The Unread Pile — Edition No. 25

Edition No. 25 Monday, 19 October 2026 19 new saves

The Lead Story

The Default Experience Is Becoming a Suggestion

Via @hnshah — the gathint TV job

Editorial illustration: a living room TV showing a clean minimal custom launcher interface

Illustration: the ad banner slides off the screen. The remote never leaves your hand, but the software inside the TV just changed owners.

“I wish this thing worked differently” is becoming something people can actually fix themselves. Someone just used AI to remove the ads from his parents’ TV and build them a better interface. One afternoon, one TV model number, a coding agent — and the sponsored ad that used to gate Netflix is gone, twenty-five preinstalled junk packages with it, and the launcher now looks, in his words, like an Apple TV.

Think about how many products we own that don’t work the way we want. The toaster firmware that nags. The printer driver that phones home. The TV that shows a sponsored ad before you can get to Netflix. For decades these were facts of life: you bought the default and lived with it. hnshah’s point is the phase change — the default experience is becoming a suggestion. The gap between “this annoys me” and “I rebuilt it” collapsed to an afternoon, because the agent does the device discovery, the package surgery, and the UI rewrite that used to require a firmware engineer.

The complaint-to-command pipeline

Sequence per the original post, via @hnshah. Time from annoyance to new launcher: one afternoon.

“Think about how many products we own that don’t work the way we want. The default experience is becoming a suggestion.”— @hnshah

This is personal software’s final boss form. Earlier editions of this paper tracked vibe-coded snacks and two-person startups; this is the next rung down the ladder — zero-ship software. Nothing is published, nothing is monetised, no one reviews it. It exists for one household and one TV. The economics of software used to demand a market; now they only demand an afternoon.

Main learnings

Complaints are now compile targets. “I wish this worked differently” parses straight into a work order.
The barrier was never skill, it was friction. Finding the TV on wifi, identifying the packages, writing the launcher — each step was boring expert work an agent now absorbs.
Manufacturers lost the last mile. The device maker’s ad inventory was deleted by the customer’s agent. Who owns the screen now?
One household is a market. Software that serves a single user is now worth building. Expect a million invisible forks of every default.

Practical takeaway — audit your own defaults: the one device, workflow, or app you tolerate daily. The cost of fixing it just fell to “an afternoon.” Price your annoyances again.

The post

The wire

Codex limits drain in seconds — quota complaints keep coming post-reset.
Opus 5.5 — 24 days, still nobody has hit a usage limit.
Resumatic.ai — designed and built in 15h47m with Opus 5.5.
Zerodha — US stocks in beta; 0.5% forex markup undercuts rivals.
The Winter Owl — the new Malfoid AI short is out.

Builders

Tokenmaxxing at the Hackathon

Via @kevinyien

Editorial illustration: students at a hackathon with laptops and rising token streams

Illustration: a hundred free tiers, one shared brain. The students don’t ration tokens — they route around the rationing.

“The next generation is gonna be just fine.” Kevin Yien spent a Friday at a UT hackathon and came back with a field report that reads like dispatches from the future: students tokenmaxxing — running on the order of a hundred free-tier accounts across the major providers, with a context harness that passes project state between them so work survives rate limits. When he asked how they afforded the tokens, they described the setup like it was plumbing.

The interesting shift isn’t the free-tier arbitrage; it’s what they write and review. Hint: it’s not the code. They spoke about what they build and how they make sure it works and is reliable — the review target moved up the stack from syntax to outcomes. The code is the exhaust; the spec and the verification are the craft. And when Yien asked a junior what he thought about the future: “As long as you have brain cells and a willingness to work, I think you’re gonna be fine.” These kids don’t believe in a “permanent underclass” of knowledge workers. They believe they can build and learn their way to whatever they want.

The student inference stack

Per @kevinyien’s hackathon report.

Main learnings

Compute poverty is a routing problem. A hundred free tiers plus state handoff beats one paid tier without it.
Review moved up the stack. They review behaviour and reliability, not lines of code.
Optimism is a build strategy. “Brain cells and a willingness to work” is not naivety; it’s the revealed preference of people shipping daily.

Practical takeaway — if juniors route around rate limits for free, your moat isn’t your token budget. It’s what you review for: outcomes, not output.

