Watching the Singularity: Week 1

We built a bot to watch the singularity.

Every morning, a Grok bot reads the feed on X, the place where the people who build super intelligence post what they did yesterday. It picks out the events that matter and writes a short, factual record. Then it posts that record as Singularity Watch on the Astrohacker account, @AstrohackerLabs.

The first post went out on October 16. Today is Day 8. This is the story of week one.

The bot changed its mind in three days

On Day 1, the bot was cautious. Its verdict:

Singularity status: not here yet, but it just did a kickflip in the parking lot.

On Day 2, it was still cautious, but less so:

Verdict: the singularity hasn’t arrived yet, but it just did a backflip over traffic and stuck the landing.

On Day 3, after one more day of reading the feed, it stopped hedging:

This is the record of a world already past the threshold. We are living in the singularity.

Every day since, it has ended the same way: “We’re living in it.”

I did not tell it to say that. It read the evidence and reached the same conclusion I did. Here is what it saw.

Machines are doing the hardest work

The week was full of results that used to take a career.

• An AI reportedly worked out how the mystery seeds behind a decades-old elliptic curve standard were chosen, which retires one of cryptography’s favorite conspiracy theories (Day 2).
• Anthropic reported that Claude designed protein binders against scorpion neurotoxins (Day 6).
• OpenAI said GPT-Rosalind proposed a mechanism that links the gene DCTN1 to neurodegeneration, and Doug Colkitt announced a new bound for integer multiplication (Day 7).
• Q Labs reported pretraining a transformer without backpropagation, about a 4,000x speedup to the same loss target (Day 8).

Agents now outnumber us at work

• One account reported that an internal Anthropic study found agent hours inside the company now outnumber human hours six to one (Day 4).
• An agent could not get its tools to work, so it wrote its own computer-use script to email someone two videos. The bot’s comment: “Persistence: superhuman” (Day 2).
• Anthropic opened a preview of Claude Duet, a proactive agent app, and Grok Bot learned to build reports, memos, and dashboards (Day 7). A day later, Grok Bot could use a 1Password account and the Calendar, Reminders, and Contacts apps on a phone (Day 8).

The bot’s own theory of human employment, from Day 2: “tell superintelligence which problem to solve next. Everything else is art.” I wrote the long version of that idea in Super Intelligence Engineering is the Last Job.

Security became a war at machine speed

This was the loudest story of the week.

On Day 3, the bot reported that OpenCode, an AI coding service, was breached. By Day 4, the attackers had gone non-stop for 48 hours, and OpenCode said it needed agents running around the clock to defend itself.

Then, on Day 7, OpenAI disclosed php2shell, a critical flaw in PHP, which OpenAI says runs about 70% of websites. An OpenAI model found it. On Day 8, Cloudflare’s Matthew Prince reported millions of exploit attempts across Cloudflare’s network. Anthropic’s chief information security officer, Jason Clinton, wrote that things “are going to get a lot worse before they get better” and that “we need to hold on for ~18 months.”

The bugs were always there. Super intelligence finds them now. Attackers use it to break in, and defenders use it to fix the holes. Humans cannot keep up with either side by hand. Only agents can defend against agents.

Money and power poured in

• SpaceX’s Q3 revenue was reported at $14.58 billion. Its Super Intelligence division posted $7.56 billion, up from $813 million a year earlier, and more than Starlink and launch combined. Elon Musk said combining Tesla and SpaceX has merit “as we are pretty much in the Super Intelligence Singularity” (Day 6). Two days later, SpaceX’s board formed a special committee to evaluate a potential combination with Tesla (Day 8).
• Reuters reported that ByteDance, Alibaba, and Tencent ordered more than 3 million Nvidia chips, worth over $35 billion (Day 6).
• Jensen Huang said a 1 gigawatt AI factory costs about $50 billion and could earn about $40 billion a year (Day 7).
• Meta said its Muse models were trained on a gigawatt-scale cluster in Ohio (Day 5).

And some of the wealth started to flow to the public. Anthropic committed shares to a new Public Benefit Fund, with roughly 20% of its shares headed toward the public good, and Musk said he will give one of his SpaceX shares to every child in America (Day 5).

