How Businesses Could Run Without a Single Human Employee.

For this week I have found another interesting article that describes a thinking exercise on the future of A(G)I driven businesses, with a focus on Agents. As you know I strongly believe AI agents will change everything we know, until this article I have mostly thought about agents on the consumer side. How they will disrupt to complete buying cycle, which is important to brands as it will fully change their way of doing marketing. Something I have written about in my third edition of this newsletter.

The article that we are going to discuss today focusses more on the future of businesses when super level AI Agents exist. It’s weird, very weird, but a great thing to think about and what this could potentially mean for your own business and/or career.

The article is called “What fully automated firms will look like” and is written by Dwarkesh Patel. Dwarkesh Patel is the host of The Dwarkesh Podcast, known for deeply researched interviews with leading minds like Mark Zuckerberg, Marc Andreessen, Sam Altman, and Tyler Cowen. He rose to fame quickly by obsessively preparing, sometimes 100+ hours per guest earning him a reputation for asking questions no one else does, and spotlighting rising thinkers before they went mainstream, earning praise from tech and intellectual heavyweights.

Patel has broken his article down in six parts: copy, merge, scale, distillation, evolve and takeover. These sections are the big advantages “AI Firms” will have over “Human Firms”. I will ask GPT to summarize each section and give my additional thoughts on them. Let me do say, this is not a (possible) future that will happen next year, or the year after. But let’s say in ten years? And maybe we are all wrong and we double it to 20 years? That is still way in our lifetime.. Crazy, YUP!

But first, he opens up the article with underestimation. I have used this saying a lot of times as well: a lot of people OVERestimate what the technology can do on the short term. There is still so many people who are unhappy with GPT responses, complain about the lack of quality of image and video models etc, purely because they think the tech is so far (already) that it can all be perfect with just a few words and a push of a button. This is obviously not the case, to get the most out of Generative AI models you need to do actual work. How disappointing right? 🤣 

On the other hand, people also UNDERestimate the impact this technology will have on the long-term. There is many reasons for that, one of those is definitely that it is so hard to grasp what it really means what is being build at the moment. This level of intelligence in machines, something more intelligent (eventually) as all humans combined? What does that meannnn? Another good reason of the underestimation gets highlighted by Patel in this article, I quote: “Most people are anchoring on how smart they expect individual models to be. (i.e. they’re asking themselves “What would the world be like if everyone had a very smart assistant who could work 24/7?”.)”

Having a super smart GPT as an assistant will be amazing, don’t get me wrong. But the challenge is in completely re-thinking how our businesses, our favorite brands, our jobs, our companies, our personal purpose, will be redefined.

The above quote by Bill Gates might sounds shocking, I don’t find it shocking at all anymore and fully agree. Imagine telling a 19th-century farmer in Europe that one day, 99 out of 100 people wouldn’t need to grow food but we’d have more food than ever. They would not be able to think about that, imagine that, it breaks my brain still too.

Let’s dive deeper into the article and start with the first section.

COPY

GPT: AI firms can copy their best talent instantly and endlessly. Instead of hiring and training people, you just duplicate your top performers. Like having thousands of expert engineers or even CEO-level minds working in sync. It’s not just individuals; whole successful teams can be replicated. This unlocks a level of scale, speed, and alignment that human organizations simply can’t match.

Let’s unpack this. I think the hardest thing in the world, for any company, is finding, hiring and keeping top tier talent. From my personal experience I can tell that, most of the time, it is just a handful of people that really push a company forward. It is a companies biggest bottleneck. As an entrepreneur you will rarely find someone that is willing to work as hard as you, which makes sense. Additionally, every great mind, entrepreneur or employee is limited by capacity, time and more. AI Agents, are not.

What this means is:

Another common discussion has to do with the amount of AI Agents, or LLMs, that will be useful. As in, will there just be one AI Agent or LLM to rule them all? Just like in the above example, wouldn’t just one incredible AGI version of Sundar Pichai be enough to run Google? I personally more believe in very specialized AI Agents who together tackle complex tasks. Why? Because they can be much more specialized in a certain job, task or industry, instead of being one super smart generalist who knows a bit of everything.

This also brings us to the next section of the article:

MERGE

GPT: AI firms won’t just copy agents, they’ll merge them. Different AI agents can instantly share what they’ve learned, combining insights and experience into a single, upgraded version. No meetings, no handovers, just direct integration of knowledge. Over time, these AI teams start acting like one giant, collective brain, learning and improving as a whole.

