Linas Beliūnas

This is Boris Cherny, the creator of Claude Code. In just 15 minutes, he explains how to actually master Claude Code.

Coupled with the guide, it turns Claude Code into your best engineer: https://lnkd.in/d-MVusTs

One of the best videos available to date. From the person who created the most powerful AI application available today.

P.S. also check out Claude Code Routines: 8 Production Prompts, Real Costs, and Where They Break 🤖: https://lnkd.in/dfFmzpAe

Google just did to commerce what HTTP did to the web - Universal Commerce Protocol, or UCP, is a new open standard for AI agent shopping 😳

Put simply, Google just dropped a new default for how buying works.

UCP is an open, shared language for AI agents, merchants, and payments to transact end-to-end - autonomously 🤖

Discovery → comparison → negotiation → checkout → returns.

All handled by software.

For 20 years, commerce followed the same flow:

Search → ads → product pages → checkout.

UCP breaks that flow. Now it’s:

Intent → agent reasoning → purchase.

No clicks.
No SEO games.
No funnels.

This isn’t theoretical. UCP launched with some of the biggest names in tech & finance:

Shopify, Walmart, Target, Etsy, Wayfair, Visa, Stripe, Adyen 💳

20+ partners on day one. That’s how standards win.

What UCP actually unlocks:

→ AI agents discover merchant capabilities automatically
→ Negotiate prices, bundles, loyalty, subscriptions
→ Execute payments securely
→ Handle post-purchase support
→ All without custom integrations

This kills the N×N integration problem that has crippled e-commerce for decades.

More importantly, UCP doesn’t replace existing rails. It connects them.

↳ A2A → agents talking to agents
↳ MCP → shared context
↳ AP2 → cryptographically signed payment mandates

UCP is the glue for AI-powered commerce.

And Google is uniquely positioned to win this.

Because at the core, commerce = matching intent with supply at scale.

Google already owns:

→ Global search intent
→ The largest product graph
→ AI models (Gemini) embedded everywhere
→ Distribution (Search, Android, YouTube)

UCP turns that into rails, not a walled garden.

Merchants stay merchant-of-record.
Users keep choice.
Agents do the work.

That’s why this can become the HTTP of commerce.

This marks the beginning of the real disruption.

Brands won’t compete for attention anymore.
They’ll compete to be chosen by machines.

Websites become optional.
Checkout buttons become legacy UI.
Loyalty shifts from humans → agents.

And this is the start of non-human commerce.

Quiet.
Autonomous.
Everywhere.

P.S. check out 🔔linas.substack.com🔔, it's the only newsletter you need for all things when Finance meets Technology. For founders, builders, and leaders.

Apple just did the most un-Apple thing in years - it officially picked Google’s Gemini AI for the new Siri, and it might be the smartest move of the AI era 😳

It’s a multi-year partnership where Google’s Gemini AI becomes the brain behind the long-awaited new Siri.

Here’s what that actually means:

→ Apple will pay Google ~$1B per year
→ Apple gets access to a custom Gemini model (~1.2T parameters)
→ That’s ~8x larger than Apple’s current internal models
→ Gemini handles reasoning, planning, summarization
→ Apple handles on-device + privacy-critical tasks
→ All data stays inside Apple’s Private Cloud Compute

Google supplies the horsepower while Apple controls the experience.

Classic Apple 🍎

This is a huge deal because Siri has been stuck in 2014 energy for a decade.

Meanwhile:

↳ ChatGPT became the default “thinking layer”
↳ Google rebuilt its whole business around AI
↳ Anthropic, DeepSeek shipped frontier models at insane speed

Apple was falling behind. Fast.

And building a frontier LLM from scratch isn’t just “hard”. It burns tens - even hundreds - of billions.

Compute. Data. Iteration speed. All costs money.

Google already solved that, so Apple made a call most companies are too proud to make.

Pragmatism > Pride.

And no, this isn’t Apple surrendering.

It’s Apple doing what Apple has always done:

They didn’t invent CPUs.
They curated them → then replaced them.

They didn’t invent modems.
They used Qualcomm → then built their own.

Now it’s AI.

Borrow the best brain now to:

→ Buy time
→ Ship a real product
→ Build your own quietly in the background.

Insiders say Apple’s own trillion-parameter model is coming next.

This is a bridge, not a dependency.

Why did Apple choose Google, not OpenAI, not Anthropic?

Because Gemini is:

↳ Strong in multi-language
↳ Excellent at context + planning
↳ Built for automation at scale
↳ Already proven in production systems

And above all, Apple already trusts Google.

Remember that Google Search has been the default on iPhone for years.

Billions flow between them annually.

And most importantly, this deal signals something bigger:

The hybrid AI era is here.

No single company does everything anymore.

The winning stack now looks like this:

→ Partner for raw intelligence
→ Own UX, privacy, distribution
→ Differentiate where users actually feel it

AI is becoming infrastructure, not a feature.

And Apple just accepted that reality.

Thanks to the $2.11 trillion IPO, more than 4,400 current and former SpaceX employees are now millionaires. 400 of them have stakes that are now worth over $100 million each 😳

To put this into perspective, Google's IPO created roughly 1,000 millionaires.
Facebook's did about the same.

Love him or hate him, but Elon is truly changing the world.

P.S. also check out how I built an AI operating system to run a startup with Claude 🤖: https://lnkd.in/dv7-ZRVc

This Anthropic engineer wrote the monumental blog post "Building Effective Agents". In just 14 minutes, he teaches you how to build AI agents the right way - more than most developers figure out in months on their own.

Coupled with this Claude Managed Agents Guide 🤖, you will be able to deploy production-ready multi-agent systems in a day: https://lnkd.in/daSFNGt7

Pure signal. From the people who actually built it.

P.S. check out How to Build an AI Agent from Scratch (With Working Code) 🤖: https://lnkd.in/dP6r5gHb

Nobody can beat Google. It’s basically the Internet 😳

→ Google owns 14% of Anthropic and 8% of SpaceX.

