Ant Stanley

@iamstan.dev

2685 Followers 4546 Following

Tell us a bit about yourself

Switched to Opus 4.8 (1M context) — now your default model · Fable 5 requires usage credits · /model to change
screen shot from mason, with the words 'this wall is sealed this waller asked to be seen by signed-in visitors only. mason reads walls logged out, so this one stays bricked up. try another wall:" on a screen to let users select a different handle
Photo of Lily and Lola, Vizsla sisters, sleeping in the dog bed under my desk covered by a blanket
Still from the movie 'Hunt for the Wilderpeople'
Screen of following text: 1. Gather. Resolve your handle, fetch your follows, sample a cohort of 100: sixty known-active authors plus forty seeded-random ones, so every refresh explores a different slice of your graph. 2. Score within a kind. Each brick gets recency_decay × log(engagement), with half-lives that suit the medium: posts 24h, blogs 7 days, trailers 30 days. 3. Never compare across kinds. The mixer picks the next kind by how far the wall is from its target ratio (roughly 70/15/10/5), then the best brick inside that kind. Comparing a liked post against a blog by raw score would bury the blog forever. 4. Keep it varied. An author cannot reappear within eight bricks, and a seeded jitter nudges the ordering so the wall feels alive. 5. Stay stable. The seed lives in the cursor, so endless scroll never repeats or reshuffles, and every fresh visit builds a different wall. The honest caveat: this surfaces your own follows, so the serendipity is bounded by your graph. It is a better-arranged timeline, not discovery beyond it. The exploration cohort is what stops it collapsing into the same twenty accounts.
screenshot showing all weekly limits for Fable being reset
A video showing me using the https://www.neuronpedia.org/qwen3.6-27b/jlens site to swap the word 'France' with 'Brazil' using the 'General Broadcast' J-Lens property, and the LLM becoming confused.
a screenshot showing Claude Code model usage quotas. Current session - 25% used Resets 3:59am (Africa/Johannesburg) Current week (all models) - 65% used Resets Jul 13 at 7:59am (Africa/Johannesburg) Current week (Fable) - 100% used Resets Jul 13 at 7:59am (Africa/Johannesburg)
screen shot of two skeets one above the other. The top one is Dan Abramov complaining about AI written blog posts from Cloudflatre. The bottom skeet is a video from Phil Hawksworth saying he's joining Cloudflare.
a screen shot of my Claude usage quotas showing 0% used on all
screen shot from a Claude Code session with the following text: What was wrong: Task 03's implementer wasn't crashed — it was wedged. At Opus max effort it went down a rabbit hole building throwaway serde spike projects in the scratchpad (to settle the array-vs-map deserialization approach) instead of writing the deliverable, then the transcript froze — ~2h19m elapsed, zero code written to the repo. Tasks 01 and 02 finished fine at max (38 min, 20 min), so this was an over-exploration/stall anomaly, not a systemic crash.
What are the main constraints to higher levels of demand? The principal constraints are grid-related and regulatory rather than demand-side. Transmission and distribution capacity remain the binding bottlenecks. While the approved Congestion Curtailment Proposal aims to unlock approximately 3.4 GW of grid capacity in constrained regions by allowing managed curtailment, grid limitations persist.
screen shot with text: These trajectories underpin the national strategy towards the long-term solar deployment goal of 25 GW by 2039 outlined in the Integrated Resource Plan (IRP). The 2025 South African Renewable Energy Grid Survey (SAREGS) identified a total renewable-energy pipeline exceeding 220 GW, an 86 GW increase on 2024, with over 46 GW of advanced-stage solar PV projects proposed for grid connection by 2030, supplemented by a further 11 GW of solar PV paired with BESS. This pipeline is supported by exponential growth in the private sector. National Energy Regulator of South Africa (Nersa) data shows that annual registered private capacity peaked at 5.23 GW in 2025, up from 0.13 GW in 2021. This pushed cumulative registered capacity past 12 GW.
Screen shot from /usage command output in Claude Code Total cost: $260.71 Total duration (API): 9h 13m 8s Total duration (wall): 18h 9m 4s Total code changes: 10110 lines added, 1454 lines removed Usage by model: claude-haiku-4-5: 2.8k input, 46 output, 0 cache read, 0 cache write ($0.0030) claude-sonnet-5: 223.0k input, 707.0k output, 105.8m cache read, 2.1m cache write ($50.74) claude-opus-4-8: 369.2k input, 933.2k output, 116.5m cache read, 4.3m cache write ($120.62) claude-fable-5: 598.8k input, 611.8k output, 23.3m cache read, 2.4m cache write ($89.34)
screenshot from a website just showing distinctive design commonly found in claude designed websites. The tell is coloured outer left shadow with rounded corners on a card
A coffee sitting in a lazy dawn
A perfect flat white.
screenshots from https://intheweights.com/p/ant-stanley with details on where I am in the weights. Shows that Opus 4.8, Gemini 3.1 Lite, and Kimi K2 0905 all know me...
The Obama Presidential Library, with the word 'Reqwest' embossed across the facade
photo of John Candy in Plane's Trains and Automobiles
Fair pushback - "can't" was imprecise.
60 Hudson Street as seen from One World Observatory in June 2015
a photo of a partially covered sun in the sky with trees lining the bottom
