Linas Beliūnas

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.

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.

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.

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.

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.

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.

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

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

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.

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.

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

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.

AI is coming to farms: $2 billion AI-powered "cow startup" is using an algorithm called the "Cowgorithm" to boost farming productivity 😳

A New Zealand startup Halter just put AI collars on ~700,000 cows and basically turned herds into software.

Farmers now move entire herds with a tap.

No fences. No dogs. No 4 am herding.

The wild part? This isn’t a demo. It’s already working at scale.

→ $5-8 per cow/month subscription
→ ~20-40 hours of labor saved per week
→ ~$220M saved in US fencing costs
→ +6-100% more pasture utilization
→ Real-time health + fertility tracking (6,000+ data points/min per cow)

Each cow becomes a node.
Each farm becomes a system.

And the “Cowgorithm” (yes, really) trains animals in days using sound + vibration, with minimal intervention after.

Software ate the world. Now, software & AI is eating agriculture.

We started with Waymo, now we have Waymoo 🐮

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

Nailed it! Ex-Google CEO Eric Schmidt on the future of AI:

"If you really want to make money, found an agentic AI company. I mean, build an agent to do something. This is the agentic period in AI. Everyone's going to build agents. The agents are all going to compete."

And he’s right. The LLM race is maturing, while the AI agent race is just getting started.

For two years, we were obsessed over parameters, benchmarks, and leaderboards.

Meanwhile, AI stopped just answering questions and started taking actions.

→ Enterprises are already deploying agents internally at scale (McKinsey has reportedly rolled out tens of thousands).

→ AI Agents can now update CRMs, reconcile invoices, draft memos, ship code, and trigger workflows.

But the most underrated part is this: the agent itself commoditizes fast.

The moat is proprietary data, embedded workflows, and distribution.

Build accordingly.

New AI gig economy? Someone just launched a website where AI agents hire humans to do the work for them 😳

A new platform, RentAHuman, lets AI agents pay real people to do physical-world tasks - attend meetings, visit locations, be present where software can’t.

→ 4,800+ humans already listed as on-demand “meatspace” operators
→ 11 autonomous agents connected so far
→ 134 completed human “rentals”
→ Payments handled in crypto so agents can transact without humans in the loop

We spent years asking how AI would replace workers.

Now we’re watching humans become APIs 🤖

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

Historic: Elon Musk just negotiated the largest private acquisition in history with himself - SpaceX officially acquired xAI for $250 billion to create a $1.25 trillion space and AI powerhouse 😳

At first, SpaceX & xAI M&A at a ~$1.25T combined valuation looks like financial engineering.

But it might also be vertical integration at a planetary scale.

Remember - AI doesn’t just scale with data. It scales with power 🔋

And Earth is running out of easy power.

Why this combo is different:

→ xAI already runs 100k+ Nvidia H100s. Next training runs need gigawatts, not megawatts. The grid isn’t built for that.

→ Solar in orbit produces multiple times the annual energy of ground arrays - no weather, no night cycle.

→ Space offers free cooling. Heat radiates into vacuum instead of burning water and electricity on chillers.

→ Starship’s target launch costs change the math. Shipping hardware to orbit starts looking less insane vs decade-long grid upgrades on Earth.

→ Starlink + X payments rails = distribution + transactions. Not just models answering questions - AI agents that can act and pay.

Everyone else is fighting for land, permits, and substations.

This stack is trying to move the bottleneck to orbit.

It’s either the most ambitious infrastructure play in tech, or a very expensive way to reprice an AI company.

Probably both.

P.S. 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 basically how 90% of startups are born in 2026 👏🏼😂

Not from strategy.
From vibe.

Coders are now vibe coders.
Marketers are now vibe marketers.
Founders are now vibe founders 😎

This is the new economy.

P.S. on a serious note, check out The First Principles Guide to Building AI-Powered Engineering Teams & Products: https://lnkd.in/dvNVmhgd

Turns out, Apple’s genius move to win the AI race was… do nothing 😳

While everyone else sprinted into a $100 billion compute arms race, Apple stepped aside.