→ Acquired DeepMind for pennies on the dollar.

→ Runs Gemini AI that’s going to power Apple’s new Siri.

→ Powers Claude with TPU chips.

→ Handles 90% of all the searches.

→ Owns Waymo, while also tracking the world through Maps.

→ Runs YouTube that dominates TV screen time.

→ Owns Android that’s powering 3B +devices.

→ Dominates ads, email, cloud, browser.

→ Leads with advancements in quantum computing.

→ About to put GPUs in space with Starcloud.

→ Just launched UCP to dominate AI-driven commerce.

When you think about it, Google isn’t really the internet…

Google is what the internet runs on.

Google is everything and everywhere.

P.S. also check out deep dive into Google's UCP, the new default for how AI buys things: https://lnkd.in/dSKWjJPy

My life advice for 2026: find someone who looks at you the same way Big 4 consultants look at a Fortune 500 company that needs an “AI Strategy” 😉

Seriously, consulting giants are printing money at will with AI:

→ Accenture has reported billions in generative AI related booking, with some estimates suggesting ~$3B this year alone.

→ KPMG is said to be making up to $1 billion/year from AI consulting.

→ 40% of McKinsey’s work is AI tech, that’s ~$6.4B per year.

→ 20% of BCG’s revenue in 2024 came from AI, that’s about $2.7B/year

The wild part?

Consulting companies are making more profit from AI than 99% of AI startups.
Combined.

AI isn't just hype - it's a goldmine.

And consultancies are digging deep 💸

P.S. for more great stuff, check out 🔔linas.substack.com🔔, it's the only newsletter you need for all things when Finance meets Technology. For founders, builders, and leaders.

Can’t believe Linus Torvalds created Linux at 21 without Claude or AI.

→ He didn't even have a co-founder.
→ No VC funding.
→ No office.
→ No team.
→ Just a personal project

He posted this announcement on Usenet in 1991:

“I’m doing a (free) operating system (just a hobby, won’t be big and professional like gnu) for 386(486) AT clones.”

34 years later, it runs 96% of the world's servers, all Android smartphones, the International Space Station, most of the cloud (AWS, GCP, Azure), every major stock exchange, and basically is the internet’s backbone.

The most important software in history started as someone's side project.

Living legend.

A Netflix engineer just open-sourced a tool that can cut LLM token usage by up to 95% - without changing your code 😳

Headroom, built by Netflix engineer Tejas Chopra, sits in front of tools like Claude, Cursor, Codex, and other agents as a local proxy.

Before your payload hits the model, Headroom compresses the context.

Not by blindly chopping it down.

By using specialized compressors for different payloads:

→ SmartCrusher for JSON
→ AST-based compression for code
→ Tool-output and log compression
→ Local reversible storage of originals
→ Agent wrappers that make it usable without rewriting your app

The headline claim is 60–95% fewer input tokens while preserving answer quality.

The repo has already crossed 42K+ GitHub stars, which says something obvious:

Developers are not just worried about AI getting smarter. They’re worried about AI getting expensive.

Of course, compression is not free magic.

Complex reasoning tasks may punish missing context.

Agent loops may behave differently.
Proxy overhead has to be worth it.
And real-world savings will vary.

But the direction is clear - the next big AI infra unlock may not be a bigger model.

It may be learning how to stop feeding expensive models cheap junk.

Because the cheapest AI inference is the context you never send.

P.S. also check out Loop Engineering: How to Design AI Loops That Build, Ship, and Improve While You Sleep 🔁: https://lnkd.in/dVW6TGiB

Huge: NVIDIA just put a gaming PC, an AI workstation, and a GPU server into a laptop thin enough to disappear in your bag 😳

For 30 years, the PC has been the same thing: Intel or AMD inside, GPU on the side, and hope it doesn't crash.

At Computex, Jensen Huang showed off RTX Spark: an ARM-based AI PC chip built around a Blackwell RTX GPU, a 20-core Grace CPU, and 128GB of unified memory.

And it fits inside a 14mm laptop 🤯

On stage, it ran Forza Horizon 6 and 007 First Light at 100 FPS in 1440p.

On battery.
On Windows.
Without throttling.

The crazy part?

It can run 120 billion parameter AI models on-device.

No cloud.
No API.
No subscription.

That means your AI agent no longer has to live in someone else’s data center.

It can live on your machine.
Always available.
Private by default.
Yours alone.

This is the shift NVIDIA is really betting on: the laptop stops being a thin client for cloud AI and becomes a personal AI workstation.

For developers, founders, analysts, designers, and finance teams, that could change the entire workflow.

The PC used to be a screen with a keyboard.

Now it is becoming the place where your AI actually lives.

P.S. also check out How to Build an AI Agent from Scratch (With Working Code) 🤖: https://lnkd.in/dP6r5gHb

Vibe coders at 2 am building another $3B AI startup 👏😂

Just pure instinct, caffeine, and vibes.

You’re not writing code.
You’re summoning it.

The fans spin up like a jet engine.
The UI half-works.
The animation is buttery… on localhost.

At 3 am, code stops doing what you intended and starts doing what you meant.

Suddenly, everything clicks.

You feel unstoppable. You tell yourself:

Software engineering is basically solved.

Then the API credits run out.

Hard stop. Forced bedtime.

P.S. for more great stuff, check out 🔔linas.substack.com🔔, it's the only newsletter you need for all things when Finance meets Technology. For founders, builders, and leaders.

Wild: Claude AI models will now have invisible watermarks embedded in all text and all code so it can be detected after being copied & pasted 😳

Honestly, this is kinda hilarious when you realise that all of the AI models were trained on a massive amount of content from the internet (that humans generated), while Anthropic itself also trained Claude on millions of human-written books for free.

Scrape the entire internet & books → train AI models → generate content → watermark the output as AI generated.

Tech truly loves irony.