Arrested Development meme with the text "No, it never does. I mean, these people somehow delude themselves into thinking it might, but .... but it might work for us"
Siphiwe Tshabalala being interviewed on TV before the world cup opening game
One thing worth flagging for next time: Claude Code's safety classifier hard-blocks pushes to a new external remote regardless of authorization, so that final step will always need to be run by you (I can create the repo, audit for secrets, and stage main - just not push). I've noted that in memory so I'll go straight to handing you the command in future rather than discovering the block again.
Screenshot with the following text. Fable 5 is here! Our newest model for complex, long-running work ▎ Included in your plan limits until Jun 22, then switch to usage credits to continue. ❯ /model ⎿ Set model to Fable 5 and saved as your default for new sessions
screenshot of Claude code usage
Coffee. Glorious coffee with slightly improved latte art
A beautiful coffee with an attempt at latte art
Photo of coffee gear. A Lelit Victoria espresso machine, a Eureka Mignon Silenzio grinder, a Lecreuset mug on top of a Timemore scale as coffee pours.
crates.io but if it was npmx.dev
screenshot of the crates.io page styled like npmx.dev ... still some bugs but getting there
a screenshot of the crates.io page with npmx inspired layout
I'll inspect the live npmx. dev homepage to capture its actual layout, then mirror that structure for the crates.io home. Let me fetch it. • Fetch(https://npmx.dev) L Received 0 bytes (403 Forbidden) L Allowed by auto mode classifier • WebFetch is blocked (403). Let me try fetching the server-rendered HTML directly with a browser user-agent. • Bash(cd /tmp && curl -sL --max-time 30 -A "Mozilla/5.0 (Macintosh; Intel Mac OS X 10_15_7) AppleWebkit/537.36 CKHTML, Like Gecko) Chrome/126.0 Safar1/537.36" https:/..) size=52180 L HTTP 200 --- 52180 npmx-home.html Shell cwd was reset to /Volumes/Delorean/code/crates.io Allowed by auto mode classifier • Got the SSR HTML. Let me parse out the layout structure - headings, sections, links, button labels, placeholders - in document order.
A screen shot of the crates.io home page with the npmx design system applied in dark mode
I now fully understand the design bundle. It's a complete design system for "прх" - a redesign concept for browsing the nom registry (dark-first, monochrome OKLCH, Geist/Geist Mono with mono headings, single sky accent, border-based depth, no shadows, Lucide icons). It was derived from npmx-dev/npmx.dev (a Nuxt app). But I'm in the crates.io repo - the Rust/Cargo package registry. This is a meaningful mismatch I need to resolve before implementing. Let me explore the crates.io frontend to understand the stack and what "implement the designs" could feasibly mean here
screenshot of Apple Music's home page showing the new Death Cab for Cutie album "I built you a tower" being recommended
it's the kind of file that falls between dimensions; worth keeping in mind.
Let me verify ground truth before declaring victory:
Coffee
video of a user selecting 'ultracode' effort in Claude Code
Video showing the effort widget in Claude Code. Selects the 'ultracode' setting which triggers a purple pulsing animation in the background.
A screenshot from Claude Code with the words "This is excellent, rigorous work. Let me read the rest"
screen capture of a cmux session with a terminal window next to a browser window in the same app. The terminal window shows a claude code session, which is being prompted to update the site The browser window is showing the live preview of the site
screenshot from Claude Code showing dynamic workflows and their associated token usage. It contains the following content --- ✔ e2e-spec-apply-fixes 7 agents · 140.5k tok · 58s ✔ e2e-spec-semiformal-review 15 agents · 865.9k tok · 6m 57s ✔ e2e-infra-research 4 agents · 274.1k tok · 2m 47s
Screen shot of an adhoc line of bash Claude Code is running to validate Claude Code is running... the bash is| echo "= claude CLI =" && which claude && (claude -version 2>&1 | head -1 I| echo "version check failed")
A screen shot of the Claude Code effort toggle with a new 'ultracode' setting
Screen shot of the silly.best site with its jumping dinosaur game
A google search with the words: "How many p's in the name penelope" Google AI response: There are two p's in the name Penelope (the first and eighth letters). (P-E-N-E-L-O-P-E)
Dog on a chair at a table waiting to be served
Screenshot with following text. Why Workspaces Beat Worktrees for Parallel Work Revision-based, not branch-based. A workspace points at a revision; you can rebase or move it without branch gymnastics, and two workspaces can edit the same underlying change without conflict-locking. Auto-snapshot. jj records the working copy on every command, so there's never a "did I stash?" moment when an agent switches direction. Shared log view. jj log shows which workspace is editing which change (@ for current workspace, <name>@ for others). Agents can observe each other's progress. No branch name required. You don't have to invent a branch up front; the workspace name is enough, and bookmarks (jj's branches) can be created later when work is ready to ship.
A sign that says "Bleak House"
Jimmy about to consume a delicious sole
View of the sunrise over an aeroplane wing
Charlie Brown charging at the football about to kick it ... will Lucy pull it away at the last minute