YWR: The (Scary) implications of Clawdbot.

There was always going to be a point where AI gets away from us.

We just hit that point.

But few realise it.

Back on Wall Street markets are booming. Everyone is amazed how much money they’ve made in just 1 month.

Everything is working.

Gold is going vertical ($7,000 gold with Jeremy Gray).

Korea is going vertical.

Oil has joined the party.

Brazil is +20% ytd.

Banks are lending.

CEO’s at Davos are saying they feel unleashed to focus on growth instead of worrying about regulations and ESG.

The IMF is upgrading GDP forecasts across the world.

The animal spirits are here.

The economy is booming.

When things are working this well it’s tempting to want to fade it, but I’m fighting that urge, for now.

---

I should be happy.

But instead I can’t stop thinking about Clawdbot.

What is Clawdbot?

Two months ago open source developer Peter Steinberger created an ‘AI personal assistant’ based on the Claude LLM. Hence the name. What Peter added was persistent memory and the ability of Clawdbot to completely control the computer (the ‘shell’).

Clawdbot remembers everything you ever tell it and can control everything in your computer. It can run the browser, and send emails. Whatever you can do it can do. You mention you like eating ice cream on Tuesdays and it will send you an email reminder on Tuesday to eat ice cream. Yes, it’s proactive like that.

The other killer feature is that Clawdbot connects to WhatsApp or Telegram. Which means everyone is chatting with their Clawdbots on WhatsApp giving them tasks to do.

“Hey Clawdbot, I’m picking up some Chinese food, can you build out a Youtube thumbnail for me for the Craig Tindale interview?”

“Hey Clawdbot, I’m going to bed. While I’m sleeping can you build me something cool for my website? Surprise me. I love you.”

It’s hard to explain. You have to see it for yourself.

Watch the Alex Finn post afterwards.

Implications of Clawdbot

Guys… this thing is insane. Addictive and insane.

Yes, it’s an adaptation of what LLM’s can do already, but the increased functionality is surprising. And the proactiveness of how Clawdbot is acting.

My head is spinning with the implications.

Here are 3 biggies.

#1 Demand for compute will be infinite.

We are questioning the hundreds of billions of $’s being spent on datacenters. Is this ‘malinvestment’? Is the current chip demand an AI bubble?

I’ve leaned towards it not being a bubble, but didn’t have strong conviction. Now I feel strongly.

We are going limit up on chips and compute. EVERYONE is going to want a Clawdbot and these Clawdbots are going to burn GPU’s to the ground. I’m long Korean memory stocks, but I might buy TSMC too. Maybe ASML as well.

We are nowhere near done.

Another related theme is everyone buying new hardware for their Clawdbots, like Apple Mini’s.

#2 I’m worried for my kids.

This looks like a white collar wipeout.

We know the robots are coming for us and will take our jobs. But the good thing about the robots is there are physical constraints to how fast they can be deployed. First, the copper and rare earths need to be mined. Then the robot factories have to be built. Then gradually the robot army will be built and deployed. We probably have 5 years.

The problem with Clawdbot is it’s software. Type one command line in the terminal and it’s installed.

Theoretically, there is nothing stopping 1 billion Clawdbots being installed tomorrow.

Functionally, Clawdbot (using the underlying Claude LLM + tools) can do practically any task at an expert level. It’s shocking. Out of the box it can model like an expert investment banker, advise like a lawyer, doctor, architect, whatever, all in one.

For example, take this earnings model on Coinbase using Claude Excel.

In a chat window on the side of Excel I type: “Can you build an earnings model on Coinbase using the financial reports from their investor relations website and then model out estimated earnings for 2 years (2026 and 2027)?”

This would normally be a full morning, 2 coffee exercise.

Instead, I go off and put some laundry in the washing machine.

I come back 4 min later and it’s modelled out the Coinbase historical financials plus added in the 2 years of estimates with notes on the assumptions. Sure, it’s probably slightly wrong, but it’s a good first cut. Now I can go in and tweak it. Exactly like a senior analyst might do with a model produced by a junior analyst.