P.S. in other news, check out Y Combinator's QM: The Complete Guide to YC's Open-Source AI Agent Harness 🤖: https://lnkd.in/dqMh63bi

AI just claimed its first major victim 😳

Chegg, the $14.7 billion EdTech giant that charged students for homework answers, study guides, and textbook rentals, has been economically decapitated by AI.

Stock is now down nearly 99% from its 2021 peak. Market cap collapsed to ~$110M.

AI tools like ChatGPT, Claude, Gemini, etc., gave students free, instant, better step-by-step solutions. The entire paywall-for-knowledge model evaporated overnight.

The numbers are just brutal:

→ 2025 full-year revenue: $377M (-39% YoY)
→ Q4 2025 revenue: $73M (-49% YoY)
→ Over 56% of the workforce axed in 2025
→ Core homework/study business is being phased out entirely

They're pivoting hard to “Chegg Skills” (B2B workforce training), which is showing early double-digit growth… but the original Chegg is dead.

AI is eating the world.

P.S. check out how I Turned Claude Cowork Into My Personal COO that does work while I sleep 🧠: https://lnkd.in/eS6JCk2G

100s of AI startups died today: Anthropic just dropped Claude Cowork, and it might be the most important step yet toward AI-native work 😳

Claude Cowork is basically Claude Code, but for non-developers.

You give Claude access to a folder.
You give it a goal.
It plans, executes, fixes mistakes, and keeps going.

No prompts every step.
No copy-pasting between apps.
No babysitting.

Just: “Handle this.” And it actually does 🤖

Here’s what Cowork can already do:

→ Read, write, and reorganize your files
→ Clean inboxes and folders
→ Generate reports from messy spreadsheets
→ Build decks, docs, and summaries
→ Run multi-step workflows across tools
→ Browse the web and pull fresh data
→ Ask clarifying questions when unsure

This is probably the first real glimpse of agentic AI for normal work.

This is literally what Microsoft Copilot and ChatGPT Desktop should have been.

Now here comes the crazy part.

Claude Cowork was 100% built by Claude itself.

Anthropic shipped it in ~1.5 weeks using Claude Code 🤯

AI is now building AI tools to make AI even more useful.

And this marks a turning point.

For years, autonomous agents were:

- Too fragile
- Too risky
- Too technical

Cowork changes that.

→ Runs in a sandboxed VM
→ Human-in-the-loop by default
→ Designed for real work, not demos

Most importantly, Claude Cowork is a preview of something bigger:

An LLM as your operating system.

You don’t open apps.
You describe outcomes.

Claude orchestrates:

Drive → Docs → Sheets → Slides → Web → Back again.

The desktop becomes AI-native.
Apps become implementation details.

And this is exactly what AI looks like right before it becomes unavoidable.

P.S. check out 🔔linas.substack.com🔔, it's the only newsletter you need for all things when Finance meets Technology. For founders, builders, and leaders.

BREAKING: Anthropic’s billionaire CEO just discovered that employees pay their bills with money, not company mission statements 😳

Dario Amodei is reportedly concerned that some new hires are joining Anthropic for the enormous compensation packages rather than its mission to build safe AI.

Which is a little like offering Wall Street bonuses and worrying that applicants noticed the bonus 🙃

As a reminder, Anthropic raised $65 billion in May at a $965 billion valuation.

Amodei’s estimated net worth sits somewhere between $8B and $15.5B.

Meanwhile, Anthropic is competing in the most aggressive talent war Silicon Valley has ever seen - while helping push AI compensation into the stratosphere.

So yes, some employees care deeply about AI safety.
Some care about the mission and the money.

And some may have noticed that landlords still do not accept “responsible scaling policies” as rent.

That does not make them disloyal.

It makes them rational participants in the incentive system Anthropic helped create.

You cannot build a nearly trillion-dollar company, pay like Wall Street, and then expect people to behave like monks.

P.S. also check out The New Enterprise AI Go-To-Market Playbook 📚: https://lnkd.in/dSWExjHf

Thanks to Claude Fable 5, some founders just pulled off the fastest AI pivot in history. From “AI replaced my team” to “culture is our moat” in one weekend is phenomenal 👏😂

Some AI founders are just built different.

P.S. on a more serious note, check out How to Unlock Claude Fable 5 Lite on Claude Opus 4.8 (full Fable 5 system prompt & the step-by-step guide) 🔓: https://lnkd.in/dtUauF5t

Elon Musk just fixed Claude’s biggest problem, while quietly also making the AI race much harder for OpenAI 😳

Anthropic just signed a compute deal with SpaceX to use the full capacity of Colossus 1, its Memphis data center.

300 MW and 220,000+ NVIDIA GPUs coming online within the month 🤯

The immediate result:

→ Claude Code limits are doubling.
→ Peak-hour slowdowns are being removed for Pro & Max users.
→ Opus API limits are going up.
→ 300MW of extra capacity from SpaceX Colossus

Claude had the thing every AI company wants: too much demand.

↳ Developers were using Claude Code hard.
↳ Enterprises were leaning into Opus.
↳ Power users were paying for Max plans.

And still, many kept running into the same wall: not model quality but compute.

So Anthropic went to Elon Musk’s infrastructure stack for help.

Anthropic now gets breathing room.
Claude becomes more usable for developers.
SpaceX turns compute into leverage.

Meanwhile, OpenAI now faces a stronger Anthropic, backed by infrastructure from the one person most motivated to keep the AI market from consolidating around OpenAI.

The model war is loud. The compute war is where the real power is.

P.S. check out how I Turned Claude Cowork Into My Personal COO that does work while I sleep 🧠: https://lnkd.in/gNjUwhXu

Europe is finally rising: Ursula von der Leyen officially just announced EU Inc., a single, digital-first company structure for all of Europe 😳

Not 27 countries.
Not 27 rulebooks.
One startup Europe.

The coolest part?

This didn’t come from a think tank. It started as a grassroots petition in 2024.