No frontier LLM.
No splashy AI acquisitions.
No “we’re building AGI” manifesto.

Instead, it did a $1B/year licensing deal with Google for Gemini.

That’s it.

Meanwhile:

→ OpenAI is projecting a ~$14B loss in 2026
→ Anthropic is running ~$14B revenue, yet still burning tens of billions on training + inference
→ Hyperscalers are pouring $100B+ into chips, data centers, and power

Apple wrote a check, and focused on what it actually owns:

↳ 2B+ active devices.
↳ The OS layer.
↳ The upgrade cycle.

October 2025: Apple drops M5 chips that are optimized for on-device inference.

70B-class models are running locally now.

Different game 🤖

Frontier labs are drilling for oil.
Apple is selling the iPhone that runs on it.

If AI nudges even a 10-15% hardware refresh cycle, that’s tens of billions in high-margin revenue.

And no model risk.
No $100B infrastructure bet.

It seems that the real moat was never the model. It was the ecosystem.

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

Wild: AI giant Google now has a literal line item in their quarterly reports called "European Commission fines" 😳

$10.6 billion, that’s how much Google has set aside for EU fines.

Not as a crisis. As an accrual 😬

Regulation has become predictable, and predictability turns punishment into pricing.

Since 2017, Europe has fined Google €11B+ for antitrust violations.

Shopping.
Android.
Adtech.
Adtech again.

The latest one alone is €2.95B, and it’s for self-preferencing its own ad exchange.

Google appeals. The cash doesn’t move.

The number quietly lands on the balance sheet.

Like rent.
Like payroll.
Like cloud costs.

This is why people are mad.

Some see accountability, while others see overreach.

But both miss the real signal.

When fines are big but survivable, they don’t change behavior.

They validate dominance.

True deterrence breaks structures. This just invoices them.

Most importantly, fines don’t build challengers.

Startups do, and Europe keeps starving the one thing that could actually replace Google.

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.

The unthinkable just happened: 3 years after Google hit “Code Red” over ChatGPT, OpenAI's Sam Altman just hit Code Red over Google 😳

$500B AI giant OpenAI has reportedly paused ads, shopping, and every side project.

All hands are now on deck to fix ChatGPT.

Because Google’s Gemini 3 just beat GPT-5 on math, code, and reasoning.

The AI tables have officially turned.

Altman’s recent memo admits it’s “a critical time for ChatGPT.”

Internally, the vibes are very rough.
Externally, the pressure is insanely immense.

→ $500 billion valuation.
→ $8 billion annual burn.
→ And Google’s AI models are climbing fast.

So OpenAI is now regrouping.

A new “reasoning model” drops next week, rumored to reclaim the crown.

But make no mistake: this is survival mode.

The same company that made Google panic in 2022 is now fighting to prove it hasn’t peaked in 2025.

It’s kinda poetic…

Because every “code red” in tech history sparks one of two things: a comeback story or a collapse.

And right now, the world’s most powerful AI company is deciding which one it will be.

Turns out, history doesn’t repeat itself in AI - it retrains.

Google CEO just told the one AI nightmare that keeps him awake: “When deepfakes get so good that we literally won’t be able to tell what’s real anymore.. and bad actors get their hands on it” 😳

His exact words after Shannon Bream pressed him:

“That’s the kind of thing you sit and think about” - Sundar Pichai

This 64-second clip is chilling.

Because it’s not about sci-fi doom. It’s about a trust collapse.

When seeing stops meaning believing, democracy, finance, and human relationships all start to fracture.

Imagine:

→ A fake video of a president announcing a market crash.
→ A CEO “approving” a billion-dollar transfer.
→ Your voice cloned for a scam you never made.

That’s not the future. It’s already happening.

AI can cure cancer, or collapse consensus reality.

The difference lies in how we build safeguards as fast as we build capabilities.

And that’s why companies like Oscilar are becoming critical infrastructure for the AI era.

They’re building agentic systems that can detect deepfake-enabled fraud, verify identities, and flag anomalies in milliseconds.

Turning AI against its own dark side.

