andy

i’m mid

[andy @andyprv]

Tesla's FSD is already approved in 5 EU countries; more will follow. It's just typical EU bureaucracy

[@levelsio @levelsio]

The French and Germans will ban Tesla FSD to protect their car industry that's being decimated by Chinese EVs and Tesla

Allowing Tesla full self driving would make it so popular that it'd probably the end of their car industry

Because no French or German car has self driving like Tesla, many customers would probably flock to Tesla

To show you: this is the state of self driving cars in France now

My favorite part is the GDPR privacy policy required to be printed on the outside! 😂

[andy @andyprv]

@jackfriks I would go with @ManusAI

Straightforward for non-tech people

@Replit was/is also a good option

They are less daunting than a terminal

[andy @andyprv]

A federal judge just set the price of a book in the AI economy: $3,000.

That's what Anthropic pays ~500k authors after training Claude on pirated copies. $1.5B total.

Here's the part everyone's gonna miss: the court also confirmed training AI on books IS fair use. Anthropic won the actual argument. They got fined for HOW they sourced the data, not for training on it.

Every AI lab on earth just got a free legal roadmap: buy the data legit, you're clear.

Clean data pipelines are about to be a real moat. Wild timeline we're in.

[andy @andyprv]

Down-to-earth version of the post here:

1/ Automated web traffic hit 53% in 2025 (up from 51%) — humans are technically the minority now. But most of that growth is scraping/API bots, not agents transacting with each other. (Imperva 2026)

2/ "Superintelligence exists" has no consensus behind it. Lab timelines for AGI/ASI still span 2027–2045+, and even Altman calls "AGI" a sloppy term. $20/mo buys very capable models — not agreed-upon superintelligence.

3/ Cloud agents handle narrow tasks well. On real multi-step professional work, leading agents still fail 60–80% of attempts, and reliability drops fast past ~35 min of task time. "Run a business unattended from your phone" is a narrow slice today, not the norm. (Mercor, METR)

4/ Voice AI and AI-native apps: real funding (voice sector ~8x'd to $2.1B in 2025, ElevenLabs $500M raise on $330M ARR) and real individual app success stories exist. No verified aggregate data on "kids at $100k MRR" as a broad trend though — treat as anecdotal.

5/ Open-weight models have closed the gap hard: MMLU gap went from ~17.5 pts in 2023 to near zero on knowledge benchmarks; best open models now lag the closed frontier by ~4 months. One of the best-supported claims in the original thread. (Epoch AI)

6/ No data shows keyboards declining, and no independent data yet confirms a mature "agent economy" (agents with logins, paying/vouching for each other). Both are plausible directions, not measured trends.

7/ Usage-based AI pricing is real: GitHub Copilot went usage-based in June 2026, and SaaS margins are compressing as inference costs replace flat per-seat pricing.

8/ Robots: Optimus hadn't started Fremont production as of mid-2026 and missed its 2025 target by ~90%. Musk himself says it's "not doing useful work," still R&D. Real progress exists elsewhere (Unitree shipped 5,500+ units in China) — but it's early pilots, not "solved."

9/ Small-team, huge-revenue outliers are real: Cursor ~$2B ARR/50 people, Midjourney ~$200M/11 people — vs. ~$130–400K revenue/employee for typical SaaS. Genuine, but read with #10.

10/ Most enterprise AI pilots don't pay off: a 2025 MIT study of ~300 deployments found 95% showed no measurable P&L impact. The "do it for me" button exists; reliable business impact from it doesn't, yet, for most companies. (MIT NANDA)

11/ Data monetization, "language barrier solved," SOPs-as-product, agent trust layers — all plausible, none has real market data behind it yet. Translation specifically still shows real error rates on low-resource languages.

12/ Google still sent ~87.6% of search referral traffic as of May 2026; all AI chatbots combined sent ~0.29%, despite usage in the hundreds of millions weekly. Usage growth is real. Traffic capture from Google isn't — yet. (Cloudflare Radar)

13/ AI-native gross margins are actually lower than SaaS, not higher: ~52% vs. 75–85%, because inference is a variable cost that scales with usage. Cheaper per-token ≠ cheaper overall as usage multiplies. Opposite of "it all goes to you." (ICONIQ 2026)

[andy @andyprv]

@levelsio Not all project will be replicated by megacorp. Some of them will not make sense financially. Riches is in the niches is more true than ever

[andy @andyprv]

@levelsio @X How did you get access to your replies for claude? Through official xapi?

[andy @andyprv]

@thejoshuageelen Buy a thermomix and you will thank me later :))

[andy @andyprv]

@aaditsh There are some great ones out there, just from the top of my head I would recommend:

•Churchill by Martin Gilber
•Unbroken by Laura Hillenbrand

The first needs no introduction, and the second is about the Olympic runner Louis Zamperini, his story is unforgettable

[andy @andyprv]

@FreedomDojo Ok Josh, I'll send you a friend request in a moment 😄

[andy @andyprv]

@syncretizm @fortelabs This right here. People tend to eat fewer meals when fasting, which brings calories down.

The simplest and easiest solution for weight loss is calorie deficit. Doesn't matter when you consume those calories, but their number.

[andy @andyprv]

@albadawee @WrongsToWrite Imagine if you also add a goodnight sleep on top of that

[andy @andyprv]

@aaditsh There are a ton of added benefits to having a healthy lifestyle:

- you become more creative
- you are able to focus better and for longer periods of time
- you get mentally sane

Our body is the greatest asset that we have, so we should cherish it more

[andy @andyprv]

@clickup You are doing a great job, keep it up <3

[andy @andyprv]

@AdmiredLeaders @fortelabs Let me give you an example, let say you are building a website, so I would have the following categories:

- development
- content
- marketing

It depends on a case by case, you can go deeper and have sub-categories, dates, etc. For me, this makes it simple to review them later

[andy @andyprv]

@parkerworth Don't focus enough on marketing and have no content strategy

[andy @andyprv]

@fortelabs I am paying attention to what people like and love around me, in order to make useful gifts, but I:

- always enter a gift idea freeze when the time comes
- I have too much information in my head on a day to day basis

The notetaking solution should fix that, thanks!

[andy @andyprv]

@MakinHisWay @WrongsToWrite I would also add that 99% will not be consistent in order to reach success

[andy @andyprv]

@AliAbdaal Easy one, YT and Twitter. Hell, I can ditch even google for these 2

[andy @andyprv]

@fre_notion I think offline functionality is what we are missing

[andy @andyprv]

@thedankoe I have a small kid, with whom we travel since she was 6 months old. Some people don't understand why we do this "she will not remember anything". That's not the point, the point is to expose her to different situations, cultures and adapt to situations