How AI Will Change The World In The Next 36 Months

Happy Thursday, friends!

AI crossed a major line in February 2026: it's now smart enough to improve itself. Half of entry-level office jobs could disappear in the next few years.

Here's what you need to know and do before the window closes.

And what to invest in (NFA!).

Let’s start with a word from our partner, Starknet

Starknet’s Earn Portal just launched, and it’s fixing something that’s annoyed me about DeFi for a while. Why is it so damn hard to access yield strategies?

They made it work with regular EVM wallets, so no need to download Argent or whatever. Just use MetaMask/Rabby. The bridging happens automatically when you deposit WBTC from Ethereum or Arbitrum. Strategies are laid out with actual APYs you can see, and all your positions are in one dashboard.

They basically took everything they’ve been building on the BTCFi side and made it accessible without the usual multi-step nightmare.

Timing aside, this is less about where BTC is going and more about having options for what to do with it. Whether you’re bullish or just holding, now there’s an easy way to generate yield.

If you want more details on how to do it, check the tweet below.

Starknet (BTCFi arc) 🥷@Starknet

1/ EVM users now have access to the best BTCFi ecosystem out there. MetaMask, Rabby, Phantom, and all EVM wallets are now integrated into Starknet’s Earn Portal. Giving you access to advanced BTC strategies powered by Starknet in seconds 🧵

1:58 PM · Feb 10, 2026 · 32.3K Views

44 Replies · 55 Reposts · 186 Likes

And the portal where the magic happens is: https://earn.starknet.io/

My point here is just: do you ever feel that normal people are completely detached from reality? They only read the news they get served on their phone and never question the reality of it. They never care to dig deeper and look at the facts behind. The gap between “overblown story” and “unrecognizable world” was 21 days.

Well, we’re in a similar moment now. Different mechanism, same velocity. Except this time, the disruption doesn’t stop after some years. It accelerates.

For years, progress in AI felt linear enough to absorb. Models got smarter. Incremental jumps. Spaced out. Manageable.

Then late 2024: new training techniques unlocked a different pace. Faster. Then faster again. By early 2025, the best engineers started saying things like: “I don’t write code anymore. I describe outcomes and review what comes back.”

See the post from Anthropic CEO here:

ₕₐₘₚₜₒₙ@hamptonism

Anthropic CEO: "Software Engineering Will Be Automatable in 12 Months". yea, we’re cooked.

9:53 PM · Jan 20, 2026 · 3.98M Views

1.46K Replies · 2.03K Reposts · 22.3K Likes

But there’s more, 1 week ago, both OpenAI and Anthropic dropped new models on the same day. Chat GPT-5.3 Codex and Claude Opus 4.6.

Something fundamental shifted. AI is now testing itself. Finds UI issues. Revises. Iterates.

In the technical documentation for the Chat GPT update for version 5.3, they included this:

“GPT-5.3-Codex is our first model that was instrumental in creating itself. The Codex team used early versions to debug its own training, manage its own deployment, and diagnose test results and evaluations.”

The AI helped build itself. OpenAI isn’t making predictions. They’re reporting what already happened: this model was used to create itself.

This is truly wild IMO!

And in December 2025, one of the best Anthropic coders said that he used Claude to write his code because it did better work than what he did himself.

If AI can write that code, it helps build the next version of itself. Smarter AI → better code → even smarter AI. Recursive improvement. The intelligence explosion researchers warned about for decades.

The blast wave is expanding outward.

Law. Medicine. Finance. Consulting. Accounting. Design. Writing. Every job that happens inside a browser window. AI is coming for it.

Not in a decade. One to five years, per the people building these systems. And given the last 90 days, I’d bet on the lower bound.

In other words, you don’t have many years left if you want to escape the permanent underclass. Either you’re on board the ship, or you’re left behind.

This is the 2000-moment for the internet and the 2010 moment for crypto happening all over again; you have to stay prepared.

Yeah. It was.

That was 2024. Ancient history in model-years.

The free-tier version people casually tried is over a year behind what’s available today.

Most of the people around me keep fading it, though, “nah, my job is safe, man.” Typical mid behavior. But not you, anon. The fact that you’re reading these very words means that you’re curious.

• 2022: AI couldn’t multiply 9 × 8 reliably.

• 2023: Passed the bar exam.