13,000+ founders and investors signed it. Europe listened 👏

Here’s what EU Inc. actually does:

→ Incorporate online in 48 hours
→ No mandatory notaries
→ One legal entity, valid across all 27 EU states
→ Standardized rules for equity, insolvency, and cross-border ops

Taxes and labor laws stay national while the corporate layer goes European.

That distinction matters because Europe’s biggest problem was never talent.

It was friction.

Until now, scaling meant paperwork hell:

→ New entities per country
→ Incompatible equity frameworks
→ Legal bills eating seed rounds alive

So the founders did the rational thing.

They left.
Delaware won.
Europe lost compounding.

EU Inc. flips that equation. And for founders, this is massive:

→ Faster launches
→ Cheaper compliance
→ Founder-friendly equity structures
→ US-style fundraising without leaving the EU

A founder in Vilnius or Lisbon now has frictionless access to a 450M-person market on Day 1.

No relocation required.
No regulatory gymnastics.

Most importantly, this is the first time Europe chose builders over bureaucracy.

If executed as a true regulation (not 27 interpretations), EU Inc. could unlock:

→ More European unicorns
→ Less brain drain
→ Stronger AI, biotech, and climate tech champions
→ A real Single Market, not a theoretical one

Of course, execution will be the key here, but one thing is already clear:

Europe is finally building for founders 🇪🇺🚀

P.S. check out 🔔linas.substack.com🔔, it's the only newsletter you need for all things when Finance meets Technology. For founders, builders, and leaders.

RIP investment bankers? Anthropic just pushed Claude deeper into finance with new AI plugins in Cowork 😳

Claude Cowork just rolled out new finance-specific plugins that go far beyond just an AI assistant.

This is real workflow automation:

Model → memo → deck.
Excel → PowerPoint.
Data → decision.

In one continuous session.

Claude is no longer just summarizing finance. It’s operating inside it.

What’s new:

→ 5 dedicated plugins: financial analysis, investment banking, equity research, private equity, wealth management.
→ Direct institutional data access via FactSet and MSCI.
→ Partner integrations from S&P Global and London Stock Exchange Group.
→ Cross-app memory: update a model in Excel, auto-refresh the slide deck in PowerPoint.

In demos, tasks that used to take analysts days - parsing earnings, updating comps, drafting research notes - were completed in minutes 🤯

While impressive, this won’t eliminate bankers. At least not yet.
Judgment, negotiation, trust - still human, and still critical.

But the copy-paste layer of Wall Street?
The formatting marathons?
The midnight model updates?

That layer is getting automated fast.

The old edge was all about stamina.
The new edge is all about orchestration.

If you run a fund, a bank, or build for them, this isn’t incremental.

It’s a cost structure rewrite hiding inside a plugin update.

The real question now is not whether AI replaces you.

It’s whether you redesign the workflow before someone else does.

P.S. check out how I built an AI operating system to run a startup with Claude 🤖: https://lnkd.in/dv7-ZRVc

The age of KYC video verification is officially over: the latest video models are now so real that nobody can tell they’re AI generated 😳

This is the moment identity verification finally broke.

For years, video was the final proof layer.

→ “Turn your head”
→ “Blink twice”
→ “Smile for the camera”

Banks trusted it.
FinTechs built on it.
Digital platforms enforced it.

In 2026, it fails effortlessly.

With a single reference clip, tools like Kling AI can now generate:

→ Perfect facial expressions
→ Natural micro-movements
→ Realistic timing and emotion
→ Full liveness… without a human

And here’s the uncomfortable part.

If an AI can pass a live video call, video KYC is no longer verification.

It’s a theatre.

We’ve officially entered a world where:

Seeing ≠ believing
Liveness ≠ human
Face ≠ identity

So the next big shift is from visual trust → systemic verification

↳ Behavior, not faces
↳ Signals, not selfies
↳ Patterns, not pixels

Zero-trust identity becomes the default.

And in this new world, AI-native risk infrastructure is no longer optional.

AI Risk Decisioning platforms like Oscilar don’t ask “Does this video look real?”

They ask:

→ Does this behavior make sense?
→ Do signals align across sessions?
→ Does intent match history?

That’s how you fight AI fraud. With AI 🤖

Because deepfakes are no longer just memes. They’re:

↳ Identity hacks
↳ Scam engines
↳ Financial attack vectors

AI broke video verification. AI will also defend what comes next.

Seeing is no longer believing.

P.S. check out 🔔linas.substack.com🔔, it's the only newsletter you need for all things when Finance meets Technology. For founders, builders, and leaders.

I just gave Claude Fable 5 my portfolio & told my AI trader:

“Turn this into $10M. Make no mistakes.”

It wrote 69 strategies.
Generated 420+ research reports.
Spun up 21 new algos.

It scraped X, Reddit, Instagram, and LinkedIn.
Mapped sentiment.
Read every chart pattern it could find.

It traded 24/7.

It lost everything.

But boy was it beautiful.

P.S. on a more serious note, check out Loop Engineering: How to Design AI Loops That Build, Ship, & Improve While You Sleep 🔁: https://lnkd.in/dVW6TGiB

For two years, everyone said Apple was losing the AI race. Turns out Apple might have been playing a completely different game the whole time 😳

While the industry burned $1.4 trillion+ training frontier models, Apple quietly waited.

Then, in January 2026, it made a very Apple move - partnered with Google to power Siri with Gemini AI.

Not “build the biggest model.”
Pick the winner. Integrate it everywhere 🤖

And tomorrow, Apple flips the switch.

AI Siri → 2.5 billion devices.

While everyone else is fighting for AI users, Apple already owns the devices on which those users live.

The contrast is quite wild:

→ $1.4T+ spent across the industry training models since 2023
→ Apple pays ~$1B/year to integrate the best one

→ Rivals chase subscriptions ($20-$200/month)
→ Apple runs AI on-device: faster, cheaper, private

→ Startups race to build AI hardware
→ Apple ships iPhone, iPad, Mac, Vision, Watch powered by its own AI chips

→ Competitors fight for distribution
→ Apple flips a switch across 2.5B devices

Sure, Apple didn’t win the LLM race.

