usage at one go. Reality: I don't have problems large enough for a SINGLE week's consumption! - From what I see of the [State of AI Design](https://stateofaidesign.com/chapters/tools) and [State of Prototyping](https://survey.uxtools.co/spring-2026), Figma is _way_ ahead of competition #ForNow, e.g. Adobe, with [Figma Make](https://www.figma.com/make/) and [Weave](https://weave.figma.com/). I was also surprised how popular Cursor is (#2 behind Claude Code #ForNow). It's also interesting that designers are coding directly #ForNow, using Figma just for edits / steering. But many _research_ tools (note takers, survey analysis/research, etc.) will likely get eaten up by AI coding agents #ForNow, given how much designers are building their own tools.