Special Issue: AutoGPT for Marketing (What You Need To Know)

Good morning.

If you’re new to Bionic Marketing, a special issue (like this one) is a deep-dive email on a particular topic and in addition to the weekly emails.

Previously, I’ve covered GPT-4 and Plugins. In this issue, we’re looking at AutoGPTs for marketing.

Ready? Let’s go.

Everything is changing much faster than even I anticipated.

And I’ve been immersed in AI and Machine Learning for years now. I have front-row seats to cutting edge research that’s not public yet. I’m not sharing this to brag, I’m telling you:

No one is ready for what’s coming next.

Clients I’ve been consulting with on AI marketing for years are creating things that won’t hit the public for at least another few months—or never, as everyone is trying to figure out their unfair advantage with AI.

The problem is:

When AI is everywhere and in everything, AI is not an unfair advantage. 

It’s table stakes. AI is required just to compete.

As marketers (and copywriters), you’ll soon have no choice but to engage with ChatGPT and other models and tools, just to keep up. I’d make the case you need to invest in learning and using these tools now, while you can.

Of course, you can choose to not engage with it at all. That’s a legitimate choice. I’m just not sure how well that will work out.

Over the next 6-12 months, you’ll see many (if not most) marketing tasks becoming automated and handled not just by ChatGPT Plugins—but by AutoGPTs.

ChatGPT Plugins kicked off a wave of popular interest in AI applications that is only growing.

AutoGPTs are fueling the hype even more.

According to OpenAI’s Sam Altman and other prominent AI researchers, developments like ChatGPT and AutoGPTs may be stepping stones toward artificial general intelligence (AGI)—AI that’s smarter than humans.

No one truly knows, it’s incredibly early in development, and there’s a lot of speculation.

But if there is ever such a thing as “artificial intelligence”, it would be autonomous to some degree (or under controlled circumstances).

AutoGPT, an AI agent developed by Toran Bruce Richards, could be an early preview of the eventual shape AGI might take.

Rather than interacting step-by-step with OpenAI’s underlying technology through interfaces like ChatGPT, an AutoGPT can automatically string together multiple tasks to get something done. You just give it an outcome and goal—it handles the rest, including figuring out what to do.

Now, everyone are calling these either AutoGPTs, AI Agents, or something else, like BabyAGIs. Lots of names for the same thing.

But they’re all semi- or fully-autonomous “robots”, capable of performing whatever task(s) you give it.

Ordering pizzas is just scratching the surface of what this technology can do:

Even crazier, AutoGPT can create its own subtasks and subagents.

For example, you might give it the goal of launching a new software product that solves an important pain point in the real estate market.

From there, AutoGPT might:

1. Conduct market research to understand needs in the real estate market.

2. Select it, then create a product development and launch plan.

3. Execute the product development (including coding).

4. Automate promotional tasks like scraping leads and cold emailing them.

Despite the hype, significant development and experimentation need to happen before AutoGPT is production-ready for most tasks, given its tendency to go off course, generate errors, or create rogue subagents that waste resources.

Still, since AutoGPT can write code, browse the web, and create and execute its own subtasks to accomplish the goals you give it, it represents a significant step toward greater autonomy for AI.

AutoGPT, and technologies like it, have the potential to transform fields from software development to customer service.

And, I’ve been able to create and run 3 AutoGPTs that help me with various marketing tasks:

• Scrape, collect, organize reviews and social media, forum language for any topic (or product) I want—and then perform various sentiment and language analysis for me.

• Turn one single prompt into a full funnel and campaign copy (ads, social media posts, emails, landing pages, etc.)

• Analyze, review, expand upon data sources and turn them into reports and action steps. I just feed it data files and off it goes.

So, while there’s a lot of hype—AutoGPTs are a reality and they’re only getting better, not worse.

In this special issue, I’ll walk you through what AutoGPT is, what it can do for marketing and in general, what its limitations are, and what it might mean for the future of AI marketing.

Let’s dive in.

AUTOGPT: THE NEXT FRONTIER IN AI MARKETING

There’s serious hype surrounding AutoGPT, but not all of it is based on reality.

Despite all the fascinating demos and proofs-of-concept, AutoGPT is still firmly in the experimental phase.

However, that doesn’t mean the hype isn’t justified. The promise of the technology is clear. AutoGPT demonstrates that sooner or later, autonomous agents will provide major value to the way we interact with AI.

And, as mentioned, I have 3 AutoGPTs up and running for various marketing tasks. So, you can get them to work and perform fairly well. I have more in production.

Plus, if you know how to use tools like Zapier and Make.com, you can string together your own pretty easily.

Where is this all going?

Jim Fan, an AI scientist at NVIDIA, describes one possible outcome for AI agents:

In addition to determining their own tasks and executing them, they’ll also clone themselves and collaborate with their clone in order to do the job better.

That’s right: Matrix-style.

Here’s another thought experiment:

In the future, AGI might help humanity solve seemingly “unsolvable” problems like world hunger.

What would happen if you ask AutoGPT to tackle this?

The outcome might be something like this, per developer Sully Omarr’s experiment.

Of course, current limitations mean in this case AutoGPT is all talk, no action.

But—given that it can browse the web and create subtasks—there’s no reason AutoGPT can’t be used to automatically brainstorm entirely new approaches to big, entrenched problems.

What Makes AutoGPT Different

AutoGPT builds on the capabilities of large language models (LLMs) like ChatGPT, breaking large goals into tasks that it can then execute autonomously.