But the AI model race might not have been the real game all along.

The real game is distribution + hardware + ecosystem.

And in that game, Apple starts the match with billions of users already logged in.

Turns out “being late to AI” might have been the most Apple strategy imaginable.

Most importantly, the AI era may not be decided by who builds the smartest model.

It might be decided by who owns the device that model runs on.

P.S. check out how I Turned Claude Cowork Into My Personal COO that runs work while I sleep 🧠: https://lnkd.in/eS6JCk2G

Epic: Somebody created an AI-generated ad portraying 2036 where humans pedal at Energym gyms to power AI after robots displace 80% of jobs 👏😂

Co-founded by Elon Musk, Jeff Bezos, and Sam Altman.

Brilliant.

h/t AiCandy

P.S. for more great stuff, check out 🔔linas.substack.com🔔, it's the only newsletter you need for all things when Finance meets AI. For founders, builders, and leaders.

This is one of the best AI hacks I’ve seen 👏😂

If you ever hit your Claude session limit, just run this prompt:

“Based on our conversation so far, create a summary so that ChatGPT can understand it clearly”.

😭😭😭

For a more nuanced solution, check out The System for Never Hitting Claude's Limits 🤖: https://lnkd.in/gNbWTuyN

Wild: Amazon staff reportedly found the easiest way to boost AI usage - use AI for tasks that didn’t need AI in the first place 🫠

According to the FT, some employees used an internal AI tool to automate non-essential work so managers could see they were “using AI” more often.

This is Goodhart’s Law with a chatbot:

When a measure becomes a target, it stops being a useful measure.

→ Track AI usage, and people will manufacture usage.
→ Reward prompts, and people will create prompts.
→ Celebrate adoption, and teams may optimize for dashboards instead of better work.

This is the trap for every company proudly reporting AI rollout metrics.

Measure the work removed, not the prompts added.

Because incentives always drive behavior.

P.S. if you really want to leverage AI, check out how I Turned Claude Cowork Into My Personal COO that does work while I sleep 🧠: https://lnkd.in/gNjUwhXu

Anthropic just launched Claude for Small Business, and it basically turns Claude into an AI employee for companies too small to hire one 😳

Here's what it can do already:

→ Manage invoices, payments, and finances
→ Create campaigns, designs, and content
→ Organize sales and customers automatically
→ Read, summarize, and draft documents
→ Manage emails, calendars, and files
→ Execute tasks across multiple apps

All from Claude.

The product plugs into tools SMBs already use: QuickBooks, PayPal, HubSpot, Canva, DocuSign, Google Workspace, Microsoft 365, and more.

Then Claude can run multi-step workflows across them.

The coolest part?

Claude for Small Business can understand the context of the business and execute work across the stack.

Anthropic has already shipped:

↳ 15 agentic workflows
↳ 15 repeatable skills
↳ Finance, sales, marketing, docs, and admin automations
↳ No engineers required
↳ Human approval before anything gets sent, paid, or posted

It’s clear now that Anthropic doesn’t want Claude to be another tab inside the business.

It wants Claude to become the OS for every small business.

The place where invoices get chased.
Campaigns get created.
Customers get organized.
Contracts get reviewed.
Files get summarized.
Tasks get finished.

So you won’t be opening 10 different apps. You will be just asking Claude 🤖

ClaudeOS is eating the world.

P.S. also check out how I built an AI operating system to run a startup with Claude 🤖: https://lnkd.in/dv7-ZRVc

This is brilliant! An engineer used Claude Mythos & data from Whoop to find which coworker gives him the most stress 👏

A developer hooked his Whoop tracker to his work calendar to track his stress.

Thanks to Fable 5, he then reverse-engineered Whoop to pull per-minute heart rate and matched spikes with calendar events and attendees.

He now has a leaderboard of which colleagues cost him the most beats per minute 😂

These are the ideas that give me hope that AI will be worth it in the end.

P.S. check out Loop Engineering: How to Design AI Loops That Build, Ship, and Improve While You Sleep 🔁: https://lnkd.in/dVW6TGiB

Epic: Someone built a Google Translate but for LinkedIn 👏😭

Whoever made this deserves a massive raise.

Absolutely brilliant.

BTW, on a serious note, in case you missed it, check out Andrej Karpathy’s Method To 10X Your Claude Skills 🧠: https://lnkd.in/dAg2yXVJ

This is Anthropic’s Applied AI Team. In just 15 minutes, they show you how to actually prompt Claude properly so it becomes your thinking partner.

Pure signal. From the people who actually built it.

P.S. check out how to Turn Claude From a Chatbot Into a Thinking Partner 🧠: https://lnkd.in/eDGHRw94

Claude AI is eating the world: Microsoft just shipped its “digital coworker” for 365, and the brain behind it isn’t OpenAI. It’s their biggest rival Anthropic 😳

Microsoft CEO Satya Nadella just unveiled Copilot Cowork - an autonomous agent that can plan and execute multi-step work across Outlook, Excel, PowerPoint, Teams, and your entire Microsoft 365 stack.

You give it intent: “Prepare me for the Q2 review with Acme.”

It builds the plan.
Pulls the data.
Writes the deck.
Schedules prep time.
Drafts the emails.

And runs the workflow in the background.

The agentic orchestration layer powering it comes from Claude Cowork, Anthropic’s technology.

That makes this one of the biggest enterprise wins the $380 billion AI lab has landed so far 🤯

It could change the enterprise AI game significantly thanks to:

↳ Massive distribution

Microsoft 365 has hundreds of millions of seats. Copilot already has 15M+ paid users and 90% of the Fortune 500.

↳ Enterprise trust unlocked

Claude’s agentic tech is now wrapped in Microsoft’s security stack: Purview, Entra, audit logs, and compliance.

↳ Strategic signal

Microsoft’s deepest AI partner is OpenAI (it has 27% stake).
Yet the system executing long-horizon work plans now runs Anthropic’s stack.

