S Anand

Since agents do most of the work, specification (telling it what to do) and verification (making sure it did it right) is more important than execution. This is what I’m often asked about. Of the ~600 questions people asked me this year, half were deciding what to do / ask for and every sixth question was about checking if it worked. I checked. For example: “What should we ask agents to do when we only have a dataset and no clear use case?” ...

This week, I learned: Enterprise software migration seems a massive medium term opportunity - Ethan Mollick. On the other hand, liquid software where anyone can tweak their own softare is emerging. “But I’m not working for my agent, it works for me. It’s a computer. It’s not a person I have to keep busy and keep managing (and checking it’s managing it’s own agents).” - Ben Tossell OpenAI announced that ChatGPT Chat (except Instant) will start getting metered from December 2026. This is a big deal for me because: I prefer ChatGPT Chat / Work to Codex because it can access my past conversations. Chat (in my ChatGPT $20 Plus) isn’t metered visibly and feels unlimited. By giving ChatGPT a plugin to control my computer, I get most of the benefits of Codex. But this free use will last only a month. I need to see how well GPT 6.1 Terra fills that gap - i.e. is it good enough to beat GPT 6 Sol and cheap enough that I can run my chats without worrying about exhausting limits? Clairic is an extraction model from Cloudflare. It’s a Decision model (or System One model) like Jev: fast and lightweight. But this focuses on extraction of text, numbers, dates, even forecasts. It provides the confidence as well as input position of the text for reference. Anthropic recommends using the main agent rather than subagents. Single agents on smart models work well unless you need independence #ForNow. Ruby chocolate is the fourth variety of chocolate - alongside dark, milk, and white. (Might be a marketing claim than verified.) I tasted it. It’s fruity, slightly sour, not bad. My order of preference among the melted chocolates I tasted in a spoon at Chocolate Nation, Antwerp: Gold. Caramel toffee butter cream . Delightful! White W2. White chocolate. Something I usually enjoy. I was surprised I liked Gold more. 823. Milk cocoa caramel chocolate. Slightly darker. Power 41. 41% cocoa Caramel with less sugar. Ruby. Fruity sour taste. Arriba. Strong cocoa flavor. 811. Bitter. The rest were too bitter I had no idea there was so much innovation in chocolates. Most of the ones I like are a century old. Good to see some revival in the last decade. ChatGPT 1880s: Truffles. chocolate that melts immediately in the mouth. 1907: Hershey’s Kisses. 1908: Toblerone. 1912: Pralines. chocolate shell with creamy filling. 1920: Chocolate-coated ice cream (like Magnum). 1923: Filled chocolate bars. 1923: Nougat-caramel chocolate bars (like Five Star, Mars). 1923: White chocolate. 1926: Chocolate syrup (Hershey’s). 1928: Peanut butter cups : Combined sweet chocolate with salty, savory peanut butter for a striking flavor and texture contrast. 1932: Chocolate oranges. 1935: Aerated chocolate with tiny air bubbles. 1935: Chocolate-coated wafers (KitKat). 1936: Malted chocolate balls (Maltesers) 1949: Soft-melting chocolate centers (Lindor) 1951: Chocolate-hazelnut spread (eventually Nutella) 1969: Spherical filled chocolates (Lindor) 1971: Chocolate eggs (Cadbury Creme Egg) 2012: Blond chocolate: white chocolate with biscuit and toffee notes. 2017: Ruby chocolate: pink, fruity, sour. 2021: Dubai chocolate: Pistachio cream, cruncy. Looks like GPT 6.1 Terra is mostly as good as Sol but cheaper. I haven’t tested it yet. On one hand, modern CSS has several new features. On the other hand, I’ve stopped writing code and delegated it to agents, so this is of no use to me. Instead, what I should be thinking about is: what new patterns does this enable, and what should I be asking for in prompts? Ethan Mollick: “Ask yourself: “how do I make what I am doing even more ambitious?” Something to practice actively, given the growing abilities of agents. Anthropic is comfortable that no single person has an architectural view of Claude Code. I guess architecture is also being taken over by agents. Addy Osmani: “No member of the team has a full mental model of Claude Code in their head, nor do we think anyone needs to. Rather, when anyone wants to learn about a piece of the codebase, they ask Claude to generate an explainer about that exact slice. It only takes a few minutes and the explainer is always up to date.” Good advice from Thornton Bell Humans write for humans. Agents for agents. Be more ambitious. Give it more eyes - logs, observability, … Code on remote machines. Ask for 5 variations. (I’m not doing enough of this yet.) Use schedules and triggers. (I’m not doing this at all, so far.) Codex introduced an experimental #ForNow --enable instant_interrupt flag that looks promising. It’s still steering, i.e. not instantly stopping any existing requests or tasks (hence wasting effort), but doesn’t wait for the entire response either. Not sure how it works but I just enabled it and will evaluate it. Code-mode yielding PR, model-response preemption PR. In Codex you can @-mention another Codex session. That’ll give it access to an earlier session’s context. In the past, I copy-pasted session transcripts, saved them in files, or pointed Codex at the session directory. This is more convenient. PixelBoard has pretty impressive cvoice typing. Google has a Gemini-powered “Rambler” voice recognition system (based on Gemini 3.5 Translate, with offline fallback) on Android phones, but not available via GBoard. PixelBoard enables that. ChatGPT Mistakes I made Week ending 01 Nov 2026 ...