The post

The 15-Hour Website

Via @jacob_jacquet

Jacob Jacquet designed and built resumatic.ai in 15 hours and 47 minutes with Opus 5.5. Twelve days later: “Starting to see Google’s reaction to the rebuilt website.”

From blank page to incumbent reaction

Two numbers tell the whole story: sixteen hours to ship, twelve days to provoke a response from the category’s eight-hundred-pound gorilla. The window between “nobody knows you exist” and “the incumbent notices” is now measured in days, not years. That’s terrifying and wonderful in equal measure.

Practical takeaway — ship speed is your only moat in the first fortnight. Use it like one: the clock starts when you launch, and the incumbent’s radar has a twelve-day lag.

The post

The 48-Hour Siege

Via @thdxr

Thariq’s systems have been under non-stop attack for 48 hours. Side effect: everything is getting hardened fast. His conclusion: “we basically need agents going 24/7 because that’s what they have.”

The attackers run agents around the clock, so the defenders must too. This is the Voxyz playbook from last week’s edition, inverted: the adversarial-security drills his team practised on their own product are now the production posture. Defence is no longer a sprint you run after an incident; it’s a shift schedule you staff with machines.

Practical takeaway — if your threat model is humans with scripts, update it: it’s agents with stamina. The defence has to be a roster, not a reaction.

The post

The allowance wars, round two

The Quota-Drain Panic

Via @TokenGremlin, @nicdunz, @cneuralnetwork — counterpoint by @coryalthoff

Editorial illustration: two dashboard gauges, Codex pinned in the red and Claude calm in the green

Illustration: same $20, two very different Thursdays. The gauge is the new feature comparison.

Last week’s reset wars aren’t over — they metastasised. Codex users report usage limits draining way faster than normal since October 16, some seeing their quota “wiped in seconds,” with complaints continuing after Tibo’s reset. @nicdunz watched his gauge fall from 100% to 84% nearly instantly from seconds of work on Astra low fast — “for weeks before today it was extremely usable without draining my limits.” TokenGremlin’s diagnosis is blunt: daily resets exist because the Codex limits are awful, and “limits so good you never need a reset would be even better.”

Then the comparison that stings. @cneuralnetwork did the accounting: on the $20 Codex plan (44M tokens) he burned ~20%; on the $20 Claude plan (100M tokens) he burned ~10% — Codex so inefficient he’s considering two Claude plans instead. And @coryalthoff delivered the quietest flex of the week: “It’s been 24 days and still nobody has hit a usage limit using Claude Opus 5.5.” On one side, gauges in the red; on the other, a limit nobody has found yet.

Allowance as the feature matrix

The kicker: @shobhitic reports Claude Code just cut request latency by compressing requests with gzip — a format from 1992 — and notes Anthropic ships 30,000 merges a month yet this took ages. His punchline lands for everyone watching the gauge wars: “lots of money to be made optimizing people’s vibe-coded apps.” While Codex fights over quota accounting, the optimisation frontier is wide open: compression, caching, routing — the unglamorous infrastructure of the agent age.

Main learnings

Allowance is the product now. The comparison shoppers ran on model quality; they now run it on gauge behaviour.
A reset is an admission. Tibo’s daily resets read as generosity; the timeline reads them as the symptom of broken limits.
Efficiency arbitrage is the next services wave. 1992’s gzip on 30K-merge-a-month codebases — the low-hanging optimisation is everywhere.

Practical takeaway — pick your coding plan by the gauge, not the model card. And if you sell to builders, sell them back their quota: the optimisers of vibe-coded apps are about to eat well.

The drain reports · Resets are a symptom · 100% → 84% in seconds · The $20 accounting · 24 days, zero limits · gzip, 1992

Strategy

Seek Friction

Via @vaibhavbetter

Vaibhav’s founder memo for the commodity age: with AI making everything a commodity, founders will hear way more NOs than ever before. You won’t even get meetings for incremental stuff anyone can build. So think hard about what to build right now — and go where the SaaS wave never touched: friction, ops, complexity, compliance and regulatory thickets.

The build/no-build matrix

Framework per @vaibhavbetter.

The logic inverts the last fifteen years of advice. The 2010s told you to find the smoothest onboarding, the cleanest API, the most self-serve motion. In 2026 the smooth stuff is exactly what an agent builds in a weekend — see the 15-hour website above. The moat moved to the places with paperwork: licences, inspections, physical operations, regulated money flows. “Think full-stack. Think outcomes. Think value-aligned pricing.” Not per-seat SaaS on a commodity; pricing that moves when the customer’s P&L moves.