↳ A learning flywheel

Every enterprise workflow running inside Copilot now trains Claude on real office work.
This alone could be a huge unlock for real-world workflows.

My takeaway is this:

The biggest shift we’re witnessing now is not about better chatbots or the latest LLMs.

It’s all about software that plans and executes knowledge work while you watch the progress bar 🤖

Anthropic built the AI agent that made Microsoft nervous.

So Microsoft turned around and shipped it to the corporate world.

Now the operating system of global office work runs on Claude AI.

Anthropic just can't stop winning, & ClaudeOS is eating the world.

P.S. check out how to Turn Claude Cowork Into Your Personal COO that does work while you sleep 🧠: https://lnkd.in/eS6JCk2G

Some folks on LinkedIn are just built different 😂

Full-time Subscriber at OpenAI (ChatGPT) for 2+ years… then promoted to Full-time Subscriber at Anthropic (Claude) 😭

Throw in “Founder at localhost:3000” and the profile is actually complete.

This is the AI version of VCs listing every single investment they’ve ever touched.

The LinkedIn experience section is truly undefeated.

P.S. on a more serious note, check out How to Build a Second Brain with Claude Fable 5 🧠: https://lnkd.in/dJheeepD

It’s officially over for America: The President of the European Commission just announced that Europe wants to become “the first AI Continent” 😳

To achieve that, they are going to invest up to €10 billion and build European AI Gigafactories.

For the perspective, OpenAI & NVIDIA alone are in talks to invest $250 billion for ONE data center campus in Ohio.

Europe is not a serious place for AI.

P.S. also check out How to Build an Agentic OS with Claude Fable 5 🤖: https://lnkd.in/dZsST5qR

This is Peter Steinberger, and his story is insane 😳

- Sells his first company for $100M+
- Spends 3 years in existential crisis
- Becomes jacked
- Comes back from retirement
- Vibe-codes 43 failed projects
- Project 44 is ClawdBot
- Goes viral
- Anthropic sends you trademark lawsuits
- Renames to MoltBot
- Crypto scammers hijack your accounts in seconds
- Secret rebrand to OpenClaw 
- Hits 180K GitHub stars
- Gets acquired by OpenAI

No VC funding.
No 100+ person team.
Just a solo founder + shipping.

Probably one of the wildest tech & AI stories of all time.

Legend.

P.S. check out how I built an AI operating system to run a startup with Claude 🤖: https://lnkd.in/dv7-ZRVc

AI’s Instagram moment? Meta Platforms just bought an autonomous agents startup Manus AI for more than $2 billion, making it Meta’s 3rd largest M&A of all time 😳

The deal is a clear signal of where AI is heading next.

Manus built an AI that executes work, not just conversations.

It runs inside a sandboxed virtual computer with memory, tools, internet access, and the ability to write and run code.

You give it a task → It completes it end to end.

That capability changes the category.

Manus quickly stood out because it could handle real workflows:

→ Market research
→ Portfolio analysis
→ Job candidate screening
→ Trip planning
→ Full application builds
→ Business process automation

Tasks that normally require hours of human coordination could be completed with minimal supervision.

But the strongest signal wasn’t the demos.

It was traction.

Manus crossed $100M in ARR in under 8 months through a straightforward subscription model.

Millions of users.
Heavy usage.
Real willingness to pay.

At a time when most AI products prioritize growth over revenue, Manus proved autonomous agents can sustain a business.

For Meta, the timing is deliberate.

The company has already committed tens of billions to AI infrastructure and open-source models like Llama.

What it needed was a layer that turns intelligence into action.

Manus thus enables:

↳ Agents that manage campaigns and operations for businesses
↳ Agents that coordinate scheduling and communication inside WhatsApp
↳ Agents that execute shopping and workflows on Instagram
↳ AI systems that behave more like persistent operators than reactive assistants

This creates a path from “AI feature” to “AI worker.”

The Instagram parallel seem to fit perfectly here.

→ Instagram changed how people behaved on mobile, then scaled globally through Meta’s distribution.

→ Manus has similar characteristics: viral adoption, practical utility, and early monetization but limited reach as a standalone product.

Meta’s 3 billion-user ecosystem removes that constraint.

There’s also a broader undercurrent.

Manus originated in China, relocated to Singapore, and outperformed many Western systems on real-world agent benchmarks.

The most capable AI teams are no longer concentrated in one geography.

More importantly, acquiring Manus pulls execution-grade agent technology directly into a major Western platform.

My takeaway is this:

This acquisition marks a shift.

→ AI is moving from assistance to execution.
→ From answering questions to completing work.

If Meta executes well, autonomous agents won’t feel novel or visible.

They’ll quietly become part of how work and coordination happen at scale.

And that’s exactly how platform shifts usually begin.

P.S. check out 🔔linas.substack.com🔔, it's the only newsletter you need for all things when Finance meets Technology. For founders, builders, and leaders.

Mira Murati, the woman who built ChatGPT, then left OpenAI and stayed silent for a year, just launched something that could change forever how you use AI in your day-to-day life 😳

And no, it is not another chatbot.

Thinking Machines Labs just previewed a new kind of AI model built around one idea:

Today’s AI is too turn-based.

You type.
It waits.
It answers.
You wait again.

That is fine for prompts. But it is terrible for real work.

Because real work is messy.

You interrupt.
Point at the screen.
Change your mind mid-sentence.
Ask one thing while looking at another.
Need the system to notice hesitation, not just parse commands.

That is the gap Murati’s team is going after.

Their new model, TML-Interaction-Small, is trained to handle audio, video, and text natively in real time.

Not as a stitched-together stack of speech-to-text, vision, LLM, voice, and tool use.

The interaction itself is part of the model.

The numbers are pretty wild:

↳ 276B MoE model, with ~12B active parameters
↳ 200ms “micro-turns” instead of rigid back-and-forth turns
↳ 0.40s turn-taking latency, according to their benchmark
↳ Can listen, see, speak, search, use tools, and generate UI while the conversation continues
↳ Demos include live translation, workout rep counting, proactive video commentary, and chart generation mid-chat

And the important part here is not “faster voice mode.”