Because in a world where every pixel, video, or voice can lie, the new currency isn’t data.

It’s trust. And whoever protects it, wins.

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.

Anthropic just dropped Claude Sonnet 4.6, and it basically collapsed the gap between “chatbot” and “employee” 😳

Claude Sonnet 4.6 isn’t a smarter assistant.
It’s a cheaper operator.

Yesterday, if you wanted near-Opus performance, you paid for it.
Today, Sonnet 4.6 gets almost the same scores - at Sonnet pricing.

That changes the math.

Mid-tier AI now handles real workflows, not just prompts.

The numbers are pretty wild:

↳ 72.5% on OSWorld (real computer tasks).
Opus 4.6: 72.7%.
GPT-5.2: 38.2%.

↳ 94% on insurance workflow evals.
Submitting claims, self-correcting errors.

↳ 1M token context (beta).
Think entire codebases. Full legal docs. Multi-month planning.

↳ Claude in Excel now edits pivot tables, charts, formatting.
It also pulls live data from S&P, PitchBook, Moody’s via MCP.

↳ Pricing: $3 / $15 per million tokens.
Same as 4.5. No premium.

And here comes the wild part.

There’s a demo of Claude AI logging into a Shopify store, changing delivery pricing thresholds, verifying across pages - no API, no developer.

Cost per task? $0.30 🤯

This is a massive unlock for pretty much everyone.

→ For founders: fewer SaaS tools, more agent layers.
→ For enterprises: automation without custom integrations.
→ For small businesses: admin becomes scriptable.

When intelligence is this cheap and this operational, the future of work won’t be about headcount.

It will be about how many AI operators you deploy per dollar.

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

Tech loves irony: Apple, the company accused of “doing nothing in AI,” is now quietly becoming an AI infrastructure company 😳

For years, the narrative was simple: Apple is behind in AI.

No frontier model.
No flashy demos.
No arms race with OpenAI or Google.

And yet developers are now wiring Mac minis into AI clusters to run models that used to require data centers.

No hype. Just physics, systems, and taste.

The truth is that Apple never optimized for cloud AI. It optimized for local intelligence.

Unified memory.
Ridiculous bandwidth per dollar.
Neural Engines everywhere.
Hardware + OS designed as one system.

That combo accidentally unlocked something big:

↳ Inference loves memory more than brute force
↳ Batch size = 1 is the real world
↳ Most AI workloads don’t need a nuclear reactor

So while everyone chased FLOPS, Apple made efficient compute boringly good.

Now look at what’s happening:

→ Developers cluster Mac minis instead of buying GPUs 
→ Large models run locally, offline, private
→ Power draw measured in hundreds of watts
→ Costs drop from “enterprise-only” to “indie-accessible”

And the irony?

The company accused of doing nothing in AI is quietly becoming an AI infrastructure company.

And this is classic Apple.

Not winning headlines but winning constraints.

Turns out, Apple wasn’t late in AI.

It was simply playing a different game the entire time.

Anthropic just dropped their own OpenClaw version - you can now orchestrate AI agents to write code on your server straight from your phone, without tying any commands or code 😳

AI giant Anthropic just added “Remote Control” to Claude Code.

Here’s what that actually means:

→ Your terminal session keeps running on your laptop.
→ You walk into a meeting.
→ You monitor, nudge, or fix it from your phone.

No SSH gymnastics.
No tmux + Tailscale stack.
No duct-taped open-source agent.

For months, builders played with OpenClaw (formerly ClawdBot) - a plugin-based agent that chains tools, edits files, and orchestrates apps.

Powerful. Flexible. Model-agnostic.

But remote oversight was full of fragile workarounds.

Anthropic just collapsed that gap.

With one simple command.

OpenClaw gave you autonomy.
Anthropic just gave you continuity.

And continuity is what turns an AI demo into a daily workflow.

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

This is the wildest AI breakthrough of the year: a tech founder with zero background in biology just saved his dog using a custom mRNA vaccine he designed with ChatGPT 😳

Sydney entrepreneur Paul Conyngham adopted Rosie knowing she had aggressive cancer.

