at 7:00am my inbox had 2,341 unread emails. at 7:18 it had zero.
i didn't read a single one. Claude did.
here's what was hiding in there:
→ a $1,840 invoice a client owed me since september. never saw the reminder
→ a recruiter who replied 19 days ago. i thought they ghosted me
→ 3 renewals on autopay this month, one for a tool i quit in spring
→ 412 newsletters. it kept the 6 i actually open
and 37 replies drafted in my tone. sent: 0. nothing leaves without my yes
how to do it in 20 minutes:
1. connect Gmail to Claude (settings → connectors)
2. start safe: "read only. don't send, don't delete anything"
3. then: "go through my unread from the last 90 days. sort into: money, people waiting on me, renewals, newsletters, noise"
4. "for everyone waiting on me, draft a reply in my tone. show me before anything goes out"
5. the part nobody does: "make a list of everyone i owe a reply to, oldest first". that's where the 19-day recruiter was hiding
your inbox isn't a to-do list. it's a pile of money and people you forgot
Grok vs Claude is the weirdest rivalry in tech right now. and almost nobody has noticed why.
i put them head to head, round by round:
ROUND 1: the model
Grok 4.8 is rolling out to Grok Bot users. 2.5 trillion parameters (per Musk), and xAI rewrote the whole training + inference stack from scratch in C/C++
Claude has Opus 5.5 and Fable 5.2 (nov 9). Anthropic says Opus 5.5 on "Medium" effort matches the old Opus 5 on "High"
→ raw ambition: Grok. polish: Claude
ROUND 2: what Elon himself says
on Grok 4.8: "not as good" as Opus 5.5
on Opus 5.5: "masterpiece"
when the guy who owns Grok calls the other model a masterpiece, that round is over
→ Claude
ROUND 3: the comeback plan
Musk says Grok 5 should pass Opus 5.5 in "2 or 3 months"
Anthropic's plan: go public. ticker ANT, trading expected tomorrow, valued at up to $1.4T
→ draw. one bets on a model, one bets on a market
ROUND 4: the twist nobody talks about
Grok was built on Colossus, Elon's supercomputer
Anthropic RENTS compute on that same Colossus. $1.25 BILLION a month, per SpaceX's own filing
so every month, Claude's maker sends Elon a check bigger than most companies' yearly revenue
→ Elon wins either way
my verdict:
today, Claude is the better model. even Elon says so
but Elon owns the arena both of them fight in
and if Grok 5 really lands in february, this whole card flips
who's ahead by summer 2027, Grok or Claude?
there's a version of Claude you're not allowed to use.
it's called Claude Mythos
same model as the public Claude. the difference: some of the safety locks are off
Anthropic wouldn't release it to the public because of what it can do: find holes in software
→ Anthropic claims it found vulnerabilities in every major operating system and every major browser
→ Mozilla reportedly patched 271 Firefox vulnerabilities with the preview version
→ only ~40 companies got access at first, through a program called Glasswing
on nov 9 they shipped Mythos 5.2. still restricted.
so the best bug hunter on earth might be an AI most people will never get to touch
should Anthropic ever release it to everyone? yes or no
you've been prompting Claude wrong. Anthropic just said so.
they published a guide for Opus 5.5 and half the habits everyone learned in 2024 are now slowing it down
stop doing this:
1. "think hard" / "think carefully". it always reasons on its own. without it, answers start faster with no measurable drop in quality
2. cranking effort to Max. start at Medium. in their tests Opus 5.5 on Medium matched Opus 5 on High
3. feeding instructions one by one. one message: the goal, the format, and when it should check back with you
4. typing tables out by hand. attach the file or a screenshot
5. asking for "a rough draft first". ask for the finished doc
6. "make it less generic". name the exact thing you hate: "no cream background, no pill buttons"
7. the part nobody does: connect your files and apps so it pulls the data itself
bonus: Sonnet on Max effort costs about as much as Opus 5.5. so stop "saving money" that way
most people will keep writing 2024 prompts for a 2026 model
wait. Anthropic pays Elon Musk $1.25 BILLION a month.