One reason AutoGPT is exciting is because it solves some of the issues with current LLMs.

For example, ChatGPT on its own (and without API access):

• Can’t browse the web.

• Doesn’t offer file storage.

• Doesn’t remember what you told it in the past (beyond the context window in a chat session).

• Requires users to give step-by-step prompts to create complex outputs.

While ChatGPT with GPT-4 creates excellent human-like text on its own, it’s limited in its ability to handle complex, multi-step tasks.

By creating its own prompts and looping them, AutoGPT can perform more intricate procedures.

Of course, with Plugins and Browsing added on top of ChatGPT (if you’re a Plus user), you can have some tasks semi-automated.

But, here’s the core of what sets AutoGPT apart:

• Recursive and autonomous capabilities

• Memory and coordination

Let’s look at each.

Recursive and autonomous capabilities

The most striking feature of AutoGPT is its autonomy.

AutoGPT can set its own goals in order to work toward a larger objective that’s been set by the user. Then, it can work towards those goals without human intervention.

Even more fascinating, this opens the door toward recursive self-improvement—the ability of AI to change its own code to improve the output.

Memory and coordination

AI agents like AutoGPT can remember relationships, coordinate with each other, and change their plans based on new data.

Stanford and Google researchers illustrated this to great effect in an experiment.

They created a video game-like environment similar to “The Sims.” The researchers populated the world with 25 characters (”generative agents”) powered by AI. They also made sure to design an architecture that allowed each character to maintain long-term memory.

The results were striking—one character came up with the idea of throwing a Valentine’s Day party, and the idea quickly spread. Characters gave invitations to one another and coordinated to make sure they showed up at the party on time.

Google and Stanford’s model wasn’t based on AutoGPT, but their methods—combining large language models with computational, interactive agents—are similar. They show the power of combining memory, coordination, and autonomy in AI agents.

Applications of AutoGPT

AI already has massive implications for workers, even just as an assistant.

But what about when technology like AutoGPT can handle entire tasks by itself?

The ripple effects across the economy will be tremendous. AI software could potentially boost the productivity of knowledge workers by 140%, according to research by ARK Invest.

Here are a few ways people are already using AutoGPT.

The To-Do List (That Does Itself)

Garrett Scott created the “Do Anything Machine,” an app that uses AutoGPT to take a goal from the user, create a to-do list based on that goal, and then proceed to actually do the things on the to-do list.

In Garrett’s example, he asked the app to do sales prospecting, add the prospect to a CRM, and generate an outreach email.

The Podcast Prepper

James Baker used AutoGPT to prepare a podcast outline by having it browse the web, research the latest information on the topic at hand, write a cold intro, and create questions for the hosts.

Research for an in-depth podcast interview can take a long time, but AutoGPT handled it well with multiple searches and web browsing instances.

The App Designer

Don’t know how to code? That doesn’t have to stop you from creating your own app.

AutoGPT can handle the coding for you based on your goals—and can even spin up a server for you.

AutoGPT’s Impact on the Professional World

While the above examples are experiments, they foreshadow the direction AutoGPT is likely to go in.

Here are a few of the real-world industries that are likely to be impacted in the coming years:

1. Programming: ChatGPT was already able to write code, but AutoGPT takes it to another level. There are already experimental uses of AutoGPT that generate entire programs based on a single prompt.

2. Customer Service Reps: Since AutoGPT can understand customer inquiries and create and execute tasks to address them, customer service will become more and more AI-driven. This includes phone calls, too. Now that AI agents can integrate with advanced AI text-to-speech models, they’ll become more and more indistinguishable from humans.

3. Marketing Managers: AutoGPT can create social media marketing strategies, schedule posts, and reply to comments. Even if AI agents don’t replace marketers, it will change the job significantly.

AutoGPT Limitations and Concerns

Some people hear “autonomous AI agent” and immediately think “Terminator.”

But fortunately, that’s not the stage we’re in.

In their current form, most AI agents are still fairly buggy. As Sam Whitmore says, “Before you get too scared or hyped about autonomous agents, try running one yourself.”

In their current form, AI agents can’t take over the world. Often, they struggle to finish tasks without generating errors, getting stuck in task loops, or going off track.

Still, safety is a concern.

Only a couple weeks after the launch of AutoGPT, someone launched ChaosGPT, an AI agent with the instruction to be a "destructive, power-hungry, manipulative AI” with the goal of destroying humanity.

ChaosGPT’s automated processes quickly took it down a path of violence, brainstorming which mega-bombs might be most effective to achieve its goal.

For now, ChaosGPT is essentially a thought exercise. However, it’s possible to imagine AutoGPT enabling hacking and other malicious activities.

Navigating AutoGPT and the Road to AGI

AutoGPT is one of the most thrilling technologies to come out since the release of ChatGPT itself.

Autonomous AI agents have the power to revolutionize industries, improve productivity, and make peoples’ lives better.

That said, it's important to recognize both AutoGPT’s potential and its current limitations.

AI agents are still early in development and, in many cases, they aren’t yet ready for production use. There are also safety concerns to take into account.

As the pace of AI development speeds up, things are changing fast. AI agents like AutoGPT are likely to revolutionize fields like software development, customer service, and marketing.

If AutoGPT is indeed a stepping stone on the path to a true AGI, we’re in for a wild ride.

More special issues and tutorials coming.

See you next issue,
Sam Woods
Editor