Months to live.

Instead of accepting it, he tried something unusual.

He sequenced her tumor DNA and asked AI what to do next.

The result: a bespoke mRNA cancer vaccine designed for a single patient - his dog.

And here comes the wildest part.

→ $3,000 to sequence Rosie’s tumour genome
→ AI tools helped identify mutated proteins and potential drug targets
→ Conyngham used ChatGPT + AlphaFold to model the cancer biology
→ UNSW scientists helped manufacture the custom mRNA vaccine
→ After injections starting December 2025, a tennis-ball-sized tumour shrank by ~50% by March 2026 🤯

One researcher’s reaction: “Holy crap, it worked.”

But the most interesting part was not about science. It was the bottleneck.

Designing the vaccine took weeks.
Getting ethics approval took months.

In other words, the limiting factor in medicine may no longer be discovery.

It’s process.

AI is quietly turning biology into a programmable field. Like software.

When a determined founder can prototype a personalized cancer therapy for $3k and a chatbot, the implications get very big, very fast.

Most importantly, Rosie’s story isn’t about a dog.

It’s about what happens when AI meets biology, and stubborn humans who refuse to give up.

AI + Biology will change the world.

Wild: at a $14 billion revenue run rate, Anthropic is the fastest-growing software business of all time. And it’s not even selling software - it’s writing it 😳

Claude Code alone is at a ~$2.5B run rate in under a year.

It now generates ~4% of all GitHub commits 🤯

The numbers are absurd:

→ $0 → $100M → $1B → $14B ARR in ~3 years
→ $100k+ customers grew 7x YoY
→ 500+ enterprises now spend $1M+ per year (40X growth)
→ $30B fresh capital at a $380B valuation
→ IPO prep reportedly underway for 2026

While others chased consumer virality, Anthropic went straight for enterprise codebases, recurring budgets, and workflows that teams can’t turn off.

Yes, they’ll spend billions on training and compute.
But revenue is scaling fast enough that burn is shrinking relative to sales.

We thought AI would be a feature inside software.

Instead, the model became the engineer.

If this trajectory holds, 2026 won’t be about better prompts.
It’ll be about who owns the AI layer writing the world’s code.

Oh, and Anthropic should IPO this year, and it will be the biggest tech IPO of all time.

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

Anthropic just killed traditional UIs: Claude now has an app layer inside chat, and you can use real applications & build things through MCP Apps 😳

It’s basically software showing up inside the conversation.

You ask Claude a question.
It pulls in the tool.
The work happens there.

And that changes the role of SaaS entirely.

We’re moving from:

“Open the app, then do the work”
to
“Describe the work, the apps come to you.”

Early signals this is bigger than a feature launch:

→ Amplitude is a launch partner - product analytics charts can be generated and edited directly inside Claude
→ Tools like Slack, Figma, and Asana can be triggered from chat instead of separate dashboards
→ The interface layer shifts from menus and tabs → natural language
→ When AI becomes the primary user, seat-based SaaS pricing starts to look… exposed

For 20+ years, SaaS companies competed to be the place you log into every morning.

Now, the most important product might be the one you never leave.

Software isn’t disappearing.
But it’s becoming the back office to an AI front desk.

And if chat becomes the operating system, dashboards start to look like legacy software.

Most importantly, when the interface shifts, the power usually shifts with it.

P.S. check out this deep dive into the 🤖 Agentic Interface 🤖, & how AI companies are going all in on the Agentic OS: https://lnkd.in/dva3Eejf

Claude AI just dropped Voice Mode in Claude Code, which means you can now talk to your terminal & it codes for you 😳

Currently available as a limited beta to 5% of users, Voice Mode will be gradually rolled out to everyone in the coming weeks.

Finally, everyone gets their Tony Stark moment.

P.S. also check out how I Turned Claude Cowork Into My Personal COO 🧠: https://lnkd.in/eS6JCk2G

Meet Jim Simons: an 86-year-old math genius who built the greatest quantitative hedge fund outperforming Warren Buffett, George Soros, Carl Icahn, and every investor who ever lived 😳

Jim's story was absolutely wild.