not a typo. per SpaceX's own IPO filing, Anthropic rents compute on Colossus I and II, the supercomputers xAI built for Grok
that's:
→ $41 million a day
→ $1.7 million an hour
→ ~$476 every second
so while you watched this video, Anthropic paid Elon about $7,000
the twist: Musk says the deal is short on purpose. his words: "i thought we might need the compute back at some point"
and last week he called Claude Opus 5.5 a "masterpiece"
so the guy building Grok is renting his supercomputer to the company that makes the model he says is better
AI in 2026 is weird
@elonmusk renew or take it back?
this is f*cking embarrassing.
i was paying $2,289 a year for 14 subscriptions i forgot existed
Claude found every single one in about 4 minutes
→ LinkedIn Premium i never opened
→ 2 VPNs. 2 cloud storages. 2 meditation apps
→ a Notion team plan for a team that left in 2024
→ an AI photo app "free trial" that ended last year
1,284 transactions → 14 forgotten subscriptions → $2,289/yr → 1 prompt
here's exactly how to do it:
1. download 12 months of statements from your bank app (CSV is best)
2. hide your account number first. date, merchant, amount is all it needs
3. give it to Claude: "find every recurring charge, group by merchant, monthly + yearly cost, flag anything that started as a free trial"
4. then: "which of these overlap?" this is where the 2 VPNs show up
5. the part nobody does: "write me the cancel steps for each one + a reminder 3 days before every yearly renewal"
it's the easiest $2k you'll ever find
still don't understand why everyone isn't doing this yet.
at 11pm i gave Claude my Downloads folder. 4,312 files. 9 years of chaos.
by 2am:
→ every file renamed so i can actually search it
→ 1,108 duplicates moved to a review folder, nothing deleted
→ my tax PDFs from 2016 to 2025 in one folder, by year
→ 38 GB of old installers flagged
→ and it found the passport scan from 2016 i swore i lost
i typed one prompt. here's exactly how:
1. give Claude access to ONE folder on your computer. start with Downloads
2. ask for a plan first: "scan this folder, don't move anything yet, show me how you'd organize it"
3. read the plan and fix what you don't like in plain words
4. then: "go. never delete anything. whatever you'd delete goes to _review"
5. here's the part nobody does: "write the rules you used into sort-rules.md" so next month it takes 30 seconds
4,312 files → 6 folders → 1 prompt → 0 files lost
save this for the weekend you finally clean your laptop
This is f*cking insane.
AWS just released a free local copy of its own cloud. 200+ services, 400+ API models
and not one line of it was written by a human
"All of AWS Emulator's code was written and tested by agents."
the wildest numbers from the story:
→ first SQS emulator prototype: ~15 minutes
→ the internal version: built over a weekend
→ a replication engine that took ~4 months the first time: ~2 hours once the agent had a spec and tests
it's public domain (CC0), runs on your laptop, no AWS account needed
and they're honest about it: the repo has a folder literally called "things-that-do-not-work"
the part nobody talks about: the humans didn't write the code. they built the checks that tell the agents when they're wrong
that's the engineering job now
save this, it's the clearest picture of where software is going
someone stole AI keys from a cloud platform and ran up ~$740,000 on other people's accounts
this actually happened last month, at Railway:
→ oct 20: an attacker escaped Railway's build isolation
→ decrypted env variables across up to ~100,000 workspaces
→ went hunting for one thing: AI API keys
→ ~$740,000 in estimated unauthorized AI usage
Railway first said it couldn't refund the spend. now it says it'll cover what the AI providers won't
if you have an AI key sitting in a .env file, do this today:
1. set a hard monthly spend cap on every provider
2. one key per project, never one master key
3. usage alerts at 2x your normal day
4. rotate any key that ever lived on a shared platform
5. the part nobody does: delete keys you don't use. a dead key is still a live bill
your AI key is basically a credit card now. treat it like one
anthropic goes public next week. at up to $1.4 TRILLION.