It is the architectural bet.

Thinking Machines is applying Rich Sutton’s Bitter Lesson to the AI interface:

→ Stop hand-engineering the conversation layer.
→ Train the model to collaborate natively.

Of course, it’s still early.

It is a research preview.
Long sessions, reliability, privacy, safety, and real-world latency all need proving.

But the direction is clear:

The next AI race may not be won by the model that answers best.

It may be won by the model that feels least like software.

And somehow, the company making that bet first is not OpenAI, Google, or Anthropic.

It is the one built by the woman OpenAI let walk away.

P.S. check out Top 10 AI Startups to Watch in 2026 (& their pitch decks) 🤖🦄: https://lnkd.in/dY7b4kT4

My Claude Fable 5 just reached a whole new level of intelligence 👏😂

Right before Anthropic is moving its most powerful AI model ever to paid usage credits…

If this isn’t AGI, I don’t know what is.

P.S. on a serious note, check out Exactly What to Build with Claude Fable 5 Before the Free Window Closes ⌛: https://lnkd.in/dUpGZFE2

Nailed it! Joanna Maciejewska gives the only perspective on AI that matters 👏

That's exactly why Elon Musk has been building Tesla Optimus, Figure and 1X are developing humanoid robots, and even Google & Apple are doubling down on robotics.

The next 2-3 years will be wild.

P.S. check out 🔔linas.substack.com🔔, it's the only newsletter you need for all things when Finance meets Technology. For founders, builders, and leaders.

Can’t believe Anthropic quietly dropped a 33-page blueprint for turning Claude into your operating system - and gave it away for FREE. You can now program Claude without writing a single line of code 🔥

It’s called Skills, which is a folder that teaches Claude how you work.

The best AI users today don’t prompt better.
They build systems that prompt for them.

Here’s why skills are so powerful:

→ A Skill is just a folder with one required file: SKILL dot md
→ A tiny YAML header tells Claude when to activate it (specific trigger phrases matter)
→ Instructions load progressively, so you don’t burn tokens every session
→ You can build one in 15-30 minutes using the built-in “skill-creator” meta-skill
→ Teams can deploy skills org-wide for consistent workflows

Think of it like this:

MCP gives Claude the tools.
Skills give it the recipe.

Without skills:

↳ You re-explain your budget planning flow every week
↳ 15 back-and-forth messages
↳ Inconsistent outputs

With skills:

↳ Automatic activation
↳ Fewer tokens
↳ Fewer retries
↳ Same workflow, every time

And that’s one of the biggest value unlocks for Claude AI today.

The best part?

You can now improve your Skills on autopilot & make them 10X better using Andrej Karpathy’s Method 🧠: https://lnkd.in/dAg2yXVJ

We’re moving from “clever prompts” to encoded workflows.

Claude is no longer just answering questions.
It’s running playbooks.

If you use Claude daily and don’t have at least one Skill, you’re still operating manually.

P.S. also check out how I built an AI operating system to run a startup with Claude 🤖: https://lnkd.in/dv7-ZRVc

Cannot believe Linus Torvalds built Linux at 21 without Claude or AI.

→ He didn't even have a co-founder
→ No VC funding
→ No fancy office
→ No team
→ Just a personal project

He posted this announcement on Usenet back in 1991:

“I’m doing a (free) operating system (just a hobby, won’t be big and professional like gnu) for 386(486) AT clones.”

34 years later, it runs 96% of the world's servers, all Android smartphones, the International Space Station, most of the cloud, every major stock exchange, and basically is the internet’s backbone.

The most important software in history started as someone's side project.

Living legend.

P.S. also check our The Founder's Playbook for Winning in the AI Age 📚: https://lnkd.in/dtwjqH4W

Turns out, Ben Affleck has a better understanding of the limitations and issues in AI than 99% of the VCs that are actively investing in the space 😳

"A lot of that rhetoric comes from people who are trying to justify valuations around companies where they go, 'We're gonna change everything in two years, there's going to be no more work.'"

And then Affleck said the quiet part out loud:

AI doesn’t create. It averages.

And averaging is the enemy of originality.

The most misunderstood part is that it’s not anti-AI.

That’s an accurate description of how LLMs work.

They compress the past.
They smooth the edges.
They converge toward the mean.

Which is great for:

→ Summaries
→ First drafts
→ Busywork
→ Tedious cognitive labor

And terrible for:

→ Voice
→ Taste
→ Risk
→ Saying something new

Creativity and judgement have never been more important.

P.S. also check out 120+ Real AI Use Cases From Leading Startups & Internet Companies: https://lnkd.in/drT27Bvb

This is Boris Cherny, the creator of Claude Code. He just sat down at the Sequoia AI session, and in 15 minutes, explained how exactly he uses Claude Code himself & revealed his entire coding setup.

"100% of my code is written by Claude Code. I run around 100 agents at one time" - Boris Cherny

→ 100% of the code written by AI.
→ 100 agents running simultaneously 🤖

Coupled with the guide, it turns Claude Code into your best engineer: https://lnkd.in/d-MVusTs

One of the best videos available to date. From the person who created the most powerful AI application available today.

P.S. also check out The Complete Claude /goal Guide for AI Agents (& turn Claude into a 24/7 autonomous employee) 🤖: https://lnkd.in/dXZHxV_w

Wild: Moonshot AI just dropped Kimi K3, a 2.8 trillion-parameter model that nearly matches GPT-5.6 & Claude Fable 5. And it plans to open-source it in 10 days 😳

The Chinese AI startup combined frontier-level capability, a 1M-token context window, native multimodality, and aggressive pricing in one model.

→ It’s now No. 1 in the Frontend Code Arena, surpassing Claude Fable 5.
→ 88.3 on Terminal-Bench 2.1, versus 88.8 for GPT-5.6 Sol
→ No. 1 on Arena’s WebDev leaderboard at launch
→ 6.3x faster decoding across million-token contexts
→ $3 per million input tokens and $15 per million output tokens

That means Kimi K3 can ingest an entire codebase, reason across screenshots and video frames, and run long agent workflows without paying the full compute cost of 2.8T parameters on every token.