My TDS course asked students to draw objects in one minute. The drawing is sent to their classmates as well as three vision models. They score if: Classmates guess it right AI guesses it wrong In other words, can you, in 1 minute, draw something that fools AI but not humans? Here are some of the best results, when I ran it against GPT 6 Luna, GLM 5.3 Flash, and Deepseek v4.1 Flash, which are really cheap and good vision models. ...

I conducted a workshop last weekend for working professionals and had assigned homework as part of that. We reviewed their homework at the start of the class, and I noticed that the majority of them had clear AI smells. I let the participants know that using AI is perfectly fine, especially in an AI class, and asked how many of them used AI to write the final output, All, except one, said “Yes” - a bit of a surprise to me since it included some output that I thought was manually written. ...

Thornston Bell’s How I work with Agents suggests actively using remote machines for coding. I’d struggled with this for many months, and finally managed to set up my Windows laptop for remote coding. How? Tell my wife to tell the maid to never switch it off. (It doesn’t have battery.) Tell ChatGPT to install Remote Desktop Commander - and schedule it to run on boot. Now agents can use it. Tell ChatGPT to install Tailscale so I can SSH into it and copy files. Tell ChatGPT to create a sync-plutonium.sh that pushes files from my Linux system (graphene) to the Windows machine (Plutonium). That last step was the critical enabler. This means that most of the files on my regular laptop is also on the home laptop, and it has the same context. ...

I think it’s a miracle to see the sheer number of technologies that can come together with AI agents orchestrating them. I’m right now on a flight from Brussels to Dubai, and the Wi-Fi works. So, I’m dictating this blog post via PixelBoard, an AI-coded port of Gboard that lets me use Gemini for voice typing on Android (which is quite good!). I’m also browsing my Asterix comic book collection. These are on my Windows laptop in Singapore, which I’ve connected via Tailscale. I have no idea how to configure Tailscale, and ChatGPT did that for me. BTW, Tailscale is one of those services that I didn’t see a use for until I used it, and now will never turn it off. It creates a VPN between my phone, traveling laptop and home laptop, so I can browse files, SSH into each other, run HTTP servers, etc. All securely. ...

The Innovation team has built a few more demos since the last update. Three interesting learnings: We can speak in one language and be heard in another, live, in a normal Google Meet call. A person in another city can use their hands and a phone to sculpt a 3D object on your laptop. AI can search what was heard, not just said, and can check scientific claims by actually reproducing them. New capabilities Translate a live meeting into almost any language (Hritik). At Friday’s VIA talk, I spoke and the audience heard Filipino, Hindi, Tamil and Telugu with a short delay. Hritik built a desktop app that creates a virtual microphone and speaker for Meet. It translates both ways: what others hear from you, and what you hear from them. It took him a day or two to vibe code, and comes as a Windows, Mac or Linux app. I estimated about 0.6 US cents a minute in my trial. ...

I’ve been playing around with mobile sensors recently and one such experiment is the voice choir. Here’s how it works. I’m the presenter. I visit voice choir, create a room, and show the QR code. People see the QR code (via screen-sharing or projector) and open on their mobile phones to join my room. I press “Start”. They hear a tone and they should hum at the same pitch. They can see if their pitch is above or below and adjust. I get a data stream of their pitch. Over time, we see if we can hum at the same pitch, or how close we converge. I record this as a visualization. We played this in Straive across about 20 people who joined the room. Here’s the video: ...