the company behind Claude was started in 2021 by a few researchers who walked out of OpenAI
the numbers:
- ticker: ANT on nasdaq
- $70-$78 a share
- 825M shares, up to ~$64B raised
- google, nvidia & 5 other partners buying another $26B on top
- trading expected nov 19
the twist: in september they were eyeing ~$2 trillion. an AI stock sell-off cut that down
and the book is already covered at this price
2021 → claude → $1.4T in 5 years
is $1.4T too much for anthropic, or still cheap?
the "can someone take notes?" job just died
anthropic launched Claude Meetings yesterday
claude joins the call, says it's there, and works while you talk:
→ drafts the plan
→ draws the timeline on a whiteboard
→ answers questions in chat
→ pulls last year's numbers from your work apps
you hang up, everything's already in claude.ai
30 min call → 1 plan → 1 timeline → 9 action items (simulated run below)
most teams will keep paying people to summarize meetings while a small group lets the meeting finish its own work
beta on Team & Enterprise
this is f*cking brutal.
canva and squarespace just cut about 20% of their people each
and TechCrunch's Equity podcast said the quiet part out loud:
"if Claude can do XYZ, what do I need Squarespace for?"
both companies made design & websites simple. now AI makes it even simpler, and the "easy tool" companies are the ones cutting
canva's CEO told staff they'll reduce "sub-specialization"
translation: fewer people, each doing more with AI
most people will read this as "AI took jobs"
the real lesson: if your whole product is "we made X easier", someone is already making it easier than you
elon musk just called a competitor's model a "masterpiece"
not grok. Claude Opus 5.5.
same day he dropped the grok 4.8 update:
→ JAX & SGLang dumped
→ whole RL + inference stack rewritten in C/C++ (+ some assembly)
→ 2.5T params, now in RL
→ limited rollout to Bot users this week
and he admits 4.8 is "not as good" as Opus 5.5
then the flex: grok 5 should pass it in 2-3 months
honestly that's the most respect xAI has ever shown Anthropic and a deadline everyone can now track
bookmark this, come back in february & check
Replying to @cradrrat
We have dumped JAX/SGLang and rewritten the entire RL and inference stack from scratch in C/C++, which took some extra time.
Training (aka pre-training) is done. Lot of cleanup was needed on the SFT data to get that mostly right. Now, 4.8 is in RL and the evals are starting to climb. We will do a soft rollout to Bot users later this week and see how it goes.
Grok 4.8 will bear no resemblance to 4.7 (thankfully). It is not yet incredible, but, this time, the foundations are built right, so I expect it to evolve to be something special.
🚨BREAKING: MY IPHONE HAD 92,316 CACHED FILES I NEVER KNEW EXISTED
I asked Claude Code to scan my iPhone for hidden files outside my normal Photos library. I wasn't expecting much. Maybe a few temporary files. Old thumbnails. Nothing unusual.
The scan finished. 92,316 cached files.
Here's what it found:
- Photos: 21,284 files - 18.2 GB
- Screenshots: 12,077 files - 4.1 GB
- Videos: 1,412 files - 21.4 GB
- Thumbnails: 4,623 files - 0.9 GB
Almost 45 GB of media sitting in cached storage.
Some of the files were things I recognized. Others had filenames I couldn't make sense of. There were duplicates, fragments, and media files I didn't remember ever seeing.
The weirdest part wasn't the amount of storage they occupied.
It was that none of these files showed up when I browsed my regular Photos library.
I started opening them one by one. Old screenshots. Video fragments. Photos from months ago. Some files had timestamps that made absolutely no sense to me.
I thought deleting my photos meant deleting my photos.
Apparently, there's a lot more to look at when you start scanning what's actually stored on a device.
92,316 files. And I had no idea they were there.
Claude Code found 145,763 photos I permanently deleted from my iPhone.
Three years of deleted photos and videos still sitting in a hidden cache. Any app with file access can read them.