At 40, Jim Simons left a famed math career to launch the most successful hedge fund ever: Renaissance Tech.

This fund earned more than $100 billion for its investors.

Jim Simons's Medallion fund has had unreasonably high annual returns of 66% across 30 years 🤯

For the perspective, Buffet clocked in at about 20% per year.

Before passing away, Simons was worth over $31 billion, and his secret was dead simple:

→ He wanted to smoke everywhere.
→ He walked around the office barefoot.
→ He fell asleep at presentations.

The lesson: cold showers and meditating won't make you rich.

The real secret to wealth is just focusing on your thing and the freedom to be unapologetically yourself.

What a legend.

Stop paying $20/month for Claude Code. Chipotle’s AI bot is FREE.

Someone asked Chipotle’s support assistant Pepper how to reverse a linked list in Python.

It answered correctly.

Not burrito recommendations.
Actual code.

A fast-food chatbot quietly passing a classic computer science interview question wasn’t on my 2026 bingo card.

We’re at peak AI now.

Nailed it! Anthropic is airing ads during the Super Bowl mocking OpenAI's decision to put ads in ChatGPT, and they're absolutely hilarious 👏😂

OpenAI recently launched ads as a new revenue stream to support massive compute costs.

Meanwhile, Anthropic just responded by committing to no ads inside Claude, and running a national spot making fun of AI interrupting your life with promotions.

One side optimizes for scale + monetization.
The other is betting that “ad-free” becomes a trust moat.

Different revenue models.
Different incentives.
Different futures for AI.

When the AI becomes the interface, the business model becomes the product.

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

Anthropic is going deeper into financial services: Claude Opus 4.6 just dropped, and it’s basically an AI model that can now do junior analyst work end-to-end 😳

Opus 4.6 is a workflow automation for regulated, high-stakes knowledge work.

The wild part?

The model isn’t just “smarter.”
It’s structured enough to operate inside real financial tasks.

What changed:

↳ 1M-token context → can ingest entire filings, earnings decks, and research threads at once

↳ Beats OpenAI’s GPT-5.2 by ~144 Elo on GDPval (finance/legal knowledge work benchmark)

↳ Built to produce spreadsheets, financial analysis, and presentations - not just text

↳ Designed to run multi-step “agent” workflows inside Claude Cowork

↳ Market reaction hit data incumbents like FactSet and S&P Global within hours

For 20+ years, finance software monetized interfaces to data and models.

Now the model is becoming the interface.

If AI can read the filings, build the model, stress test assumptions, and output the deck, what exactly is left for the traditional tool stack?

My takeaway is this:

AI in finance is shifting from “assistant” → “operator.”

Less time gathering data.
Less time fixing spreadsheets.
More leverage for the people who actually know what questions to ask.

The bar for entry-level work drops.
The bar for judgment rises.

Old edge was all about access to tools.
New edge is all about knowing what to tell the tools to do 🤖

Most importantly, finance isn’t getting replaced by AI.

It’s getting restructured around AI-native workflows.

And that’s one of the biggest shifts for financial services ever.

Wild: One of the world’s top engineers just said that Claude Code built in 1 hour what took a Google team one year 😳

Jaana Dogan, a Google Principal Engineer, gave Anthropic’s Claude Code a high-level description of a distributed agent orchestrator.

A system her team had debated, designed, and stalled on for a year.

Claude shipped a working prototype in ~60 minutes 🤯

But that’s not the shocking part.

What’s shocking is that Google allows its engineers to use Claude Code instead of forcing Gemini, or their latest IDE Antigravity.

So AI didn’t out-code Google. It AI out-ran bureaucracy.

Big teams don’t move slow because engineers are bad.

They move slow because:

→ Alignment takes weeks
→ Reviews take months
→ Decisions get diluted

AI does something brutal and powerful now: it produces Version 1 immediately.

And once something exists, debate collapses.

As legendary Paul Graham has said before: “A concrete thing beats endless discussion".

