The AI-Operated Business

Recently, I’ve been thinking through the implications of AI Assistants on work, in general. Things like:

• Which tasks can AI Assistants reliably complete, today?

• What’s the cost discrepancy between AI labor and human knowledge work labor for said tasks?

• Which tasks are within reach of future AI models? Which are less likely to be replaced?

• What will be the new bottlenecks when AI Assistant labor is cheaply available?

• What will this all mean for how new businesses are structured?

Over the coming weeks, I’ll share some of my thoughts on these questions.

As standalone questions, they may not seem deeply interesting (or relevant to you in your day to day). But in combination, they lead me to conclude that a new kind of business model will likely emerge.

Not because AGI is ‘taking over’. Not because humans will no longer be needed.

But because the incentives of a world where AI labor is cheaply available will necessarily push people toward figuring out ways of maximizing the work done by AI Assistants.

Just as the incentives of an online world pushed businesses online. An AI-Assisted world will push more businesses into leveraging AI Assistants.

Those businesses which take this to the limit are of special interest to me. Those are what I’m calling AI-Operated Businesses. (And I’m experimenting with building a couple of my own.)

But I’m getting ahead of myself.

In this post, I’ll be taking a closer look at just the first of the above questions.

If you’re not sure why you’re getting this, you either signed up on the Bizway blog or signed up for a Bizway user account—if you don’t want to receive these messages in future, there is an unsubscribe button at the bottom of this email.

1. What can AI Assistants actually do?

For a start, they can write decently well. They can comprehend and summarize text given to them. And they can write code and perform simple math functions.

Considering our most popular standardized test (SAT) for students includes as its 3 sections:

• Reading comprehension;

• Writing; &

• Math

These are no small categories.

If you add to this the ability to generate original images from simple text inputs (like the thumbnail image above), parse images for meaning and generate audio from text—it’s clear that these models are already incredibly capable.

The real limitation, as far as I can see, is to do with sequences of tasks and complexity.

It’s one thing to be told: ‘Write an 800-word article about X, including sections on A, B, C.’ and for the model to spit it out.

It’s quite another to say: ‘Come up with a plan for a new, profitable business and implement every step along the way’.

AI Assistants (for now) require a lot of hand-holding and human oversight.

But once you’re able to clearly scope the work that needs doing, they are incredibly helpful—I think it’s beyond argument that they are faster, more widely knowledgeable and have a longer stamina for work than we do on such tasks.

To give some more concrete examples of tasks that AI Assistants can actually do, here is a shortened list of my current AI Assistants in Bizway:

Some I use less frequently.

Some I use persistently (running on automations), so that they complete 500+ tasks per week on my behalf.

The point is: there are indeed tasks which can be 100% outsourced to AI assistants.

In general, these tasks relate to:

• Writing text;

• Explaining concepts;

• Extending ideas (brainstorming);

• Writing code or translating from one language to another;

• Structuring ideas into outlines and plans;

• Making simple calculations and forecasts (pricing, financials, etc.);

• Understanding and responding to text;

• Generating images or video;

• Parsing and interpreting images or video;

• Generating audio;

• Parsing audio.

The implications of this are what interest me. Not the current limitations of what AI models can and cannot do.

For me it’s enough to note that there are in fact tasks which can be 100% outsourced to AI Assistants, at 1/100th the cost of hiring a person to do it or doing it myself.

The quality might not be comparable to a human expert.

The tasks that fit this description (100% automatable and 1/100th the cost) may not all be especially valuable.

But if there were just one task with a 100X discrepancy, and a relatively low loss of quality—that task might just be worth building an AI-Operated Business around.

My belief is that there are many such tasks, and so it’s only a matter of time before these businesses are discovered and built.

The precise cost discrepancies and how the math might pan out on such AI-Operated businesses is what I’ll be exploring further in my next post.

For now, I’ll leave you with a question:

Which tasks do you perform in your work that meet the criteria ‘100% AI automatable at 1/100th the cost’?

(And the follow-up: What’s stopping you from outsourcing it to an AI Assistant, today?)

Best,

Gerrard

PS: If the above sparked any thoughts or ideas, and you’d like to share them, I’ll be very glad to read every response and reply where I can.