I plugged my iPhone into my MacBook to run a backup. Claude Code was doing the security audit from my last post and flagged a directory I wasn't even looking for.
PhotoData/Thumbnails. Over 146,000 files. Photos, videos, screenshots. Every single one was something I deleted.
Photos from 2023, 2024 and 2025. The ones I deleted the second after taking them. The ones I cleared from Recently Deleted. The ones I deleted three separate times because I wanted to be sure they were gone. All still there.
Passport photo I took for a visa application in 2023.
Screenshots of my bank app with full account numbers visible.
Private photos I sent someone and immediately deleted after a breakup.
Screenshot of my crypto seed phrase I took temporarily in 2022.
Photo of my ID and credit card I took to remember the numbers and deleted 5 seconds later.
Screenshots of my ex's private Instagram stories I saved and permanently deleted in 2023.
Some were full resolution. All organized by date going back to 2021.
When you delete a photo on iPhone, iOS removes it from your camera roll and Recently Deleted. It keeps a compressed copy in a Spotlight search index cache so your phone can find things when you search. Apple never clears it. Not after 30 days. Not after a year. Never.
They told you it was permanently deleted. It wasn't.
I checked which apps on my phone had Files and Folders access. 23 apps. Weather apps. QR code scanners. PDF readers. A flashlight app I downloaded in 2022. Any one of them could have been reading every photo I ever deleted this entire time.
Went from 23 apps to 3 in about two minutes.
Full step by step below. Takes 2 minutes. Save it so you don't lose it and send it to anyone with an iPhone.
Readers added context they thought people might want to know
The author is posting AI-generated engagement bait text with various random claims of deleted photos or videos on a daily basis.
1. x.com/coolerwave/sta… - "BREAKING: Claude Code found 12,847 photos of my face on my iPhone
2. x.com/coolerwave/sta… - "Claude Code found 214,856 photos and videos I permanently deleted
the easiest way to lose money in crypto is to need the money back quickly
suddenly every -20% candle becomes an emergency
every pump feels like your last chance
every random twitter account becomes an analyst
if you can't sit through volatility, you don't have a thesis
you have a position.
the real test of a trading system isn't whether it makes money when you're watching it
it's what happens at 3 a.m. when the feed goes stale, an order doesn't fill, the risk monitor dies and you're asleep
that's when you find out what you actually built a system
or a collection of things that usually work
I just took a NO position on an official US Iran ceasefire by March 31 on Polymarket.
Entry around 69c.
My reasoning is basically the same as for the March 15 market.
Nothing meaningful has changed.
Trump keeps demanding full nuclear concessions.
Strikes are still happening almost every day.
Neither side looks ready to slow things down.
Talks are going nowhere and mediators have not moved the needle.
And the resolution requires an official public agreement.
Right now there is not even a hint of that.
Market still prices YES around 25 percent.
To me that feels high for March.
🧵 My take on the Polymarket question: “Will Iran close the Strait by March 31?” — I’m betting NO.
Yes, Iran’s Revolutionary Guards have issued very strong warnings and attacks have disrupted shipping through the Strait of Hormuz. Many companies have pulled vessels, and insurers are canceling war risk cover - essentially making the area too risky for normal traffic right now.
But a true 80%+ sustained drop in moving averages caused directly by Iran’s actions - that’s a high bar. Shipping data shows near-halt activity at the moment, but that’s largely due to multiple parties choosing safety first - not an enforced legal blockade.
Fully closing the Strait would hurt Iran’s own export economy and risk broad military escalation. Tehran also relies on that route for its oil trade, including shipments to big partners.
Right now we’re seeing a de-facto disruption, not a sustained, enforced, measurable collapse as defined by the market. So my view is that by March 31 the conditions for a “Yes” don’t seem likely, even if tension and delays remain.
I think the market may be pricing emotion and risk premiums - not the technical trigger needed to resolve YES.
🔍 So I’m leaning NO - shipping will still find some way to move or resume de-facto once immediate pressure fades.