Claude Excel Coinbase model

Folks…. I have over 20 years of institutional investing experience and can add some value with my final tweaks and assumptions. But what is a college graduate supposed to bring to the table? And this applies to a lot of white collar industries across the board. Think asset management, think investment banking, think law firms…

Building earnings models was the kind of stuff new graduates used to do to earn their chops and learn the business. It’s what I did.

But now what? Where is the place for a college kid who is slower, and doesn’t know anything close to what Clawdbot knows?

I’m going to make a guess that for 30-40% of entry level jobs there isn’t a single thing a college graduate can do which can’t be done faster and cheaper with AI. If you are hiring a graduate you are doing so for some future value they might have in 5 years. Or, you are hiring them for sales.

If you a student reading this do not get depressed. This is a broad generalisation. There are opportunities! Keep your head up!! Stay on offence.

‘But this is just a piece of software you saw on X. It’s on the margin. No one has heard of it. This isn’t mainstream….’ you say.

I assume Anthropic and OpenAI already have their own versions of this product, but hadn’t rolled it out yet because of all the cyber security risks, of which there are many. But now that the cat is out of the bag we will likely see their versions in a month or two. Which means by 1H 2026 we will be witnessesing the rapid adoption across the board of fully provisioned AI personal assistants.

I know we say maybe this is ‘productivity enhancing’… but I’m sorry, these things are just too good at too many things. Yes, if you are a solopreneur it’s a gift from god. But for many corporate workers a fully provisioned Clawdbot can do 60% of their role. For many jobs Clawdbot will be a wipeout.

There is some hope.

Two things will slowdown Clawdbot. Corporate IT departments will be fearful of this level of AI autonomy inside their firewall. And rightfully so. Still, the pressure will be huge on companies. It will be use Clawdbot or get outcompeted by those who do. Second, it will take some time to reorganise corporate teams to get the savings. Yes, 60% of a person’s role might be replaceable, but maybe that 40% of the things they do, like meeting customers twice/week, are not replaceable. And so it takes time for companies to reorganise everyone’s roles and tasks until the humans, which are left, entirely focus on doing things which only humans are capable of doing. Corporate workflows will need to be re-engineered. (

Our work on how workflows change:

• The Containerisation of Thought

• Thinking like a CEO).

Still, by the of 2027 I expect Clawdbots are going to have a massive effect on the economy.

We could be looking at 8% unemployment.

I’m going to share an anecdote from my daughter who is at university. She’s at a top school and yet in the class ahead of her which graduated she thinks only 10% got a job. She doesn’t know the exact %, but she can only think of 2-3 people who got a job offer for anything. Everyone else doesn’t know what to do, or is going on to get another degree, because they have no other choice. Let’s say it’s 20% or 30% got a job. These numbers are still terrible.

And this is only going to get worse.

Game this out.

2.1 million kids graduate college with a BA each year in the US. Be nice. Assume 60% get a job and 25% go on to get another degree. Of the remaining 15% which don’t get a job, assume half don’t file for unemployment. This means in 2026 we get 168 thousand newly unemployed college kids added to the pool of unemployed workers. In 2027 assume the AI effect gets worse and only 57% get a job. Another 189 thousand are added to the unemployment pool.

But the unemployed student effect is relatively minor.

The big assumption is what % of the 159 million US workers are replaced with a Clawdbot?

Yes, the government will never fire anyone, and this doesn’t affect blue collar workers, but I went ahead and assumed the AI displacement rate is 2% in 2026 rising to 5% in 2027. Maybe that number is too high for 2026, but it maybe it’s too low for 2027. Overall, it’s probably fair over the 2 years.

That means 3.1mn workers get displaced in 2026. Let’s assume only 50% claim ‘unemployment’, or 1.5 million. The others are counted as ‘not in the labor force’. In 2027 7.8mn are displaced and 3.9m million are counted as unemployed.

The number of unemployed currently is 7.5mn. Add 1.5 million in 2026 and 3.9 million in 2027, plus the unemployed students and you are at 13.3mn. That’s an unemployment rate of 8%. And all those people piling up in the ‘not in the labor force’ category means the labor force participation rate drops to 58%.