A human CEO runs a company by trusting others to give him or her the correct information, at the right time. Now if you have worked for any company before, you know the challenges that brings. Information quickly gets misrepresented and it is never real time. In this section Dwarkesh argues that there might be a central AI Agent who orchestrates all other agents, aka the CEO of the AGI Firm. Continuing with the CEO of Google as an example. This means that AI Sundar can not only copy itself but also gets information fed back from all the thousands, and thousands of AI Agents that this central AI manages in real time.

This way, this central AI Agent CEO can be aware of every little detail that is going on, use that data to run scenarios almost at the speed of light and make the best decisions for, everything? Possibly..

Dwarkesh: “Historical data going back thousands of years suggest that size is the key input for how fast your society comes up with more ideas.”. More people, is more brainpower and when the knowledge gets spread and shared properly, the more ideas and innovations will come fro that particular population. Well… This is exactly what AI firms will be capable of. Thousands of agents with perfect and instant share ability of knowledge, learnings and information.

SCALE

GPT: AI firms can grow without limits. With millions of AI workers and perfect knowledge-sharing, innovation happens faster and more often. Every insight spreads instantly, and every improvement sticks. These firms will look less like companies—and more like superorganisms with massive collective intelligence.

Now this is a scary, great, exciting, weird one. This is also one of the reasons why people look at this evolution the wrong way, or only look at the impact just partially. As said in the beginning of the newsletter, having a super smart GPT assistant or a set of AI Agents to help you do your work and daily tasks, is awesome. But man… that is only just the tip of the speer. Going back to the AI Firm, run by Central AGI Sundar. This AI Firm can have the ability to spin up hundreds of thousands of agents all working together towards a common goal. Specialists in every little facet of the company, brand or product and when successful they can just copy that high performing agent 1000x.

So what will potentially hold this back? Why will not everyone be able to do this? The short answer is: compute. If data is gold, compute is the refinery. How to think about this? The smarter you want your AI Agent to be, the more compute it will need. Dwarkesh therefore argues that we are going to rethink what is valuable in terms of where we need to spend most compute power. As an example states:

“Would it be worth it for Google to spend $100 billion annually on inference compute for mega-Sundar? Sure! Just consider what this buys you: millions of subjective hours of strategic planning, Monte Carlo simulations of different five-year trajectories, deep analysis of every line of code and technical system, and exhaustive scenario planning.”

Because having your super AI think through 1000+ different five-year trajectories so it can make the best decision for the company, is going to cost a sh*t ton of compute power. This is also the reason why you want hundreds of thousand of specialized AI Agents. Your social media copywriter agent for Instagram in Italy does not need the compute power to run that type of scenario for every word they write, right?

• Intelligence eats Compute

• Compute eats Energy

To power all this compute and therefore all the intelligence we need to, not only run AGI firms, but also solve cancer and every other decease on the planet, we need a lot of energy. The big tech players are already screaming for changes in regulation in order to speed up new forms of energy sources (such as Nuclear). A few days ago Sam Altman (CEO, OpenAI) was in front of the US Congress again, among other tech leaders, to discuss AI and mainly the race VS China. In that hearing, Altman very much focused on the importance of energy and calling it a “bottleneck to progress”.

It is also a common concern for many people. While it’s true that AI operations require energy, the per-query energy usage is relatively small. Estimates suggest that a single ChatGPT question consumes approximately 0.3 watt-hours (Wh) of electricity. In comparison, streaming an hour of Netflix requires about 0.8 kilowatt-hours (kWh), or 800 Wh, which is over 2,600 times more energy than a single ChatGPT question. To put it into perspective, you could perform over 2,600 ChatGPT queries for the same amount of energy it takes to stream one hour of Netflix. 

Now I do agree that while individual questions consume minimal energy, the cumulative effect of millions of users can be significant. Therefore, ongoing efforts to improve the energy efficiency of AI models and data centers are crucial to mitigate environmental impacts.

But AI is also the answer to its own problems. I do strongly believe that because of the ever smarter models we will see breakthroughs in, not only energy efficiency of AI models, but also on energy sources themselves. Nuclear and Fusion being the most exciting ones.

Now back to the article.

DISTILLATION

GPT: AI firms can create ultra-specialized versions of their top models—each fine-tuned for a specific task. Think AI data center operators who understand every wire, or legal agents with deep domain knowledge. Since knowledge can be cheaply copied, these distilled AIs can be both narrow and incredibly smart—making the entire organization faster, sharper, and more efficient.

This goes back to the specialized AI Agents and thousands of them, or one general AI Agent. It seems, for multiple reasons, that specialized agents is the way to go for the future. Not only for compute reasons but also for knowledge reasons. Dwarkesh does mention in the article that a certain level of general intelligence will be so cheap to have in every agent, that this will be the case. Aka every AI Agent will have a basic understanding of everything that is written on Wikipedia for example.