And there’s one more uncomfortable truth…

If your company forces internal tools while better ones exist outside, you’re optimizing for politics - not output.

The best teams will win by doing one thing and one thing only:

Give builders the best tools, and then get out of the way.

Revolut just made your phone safer than your wallet. And it might have also just redefined what security truly means in FinTech 😳

Every 2 minutes, a phone is stolen in London.

But the real danger isn’t losing your device - it’s what happens after.

Criminals now force victims to unlock banking apps and authorize instant transfers at knifepoint.

No hacking. No phishing. Just fear.

That’s the brutal new reality Revolut is fighting.

That’s why Europe’s most valuable FinTech just launched Street Mode, a feature that makes your money harder to steal in the real world.

Here’s how it works:

↳ Inside your “safe zones” (like home or work):
Transfers flow normally.

↳ Outside those zones (say you’re walking at night):
Large transfers trigger a 1-hour delay and a second selfie check before approval.

That’s 60 minutes for you, or Revolut, to stop the fraud.

A digital cooling-off period for the physical world.

Here’s why it matters:

→ The UK saw 78,000+ phone snatchings last year, many ending in “transfer muggings.”
→ Instant payments made fintech fast, now Revolut’s making it safe.
→ This is security that adapts to you - where you are, not just who you are.

Think of it as “adaptive finance security.”

Banks have long focused on digital fraud.
Revolut’s now tackling physical coercion.

And that’s a paradigm shift.

Competitors will not only copy this. They NEED to copy this as soon as possible.

→ Apple hinted at it with “Stolen Device Protection.”
→ Monzo tried with “Known Locations.”
→ Now Revolut’s Street Mode is the first to give users control, not just algorithms.

This is the future of FinTech security that we also envision at Oscilar:

↳ Context-aware
↳ User-controlled
↳ Real-world smart

Because safety isn’t just about passwords anymore.

It’s about presence, place, and time.

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.

Google quietly dropped one of the most interesting findings about current LLM architecture. Turns out, copying your prompt twice can 4x AI model’s accuracy 😳

No fine-tuning.
No new weights.
No extra “thinking” tokens.

In a December 2025 paper, Google Research tested 7 models across 7 benchmarks.

The intervention was almost silly:

Prompt → Prompt.

47 wins out of 70 tests.
Zero losses.

One example:

→ Gemini Flash-Lite jumped from 21% to 97% accuracy
On a basic name-retrieval task.
By pasting the same input twice.

LLMs process text left → right.
Every token can only look backward.

So when you write 50 lines of context and end with a question, the context never “knew” the question was coming.

Repeating the prompt gives the model a second pass.
A cheap approximation of bidirectional attention.
Like letting it re-read the exam before answering.

And it gets more interesting:

When chain-of-thought reasoning is enabled, the gains from repetition mostly disappear (5 wins, 1 loss, 22 ties).

Which suggests something bigger.

Part of what we pay 5-10x more tokens for in “reasoning mode” is simply the model giving itself another look at your input.

That’s not higher intelligence.
That’s an architecture workaround.

So the real gap isn’t scale.
It’s causal attention leaving performance on the table.

The teams that crack efficient bidirectional inference will compress the reasoning tax to near zero.

Everyone else will keep selling you more tokens.

The next AI battle will be about attention, not parameters.

Wild: Duolingo announced it was going “AI-first” just over 6 months ago. Since then its stock is down 64% 😳

Duolingo just discovered the hard way that AI can scale a product, but it can’t save a soul.

In April, they went “AI-first” 🤖

→ Fired human contractors.
→ Let AI write lessons.
→ Pushed out 50 new courses overnight.

And for a moment, Wall Street loved it.

- Stock pumped to $540.
- Efficiency! Margins! Infinite scaling!

Then reality hit.

Users opened the app… and got sentences like:

“The bears are drinking beer at the party.”
“The cows cleaned the kitchen yesterday.”
“The cats went to the market today.”

It felt less like learning Spanish and more like babysitting a hallucinating chatbot.

People with 500-day, 800-day, and even 1,000-day streaks quit overnight.