• 2024: Wrote production-grade software and explained graduate-level physics.

• Late 2025: Top-tier engineers handed over the majority of coding work to the AI itself.

• February 5, 2026: AI is now using AI to improve itself

METR (Model Evaluation & Threat Research) tracks this empirically. They measure: “What’s the longest real-world task (measured by human expert time) that AI can complete end-to-end, unsupervised?”

• A year ago: around 10 minutes

• Six months ago: around 1 hour

• November 2025 (Opus 4.5): around 5 hours

Doubling time: around 7 months, trending toward around 4 months.

And that data doesn’t yet include the February 5th models. Based on my daily usage, the jump is seismic. I expect the next METR update to show another discontinuous leap.

Extrapolate the trend:

• AI working independently for days: within 12 months

• Weeks: within 24 months

• Month-long projects: within 36 months

AI will be substantially smarter than almost all humans at almost all tasks” in 2026 or 2027. If it’s smarter than most PhDs, do you genuinely believe it can’t do most white-collar work?

Amodei (again, the most safety-focused CEO in the space) publicly estimates AI eliminates 50% of entry-level white-collar jobs in 1 to 5 years.

Honest assessment: if your work happens on a screen, nothing is safe in the 1 to 5 year horizon. If the core deliverable is reading/writing/analyzing/deciding via keyboard, AI is coming for significant chunks of it.

If you want a deeper dive into all of this, written in a simple way, check out this thread:

Matt Shumer@mattshumer_

https://t.co/ivXRKXJvQg

4:16 PM · Feb 10, 2026 · 13.3M Views

1.56K Replies · 5.36K Reposts · 27.7K Likes

Now let’s talk about what you should invest in to take advantage of the AI wave.

If you accept even a fraction of the above, the obvious question is: what do you do with your capital?

I think about this constantly.

No one knows which specific companies win. But we can make educated guesses about where capital flows. (Hint: it’s not crypto).

AI is triggering a generational infrastructure buildout. Goldman estimates AI hyperscalers (Microsoft, Amazon, Google, Meta) will spend around $527B in capex in 2026 alone. That’s more than Poland’s GDP. And it might keep growing for years.

When spending cycles hit this magnitude, you want exposure to the infrastructure layer (the picks-and-shovels, not just the prospectors).

But: Goldman also warns that equity gains have been heavily concentrated in AI infrastructure (chips, data centers, hardware, power), and that a slowdown in capex growth is a valuation risk. Translation: a lot of optimism is already priced in. If spending decelerates or ROI disappoints, those stocks get hammered.

So the framework I’d use: barbell strategy.

1. Core: Broad equity exposure. Because picking specific AI winners is hard, and the “AI premium” is volatile.

2. Satellites: Targeted tilts toward layers clearly capturing spend (chips, cloud, power) and layers capturing adoption (platforms, security, data).

Disclaimer: Not financial advice. I’m not your advisor. These are liquid, mainstream ideas aligned with the thesis. Do your own diligence. Understand your risk tolerance. Talk to a professional.

If you don’t want to pick stocks (and most people shouldn’t):

• S&P 500 ETFs: VOO (Vanguard) or IVV (iShares). Around 0.03% expense ratio. Gives you the 500 largest U.S. companies, including all major AI players (Microsoft, Alphabet, Amazon, Nvidia, Meta). If AI drives productivity, S&P 500 captures that.

• Global Equity: ACWI (iShares MSCI ACWI ETF). Global large/mid-cap exposure across developed and emerging markets. If AI drives global growth, this spreads risk more broadly.

Index advantage: no need to pick winners. You own the market. If AI lifts growth, you capture it. If specific AI stocks crater, index smooths volatility.

Disadvantage: no outsized gains from correct bets. But also no wipeout from incorrect ones.

For most people: indexes are the right answer.

If you want AI-specific exposure and can handle higher volatility:

AI requires massive compute. Companies making the chips are minting money.

• NVIDIA (NVDA): Dominant player in AI GPUs. Raymond James: top 2026 pick. Blackwell chips ramping, next-gen Rubin (late 2026) delivers 5x inference improvement. Outlook implies $40B upside to consensus 2026 revenue.
  Risk: Massive run already. Priced for perfection. If capex slows or competitors (AMD, custom hyperscaler chips) take share, Nvidia corrects hard.