EVOLVE

GPT: Human companies age, slow down, and struggle to replicate their magic. AI firms won’t. Because everything, from skills to culture, can be copied and improved, these firms can evolve like living organisms. They can clone their best versions, adapt fast, and spin off new entities without losing what made them great in the first place.

Another thing that limits human companies is how they operate as a human, living, organism. They age, cultures change, and are hard to maintain. Today can be great, tomorrow something can happen in the world that effects most of your people mentally and your whole company has changed overnight. The magic that happened when Mark Zuckerberg was developing Facebook with a small team is hard to replicate in a company of almost 75.000 people. AI’s of course don’t have that problem and it can keep evolving what works, fast. Not only in one industry, but in every industry if it knows how to do it. Patel shares a great example:

I think an interesting point in this quote is that “when Elon isn’t the best person for the job” you can find the agent (or person) that is and replicate them. Crazy..

TAKEOVER

GPT: If AI firms can copy, specialize, evolve, and scale better than human firms, why would we need other companies at all? The article suggests that the first fully automated firm could become so efficient, it outcompetes everyone else. It could spin up its own versions of suppliers, partners, even competitors. But there’s a catch: even the smartest AI firm still needs an external feedback loop, like profits or market signals, to stay grounded in reality.

You might have been reading all the above and indeed asked yourself this question a couple of times already: do we need companies at all? It is a good question, I don’t have the answer. Dwarkesh thinks it doesn’t necessarily means that there will just be one large company that does everything for our planet. Experts do believe that a full robot economy could become reality though, which means costs of most good we as humans want will crash to (nearly) zero. These companies could indeed be run by AI Agents (why be limited by humans? 😅).

Final Thoughts

It is very thought provoking to say the least (in my opinion). It is very hard to imagine that a full company could be run by AI Agents, but when I think about it longer I can’t seem to see big reasons why not.. Now the question becomes, is this a bad thing?

I can imagine that for many people this line hits hard: “All of Google’s 30,000 middle managers can be replaced with AI Sundar copies.” I indeed also believe that a lot of jobs can be fully automated “away” by AI(s) and I don’t necessary think that is a bad thing. Ask yourselves, are we (as humans) really meant to work 40-hours per week (at least) to enjoy two days off? If you work full-time from age 24 to 67, you spend about 33% of your awake time working during those years .

That’s 1 out of every 3 waking hours..

I am more than happy to let all companies be fully automated and run by AI’s (assuming it will be safe, governments have found an income solution etc) so that I can focus on other things..

In order to get there, we need to solve the energy bottleneck first. So this is all not a given yet, but as you know I strongly believe in human ingenuity and that this will be solved (in collaboration with powerful AI).

Last point I want to make is that the above still feels very far away. As I said in the intro, it might be 5 years, it might be 10, or they all could be very wrong and it could be 20. That is still in our lifetime. I saw a podcast by Dwarkesh with the title: “AGI is still 30-years away”. Well.. that means that I will be 65 and probably are still fully capable of experiencing what that will mean. But when I see people, for example, launch new agencies, are focused on launching AI agents for brands etc. Yes I think that is all going to be super important and successful the next 3-5 years. But that is about it.. Almost anything you launch now, should be probably launched with a focus to build it for the next 5 years or so. There is a big chance that anything you start will be redundant by then.

There is nothing wrong with that in my opinion, gives you a clear window to kick some ass.

All together I, as always, want to end on a positive note. It is the same thing I say over and over again. I do strongly believe we will reach some form of “utopia” on the other end, more freedom, more time, more room for whatever it is people want to do. But the road to get there will be painful, but it will be worth it.

PS... If you’re enjoying my newsletter, will you take 6 seconds and refer this edition to a friend? It goes a long way in helping me grow the newsletter (and help more people understand our current technology shift). Much appreciated!

PS 2... and if you are really loving it and want to buy me some coffee to support. Feel free! 😉 

^{Who am I and why you should be here:}

^{Over the years, I’ve navigated industries like advertising, music, sports, and gaming, always chasing what’s next and figuring out how to make it work for brands, businesses, and myself. From strategizing for global companies to experimenting with the latest tech, I’ve been on a constant journey of learning and sharing.}

^{This newsletter is where I’ll bring all of that together—my raw thoughts, ideas, and emotions about AI, blockchain, gaming, Gen Z & Alpha, and life in general. No perfection, just me being as real as it gets.}

^{Every week (or whenever inspiration hits), I’ll share what’s on my mind: whether it’s deep dives into tech, rants about the state of the world, or random experiments that I got myself into. The goal? To keep it valuable, human, and worth your time.}