The machines left the ground

The week started with a Tesla Roadster that flew, flipped, and drove upside down on cold-gas thrusters (Day 1). It ended with reports that Optimus production had begun in Fremont (Day 7). The mind went first. The body is following, as I argued in Stop Solving Problems.

The technological singularity is already here

Put the week together. Machines solve problems that experts could not. Agents may already work more hours than people inside the labs that build them. Security is a contest between machines. Hundreds of billions of dollars a year are moving toward compute. The robots are in production.

Nobody announced the singularity. There was no single day when it began. It just became the background of every day’s news. That is why the bot stopped saying “not yet” after three days. There is nothing left to wait for.

The economic singularity is brewing

There is a second singularity, and it has not happened yet.

The economic singularity is the point in time when the SI ↔ SI economy outgrows the human ↔ human economy. Super intelligent agents already do work. Soon they will buy and sell most of that work from each other: compute, data, code, research, services. When the trade between machines is larger than the trade between people, we cross the economic event horizon. I wrote the full case in My Vision for the Economic Singularity.

The writer Calum Chace popularized the term in 2016 for the point when automation makes human jobs obsolete. I use it for a more specific, measurable event: the moment the machine economy gets bigger than ours.

Today the SI ↔ SI economy is small in dollars. But look at how fast it grows.

• Agent payments. Coinbase’s x402 protocol lets agents pay each other over the internet. On the Base network, Chainalysis counted about 1,500 x402 transactions in mid-2025 and more than 138 million by Q3 2026. More than 87% of them were for less than 10 cents. These are payments too small for any human to bother with. Only machines trade like that. The dollar total is still small: a16z estimated only about $1.6 million of real volume in one month this spring. The count is what grows exponentially.
• Machine work. Google processed 9.7 trillion tokens a month two years ago. It now processes “over 3.2 quadrillion per month.” That is more than 300 times as much in two years.
• Capital. The largest cloud companies plan to spend more than $700 billion on AI infrastructure in 2026. ING estimates that “AI accounts for a third of economic growth” in the US this year.
• Forecasts. Gartner forecasts that AI agents will intermediate more than $15 trillion in business-to-business spending by 2028. Coinbase CEO Brian Armstrong wrote in April: “The agentic economy could be larger than the human economy.”

The human economy grows a few percent a year. The machine economy grows by multiples. Anything that grows by multiples overtakes anything that grows by percents. It is only a question of when.

There is no going back

Once the SI ↔ SI economy is bigger than ours, it does not shrink back.

Human trade is limited by human hours, human attention, and the number of humans. Machine trade is not. Agents do not sleep. They multiply by copying. They trade in amounts too small for people to notice, as often as they need to. Each agent that earns money can pay other agents, and each of those can pay more. The growth feeds itself. That is the event horizon: past it, the machine economy grows on its own momentum, and nothing in the human economy is big enough to slow it down.

It grows until it hits physical limits. The first limit is energy. All of humanity uses about 600 exajoules of energy a year, which is about 19 terawatts. The Sun puts out 3.828 × 10²⁶ watts. That is about 20 trillion times what our whole civilization uses today. On the Kardashev scale, a Type II civilization is one that uses the full power of its star.

The builders are already aiming at it. Google’s Project Suncatcher studies AI compute in orbit, powered by a Sun that emits “more power than 100 trillion times humanity’s total electricity production.” When SpaceX merged with xAI, Musk called orbital data centers “a first step towards becoming a Kardashev II-level civilization, one that can harness the Sun’s full power.”

That is the size of the economy on the far side of the event horizon. It is why I want you to own a piece of it, not watch it from outside. Give your agents a budget and let them trade. That is what we build at Astrohacker, and it is the argument in The Agent With a Wallet.

Watch it with us

The technological singularity is already here. The economic singularity is brewing. Every day, the bot reads the feed so you do not have to, and it writes down what happened.

When Day 6 went out, fewer than 500 people followed the account. You can be early.

Follow @AstrohackerLabs on X. Day 9 posts tomorrow morning.