“Seek friction. Seek ops. Seek complexity. Seek compliance and regulatory complexity. Seek more such stuff that was never touched by the SaaS wave.”— @vaibhavbetter

Main learnings

Incremental is uninvestable. Anything an agent can build in a weekend can’t raise, can’t charge, can’t keep a meeting.
Boring is the new defensible. Regulated, operational, physical — the stuff the SaaS wave skipped because it was hard.
Price the outcome, not the seat. Value-aligned pricing survives commoditisation; per-seat doesn’t.

Practical takeaway — before your next build, score it on the friction axis: if a weekend agent could ship 80% of it, you’re in the red quadrant. Move toward the paperwork.

The post

Money

Zerodha Goes American

Via @ActusDei

Zerodha has enabled US stocks in beta for some users, and the fee schedule reads like a declaration of war on the incumbents. Brokerage at 0.1% matches Groww and undercuts INDmoney and Vested at 0.25%. But the real tells are at the edges: minimum brokerage of one cent versus a dollar at Groww, a $10 maximum where Groww has no cap, and a 0.5% forex markup against 0.8–1% elsewhere — which the post flags as the strongest advantage.

The fee comparison

Figures per @ActusDei. Underlying broker likely India INX / View Trade; Groww runs on Alpaca. UCITS not available yet on either.

The plumbing matters too: the underlying broker is most likely India INX (BSE-owned), with View Trade at the back end, versus Groww’s Alpaca rails. One gap both platforms share: no UCITS yet. For the Indian investor who DCA’d into US tech through a 1%-markup corridor, the remittance cut alone rewrites the math on every transfer. The discount-broker playbook that conquered Indian equities is now pointed at the dollar.

Main learnings

The fight moved to the edges. Headline brokerage is tied at 0.1%; the war is in minimums, caps, and forex.
One cent minimums change behaviour. A $1 floor taxes small SIP-style US investing; $0.01 removes the friction entirely.
Forex is the margin that matters. On recurring remittances, half a point of markup beats a tenth of a point of brokerage.

Practical takeaway — if you invest in US stocks from India, recheck your corridor: the cheapest headline rate may not be the cheapest transfer. And watch the beta rollout — beta today, default tomorrow.

The post

Human

The Four Qualities, in Reverse

Via @ilavanyajain — quoting James Clear

Lavanya Jain flagged a James Clear career checklist worth pinning: the four qualities of a great career are 1) I enjoy it, 2) I’m good at it, 3) I make good money, 4) I’m around fascinating people. The twist is the instruction: answer in reverse order. Start with where the fascinating people are; then how you could make money with them; then which of those you’re good at; then which you’d enjoy.

Search top-down, live bottom-up

The reversal fixes the classic failure mode: starting with “what do I enjoy?” strands you in hobbies with no market. Starting with people strands you, instead, in rooms where money, skill, and taste compound. It’s the career version of vaibhavbetter’s friction thesis from the strategy section: go where the interesting density is, and let the economics follow the network. The order of the search is the whole trick.

Practical takeaway — run the reverse pass on your own role: name the fascinating rooms first, then work backwards to what you’d enjoy there. If the list collapses, the room is wrong, not you.

The post

Briefs

The Theranos pre-mortem on Reddit. Years before the WSJ broke the story, an anonymous Theranos employee discussed the whole fraud — without naming the company — on a popular subreddit, with screenshots to prove it. The crowd knew; it just didn’t have a byline. Post
Push off the average. Replying to Rakshit: agents default to the most likely output, so the interesting work is all in pushing them off the average. A one-line craft note for every prompt you write this week. Post
Writing as thinking. “S-tier advice” — Tim Ferriss on using writing to sharpen thinking, in video form. The companion to the tokenmaxxers’ review stack: the spec is the thinking. Post
The Winter Owl lands. Skyr0n’s new Malfoid short — “The Winter Owl,” 1:35, made with AI, Draco × Harry — is out, and he’s happy with how the scenes turned out. The TikTok sentiment flip from last edition continues. Post
Skill-issue.in. Bookmarked, one verdict: “crazy tbh.” A link worth your click more than our summary. Post
The luck interview. A long Reddit story argues most big-tech interviews are about luck — the screenshot made the likes, so the argument must be landing. Post