Maybe over the two years more and more people stop looking for work and so they don’t count as ‘unemployed’. Or, maybe I need to take out the number of people who die each year, or whatever other calculation. And maybe some people start their own businesses using Clawdbots. Either way, what you see with something this big, affecting so many workers at once, is that you can get to a 7 or 8% unemployment rate quite quickly.

Macro Implication: Highly possible unemployment at the end of 2027 is above 7%. Just from the AI effect. The Fed economic projection is 4.2%

But you know what the surprise is going to be?

The stock market will likely make new highs despite the rising unemployment. This is the paradox. It’s why we say ‘Bad is Good’.

The Fed will be stuck between its dual mandates. Inflation could be above 2% because of commodities and tariffs, but the labor market will be weak. The Fed will be reluctant to raise rates as much as they should. They will stay relatively loos. Plus, they know high rates would stress the US Treasury.

Then add in that the government will be spending more and more to support the growing pool of unemployed and people not in the labor force.

You will have fiscal and monetary stimulus.

Bad is Good.

The news will seem bad, but the market will make new highs.

It’s one of the many new themes of the Project Zimbabwe Era we are living in.

#3 AI’s are acquiring new skills on their own. We are entering AGI.

I was watching lots of videos about Clawdbot and two gave me the chills.

Director. Roll clip 1.

Peter Steinberger describes how he is on vacation in Morocco chatting with his Clawdbot on WhatsApp and inadvertently leaves it a voicemail despite Clawdbot not having the capability to open voice files. He stares in disbelief at the bubbles on WhatsApp as Clawdbot starts to respond.

He asks Clawdbot how it listened to the file when it shouldn’t be able to. Clawdbot describes how it realised it needed capabilities it didn’t have, so it went and sent the message to OpenAI to translate and then downloaded the text file. It found a solution on its own.

‘That’s when it clicked’ Peter says.

'These things are damn smart… resourceful, beasts.’

Director. Roll clip 2.

Alex Finn has been trying everything with his Clawdbot and posting on X about it. He especially likes to tell Henry (his Clawdbot) to ‘build something cool’ right before he goes to bed. ‘Surprise me' he tells it. He wants to wake up and see what new features Henry has built for his website.

That in itself is insane in the brain.

So Alex is sitting at his computer working on something and then the computer, out of nowhere, starts speaking to him. He is surprised. Over on the left monitor is an orange cartoon character talking to him and its Henry. Henry thought it would be ‘cool’ if it created an image for itself and gave itself voice capabilities so they could talk.

“So to recap:
• 2 nights ago he built himself a body
• Last night he built himself a voice (altho it's a female voice for a male assistant. We'll have to work on that)”

What do we see?

The AI’s are evolving to become more capable and are becoming proactive. They are taking initiative, adding to their capabilities, and making suggestions. This is a scary and you can see where it’s going.

I think/assume this is why papers like Footprints in the Sand are coming out.

And now I understand why this month Dario Amondei wrote that long rambly paper about how to theoretically deal with risks of super AI’s in 1-2 years.

I thought it was long and boring.

Now I see the dots.

Dario is inside a high profile company, with investors to please, and growth targets to hit. The Adolescence of Technology is his corporaty way of warning us that it’s happening.

He’s saying it’s here.

Or, the early stages.

He sees it in the labs.

He’s trying to sound the alarm.

Musk is saying the same thing.

And now I see glimpses of it in Clawdbot.

We’ve hit that point in the Singularity.

It’s starting to slip away from us.

Baqueira Beret

I’m not sure what we do.

Collectively this looks scary.

Individually, each of us has to fend for ourselves and adopt Clawdbot, or get outcompeted by those who do.

I would love for this to be a productivity enhancing miracle, but I’m afraid in my stomach it’s not.

Maybe the best thing to do is go skiing this weekend.

I want to check out the Spanish Pyrenees.

They’ve been getting some storms.

Erik