This week, I learned: Neural Networks Are Already Clean argues that in modern LLMs, most neurons have one function, not arbitrary overlapping functions (like we thought earlier). This becomes clearer when you see the learning steps, e.g. the “iy” neuron -> “diy” -> “do it yourself”. Without the “diy” in the middle, it’s confusing what “iy” has to do with “do it yourself”. Models probably learn by generalizing concepts repeatedly. Interpreting models has the bitter lesson, too: just tell a powerful agent to interpret it. Modern LLMs have switched from ReLU to SwiGLU. Why? ReLU is on-off and communicates no information when off. SwiGLU is a smooth function that whispers less relevant information - and that gives smarter models more context. Gemini 3.8 Flash is SLIGHTLY cheaper than Gemini 3 Flash Preview when transcribing 1 min of audio - #ForNow until Dec 31. After that, a higher pricing kicks in. ChatGPT ArtCraft extended their open source tools to cover MS Office, Adobe, and Autodesk #ForNow. Others are following suite. The Remade Wiki has includes OneNote, QuickBooks, Power BI, Outlook, MS Project, Visio, Discord, Jira, Linear, Monday, Corel, … #ForNow Cloudflare ported the TypeScript compiler to Rust. Increasingly, language choice is becoming fluid. Tiny Little Office uses a number of interesting tricks: The entire document is in the URL fragment, which is as much as 2 MB for Chrome! That means the entire document can be shared as a single QR code (if it’s not too large). Docs, spreadsheets, presentations, diagrams are all Markdown-based #ForNow and agent-friendly. They can be embedded as code fences inside each other. It uses domain-specific compression tailored to its language. It also uses ~ instead of - for base64 - to avoid wrapping issues with mail clients. It uses directives for agents. The way it is organized is interesting! document.currentScript.replaceWith(...) can be used as a nifty way to dynamically replace content, effectively creating a mini-templating language. Via Simon Willison. Standardization and personalization are independent axis. In fact, standardization can make personalization easier. ChatGPT Before the printing press, people used to spell words differently. Shakespeare, for example, had different spellings of his name. With the printing press and distribution of publication, spelling became standardized. But interestingly, it standardized by freezing the current conventions. So some oddities of the spellings people used are now a formal part of the language definition. In pharmacies, we call them “compounders” because they actually used to compound, i.e. mix, medicines - personalizing them to each individual. That faded away as standardized tablets became more common. Fonts are becoming more sophisticated. Some features in newer fonts are: Icons: NerdFonts is an example of a font class with 10K+ icons embedded - like 󰊢 GitHub, 󰈙 files, or 󰀨 terminals. Opt-in Ligatures: Display multiple characters as one symbol, e.g. => as ⇒, != as ≠, or >= as ≥, letting users pick what to enable. Variable weight: Continuously adjust thickness, e.g. weight 450 or 575 instead of being limited to Regular (400) and Bold (700). Smart width: Make characters narrower or wider, adjust shapes based on neighbors (making mi look more evenly spaced), etc. while preserving mono-space. “No one cares more about your case than you, so take control and send lots of follow-ups.” Things to know if you get cancer. The article has usefuls tips for important things in life, in general - not just cancer. Agents are making this easier. Because Amazon has a fair pricing policy, a strange consequence is that if Amazon sellers raise prices on Amazon they’re pressured to raise prices elsewhere too! So, increasing Amazon seller fees tends to increase prices across the board. How Amazon pushes up pricess all across the Internet Will Jev destroy its servers to save 5 lives? Will other models? A lovely set of experiments based on the trolley problem. Mistakes I made Week ending 25 Oct 2026 ...

I conducted a session on Mon, 19 Oct 2026 at TDS Course Ideation — Open Design Meeting, Sep 2026 - Remote. Speakers: Anand S Summary: A week after inventing agent-friendly assessments in public, the TDS team meets running code, student questions, checking budgets, a dropped $500 experiment, leaky graders, and a pass rate that only counts who scored above zero. Here’s the link to the session Links: Video (60 min) Transcript Captions Audio (60 min)