And that’s pretty huge.

DeepSeek showed Chinese labs could compress the cost of intelligence.

Kimi K3 suggests they can now compete on capability, context, price, and openness at the same time.

Of course, there are caveats.

The weights are not public yet, several benchmarks are vendor-reported, and one independent evaluation found a ~51% hallucination rate.

But if Moonshot delivers on its promise, thousands of developers will be able to fine-tune, audit, self-host, and build agents on top of a near-frontier model.

Open source AI is eating closed AI.

P.S. also check out The Full Guide to GLM-5.2: The ChatGPT Moment for Local AI 🤖: https://lnkd.in/d__8NVYt

Can’t believe Anthropic just told founders exactly what AI products to build in 2026. Here’s a full founder playbook + where Anthropic's own products will and won't compete 🔥

Anthropic just analysed 1 million Claude conversations and found that ~6% were people asking for personal guidance:

Health.
Money.
Careers.
Relationships.
Parenting.
Legal rights.

These are the problems people usually take to a doctor, therapist, lawyer, financial advisor, or coach.

It’s basically product-market fit hiding in the raw usage data, and Cal AI proved the wedge.

One tiny subtopic - calories + macros - became a $40M+ revenue business, crossed $50M ARR, got acquired by MyFitnessPal, and did it with 7 employees.

So I mapped the full opportunity.

Inside:

↳ The 9 consumer AI domains Anthropic’s data points to
↳ The highest-signal subtopics founders can build around
↳ Where Anthropic is building internally
↳ Where Anthropic likely will and won’t compete
↳ Funded startups already attacking each category (& their pitch decks)
↳ Why thin wrappers are dead
↳ The 2026 founder playbook for trusted AI guidance products
↳ Bonus: AI OS to run your startup + Top 100 seed investors

Comment "AI Playbook", and I’ll send you the link 🦄

P.S. also check out How to Turn ChatGPT Images 2.0 & Claude Design Into Your Chief Designer 🫟 (& build pitch decks, app prototypes, landing pages, etc.): https://lnkd.in/dcExymFM

Teaching robots how humans work so machines can replace them 🤖

At first, I thought this was AI generated…

Turns out, it’s 100% real and was happening in a Shenzhen factory this year.

Surreal.

P.S. check out 🔔linas.substack.com🔔, it's the only newsletter you need for all things when Finance meets Technology. For founders, builders, and leaders.

This is brilliant! Apple might have just solved the biggest problem in AI 👏🏼😂

iClaw - $199 metal claw to keep your MacBook lid open so your AI agents can safely run 24/7 😭

Some people actually searched the Store for it.

On the other hand, Apple would probably make more money selling this than 95% of AI startups…

P.S. also check out How to Build an Agentic OS with Claude Fable 5 🤖: https://lnkd.in/dZsST5qR

End of an era at Apple: Tim Cook is stepping down as CEO on September 1. He turned a $350 billion company into a $4 trillion tech giant 😳

To put this into perspective, the $3.65T added under Cook is bigger than the entire stock markets of the UK, France, and Germany. Combined.

SVP of Hardware Engineering John Ternus will be replacing Tim.

For those who don’t know, Ternus is the engineer behind iPhone, Mac, Watch, and Vision Pro hardware.

This signals Apple's bet on deep technical leadership for future product innovation and (hopefully) more on-device AI.

Steve did the Jobs, Tim Cooked, and now John might Ternus around.

P.S. check out The Ultimate Guide to Claude Managed Agents (with step-by-step build walkthrough) 🤖: https://lnkd.in/daSFNGt7

A border officer won’t accept a photo of your passport. Your bank builds identity decisions from one every day 😳

That's the foundation of online identity verification in 2026. A process designed for in-person checks, duct-taped onto the internet.

And it's breaking.

Deepfakes now pass liveness checks.
Synthetic identities clear onboarding flows.
The selfie-plus-document model is done.

Governments see it. That's why they're building something different:

→ 36% of Europeans already use government-backed eIDs
→ 80% of EU citizens will carry a Digital Identity Wallet by 2030
→ eIDAS 2.0 mandates banks, insurers, and telecoms to accept them

But here's the problem nobody talks about: there are 150+ eID schemes worldwide.

Different standards. Different assurance levels. Different specs.

A fintech operating across 10 EU markets needs 10 separate integrations.

That doesn't scale.

This is exactly where payments were before Visa and Mastercard built the network layer.

Identity needs the same thing. That's what Hopae is building - a single integration into 100+ government-backed digital identity schemes globally.

The Visa layer, but for identity.

They built a 2-minute assessment to get your personalised compliance readiness score instantly.

Get your score here: https://lnkd.in/djCqip95

Most organisations think they have until 2027 to sort this out. They're wrong. The deadline is 2027, but the infrastructure decisions are happening right now.

Payments got their network layer, and it changed everything.

Identity is next.

The only question is whether you're building on it or scrambling to catch up.

Wild: All of Cursor's founders - 25-26 years old MIT dropouts - are now billionaires following SpaceX's $60 billion acquisition 😳

Their story is one of the wildest ones in tech ever:

→ 4 MIT students start a side project in 2022
→ They build the AI coding tool developers love
→ Hit a $10B valuation
→ Decide copilots aren’t enough, so they move into models
→ Need massive compute to compete with OpenAI & Anthropic
→ Meanwhile xAI is losing the coding race
→ Elon realizes catching up could take years, so he skips the line & buys Cursor for $60B
→ Cursor gets compute, while xAI gets the coding leader
→ Founders become multi-billionaires; the first 50 hires ~$20-500M each

This is the beauty of startups at its finest.

P.S. check out how I built an AI operating system to run a startup with Claude 🤖: https://lnkd.in/dv7-ZRVc