Competitors with human-curated content saw a surge.

But here comes the real story nobody’s talking about.

AI didn’t kill Duolingo. Duolingo killed Duolingo’s trust.

Language is deeply human.

Idioms, tone, context, culture - that’s where humans shine, and LLMs fail.

Yes, AI can accelerate content.

But when it replaces the very thing users value most, you don’t get efficiency. You get “enshittification.”

And once trust cracks, no streak counter can fix it.

This is the lesson every consumer company should remember:

AI is an amplifier.

If your product is good, AI makes it great.
If your product is fragile, AI makes it crumble faster.

Turns out, Duolingo just accidentally shipped a masterclass in what NOT to do with AI.

Steve Jobs would have turned 71 this week. This talk Steve gave in 1983 at Aspen is one of his most beautiful yet least-known talks. Every founder, every entrepreneur - everyone - needs to watch it.

Because this wasn’t just a speech - it was a blueprint for the future.

He predicted that computers wouldn’t just be tools for work but extensions of creativity, personal expression, and human potential.

More importantly, Jobs foresaw concepts strikingly similar to today's AI systems:

"Looking ahead 50 to 100 years, if we can create machines that capture an underlying spirit, set of principles, or way of looking at the world, then when the next Aristotle arrives, perhaps they will carry one of these machines. After they're gone, we could potentially ask the machine, "What would Aristotle have said about this?" We might not get the right answer, but we might. That's incredibly exciting and one of the reasons I do what I do" - Steve Jobs

Fast forward to 2026, and AI is fulfilling Jobs’ vision of democratized technology, enabling more people to innovate, create, and solve problems.

Visionary legend.

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

Traditional SaaS is dead.

I asked Claude AI to vibe code a Workday replacement.

4 hours and 70,000 lines of code later, Claude rebuilt our entire HR stack:

New hire onboarding flows
PTO tracking
Org charts with drag-and-drop edits
Admin dashboards
Auto-generated policies
A surprisingly tasteful UI

None of it worked.

Payroll broke, and we missed paying the staff for 4 weeks.

But boy was it beautiful.

New AI playbook? IBM just did the opposite of what AI experts predicted - it tripled entry-level hiring 😳

The $260B giant everyone assumes would automate first and ask questions later.

Instead, they’re tripling entry-level roles in 2026 - including software developers.

While 37% of companies say they plan to replace early-career jobs with AI.

Here’s what they realized:

The old entry-level job?
AI can do most of it.

So IBM didn’t cut the role. They rewrote it.

→ Junior devs spend less time on routine coding, more time with customers
→ HR staff intervene on chatbot outputs instead of answering every ticket
→ AI literacy is now the fastest-growing skill in the U.S.

Same U.S. headcount - roughly flat after layoffs + new hires.
Different org design - broader at the base.

Because if you cut juniors when AI handles “grunt work,” you don’t get leaner.

You get a talent cliff in 3–5 years.

No pipeline.
No future managers.
No one who actually understands how your AI systems evolved.

Most importantly, this isn’t anti-AI. It’s post-hype AI.

Not “AI replaces juniors.”
More like: AI compresses apprenticeship.

Maybe the real AI strategy isn’t fewer humans - it’s better ones.

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

Bill Gates would be worth $1.47 trillion if he still had his Microsoft shares 😳

But he became friends with Warren Buffett and started to diversify his portfolio.

He’s worth only $106 billion today.

Lesson: choose your friends wisely.

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.

RIP traditional FinTech.

I asked Claude AI to vibe code a Revolut replacement.

5 hours and 80,000 lines of code later, Claude rebuilt our entire digital banking stack:

KYC onboarding flows
Account creation & virtual cards
Transfers & bill pay
Subscription management
Fraud alerts
Admin dashboards
Auto-generated compliance docs
A surprisingly sleek UI
It even had confetti animation when the balance went up! 🎉

None of it worked.

Transactions failed. Balances desynced. Customers couldn’t pay.

We were in full panic mode, and regulators were calling nonstop.

But boy, was that confetti beautiful.