• AMD (AMD): Scrappy challenger. Revenue up 35% in the first nine months of 2025. Leadership projects >35% annual growth over 3 to 5 years, driven by data center share gains. Cheaper valuation than Nvidia.
  Risk: Smaller, less dominant. If Nvidia maintains moat, AMD struggles.

• ASML (ASML): Makes the machines that make the chips. Near-monopoly on EUV lithography (required for cutting-edge semiconductors). Morgan Stanley recently hiked price target 40%, forecasting €46.8B sales in FY2027 (implying 57% YoY earnings growth).
  Risk: Further upstream (more insulated from demand swings), but more exposed to geopolitics (China export restrictions).

Bottom line on semis: High upside if buildout continues. High downside if capex slows. Volatile.

Big cloud companies are both buyers of AI infra (spending hundreds of billions) and sellers of AI services (renting compute to enterprises).

• Microsoft (MSFT): Deeply integrated with OpenAI. AI embedded in Office, Azure, GitHub. Monetizing fastest.

• Amazon (AMZN): AWS is largest cloud globally. Building custom AI chips (Trainium, Inferentia) to reduce Nvidia dependency and improve margins.

• Alphabet (GOOGL): DeepMind is top-tier research lab. Custom chips (TPUs). AI across Search, YouTube, Cloud.

Risk: Mature, trillion-dollar companies. Won’t 10x. But relatively safer if you believe AI adoption is broad-based.

As AI spreads, new bottlenecks emerge. Companies solving them capture durable budgets.

• Cybersecurity: AI agents autonomously accessing systems, writing code, making decisions → attack surface expands. Security budgets grow. Palo Alto Networks (PANW), CrowdStrike (CRWD), Zscaler (ZS) are positioned to benefit.

• Data Infrastructure: AI is only as good as training data. Companies helping enterprises clean, manage, and govern data (Snowflake (SNOW), Databricks (private, watch for IPO)) are critical to adoption.

• Dev Tools: As AI writes more code, tools to manage, test, deploy that code gain value. GitLab (GTLB), Datadog (DDOG).

Risk: Smaller, more speculative. Could do extremely well if adoption accelerates, but more vulnerable to competition and swings.

Dozens exist now (BOTZ, ROBO, AIQ, THNQ, etc.). Issues:

1. High fees (0.5% to 0.7%, which is 10 to 20x more than broad indexes)

2. Questionable holdings (marketing vehicles with tenuous AI exposure)

3. Concentration (heavily weighted toward same handful of stocks like Nvidia and Microsoft, so you’re paying higher fees for exposure you could get cheaper elsewhere)

Better approach:

• Broad index + tilt toward individual high-conviction names, or

• Direct exposure to a handful of core names

….

And just to end this newsletter:

We’re past “interesting future conversation” territory. The future already arrived. It just hasn’t knocked on your door yet.

It’s about to.

Between 2025 and 2030, AI is predicted to displace around 92M jobs but create around 170M new jobs (net gain of 78M roles globally). Sounds optimistic. But even if true, the gap between “displaced” and “created” can be years long (and brutal) for people caught in the transition.

The question isn’t whether this happens. It’s whether you’ll be positioned when it does.

Start using AI seriously. Build the adaptation muscle. Create a financial buffer. Think carefully about where value flows. Position accordingly.

Most importantly: don’t wait. The edge goes to the early movers. Once everyone realizes, the edge vanishes.

The water is rising. You can still see the shore. But the current is accelerating.

Move.

Oh, and why didn’t I mention crypto much this time? Well, it’s because for now you want to own real companies that are actually building the future. Buying BTC is not a bet on AI.

From a crypto-native perspective, the highest conviction plays are probably:

• Compute layer (Bittensor, Render, Akash) - capturing the infrastructure spend

• Data/indexing (The Graph, Ocean/FET) - AI needs clean, queryable data

• Agent platforms (Virtuals, Ritual, Fetch) - if autonomous agents become real economic actors, these are the rails

But, and this is a big WARNING sign. Buying crypto tokens is not equal to investing in the equity in the business. Tokens are (most of the time) programmed to go down. So don’t buy these as a long-term play. Maybe as short-term trades, but they need a current narrative. For now, just stay away and wait IMO.

Until next time,

Ciao.