• Bionic Business
  • Posts
  • Issue #55: Agent Upgrades—Can they run your business now? (Plus new o3 model)

Issue #55: Agent Upgrades—Can they run your business now? (Plus new o3 model)

Good morning.

AI assistants are becoming operators in their own right.

If you’ve ever wished for a personal research team, a strategy consultant, and an execution engine all rolled into one, OpenAI’s Operator and the latest wave of AI assistants—like Perplexity, Google’s Gemini, and Astral—are getting dangerously close.

They’re decision-makers, fast learners, and, in some cases, proactive problem solvers.

Then there’s DeepseekR1—a model that recently dropped and is already turning heads with its retrieval capabilities. If you thought AI was good at finding answers before, this takes it to another level.

So what does all this mean for you?

It’s time to shift from using an AI or two—to actually directing several.

The winners in this next phase won’t be the ones who automate everything mindlessly. They’ll be the ones who know how to direct AI effectively, set constraints, and push the boundaries of what’s possible.

That’s exactly what we’re getting into today—how to work with these AI operators instead of getting left behind by them.

Let’s dive in.

—Sam

IN TODAY’S ISSUE 👨‍🚀 

  • The AI Assistant War has begun.

  • China setting standards with Deepseek R1.

Let’s dive in.

Operator: OpenAI’s Next Move Toward AI Autonomy

Alright, let’s start with the obvious one (from OpenAI).

If you’ve been paying attention, OpenAI has been inching closer and closer to something much bigger than ChatGPT.

Operator is their first real step into building AI agents that don’t just assist—you can actually delegate work to them.

If you want a full explanation of how Operator and these other AI assistants work, watch Operator’s preview

(Fair warning: OpenAI needs to invest in a new microphone)

Operator is designed to take action, execute tasks, and operate (hence the name) in a way that starts to blur the line between “tool” and “team member.”

Insiders at OpenAI say this is the foundation of AI agents that will act on your behalf—booking flights, analyzing documents, even automating entire workflows.

Although this is a huge advancement, this is just the start.

Operator is already “outdated” compared to other Assistants.

Here’s what I think:

This is OpenAI laying the groundwork for the future.

The real power move here isn’t Operator itself—it’s what Operator represents: a stepping stone toward AI that you don’t just talk to, but manage. And that means you’ll need to start thinking less about “using” AI and more about directing it.

Once systems and assistants like Operator are combined, the capabilities will be even more powerful than they are now.

Look at how Rowan Cheung is using it:

So far, developers have used Operator to perform a lot of tasks that aren’t just scheduled tasks, or questions, it’s a full task being performed like an employee would.

As I said before, Operator and the rest of the AI assistants are ready to be delegated to, not used to work with.

Most of the use cases you’re seeing so far, are typical “consumer app” stuff, like booking a flight, reserving a table at a restaurant, and so on.

There’s a better use of Operator: almost any task you’re doing, let’s say, in your browser can be handled by Operator.

Now think about the work you do as an entrepreneur, marketer, or copywriter.

Yeah, that can be done either in pieces or completely by Operator.

Now, when you throw their latest model o3, that was quite literally made available yesterday…

Now we’re cooking.

The o3 model (available as o3-mini and o3-mini-high for now), is pretty “smart”.

I’ve been using it for a few days (got access before public release).

And I’m a bit surprised, to be honest.

This release marks a significant advancement, particularly for STEM-related tasks.

  • 24% faster response times compared to o1-mini.

  • Improvied accuracy in coding and mathematics tasks.

  • Outperforms o1 in various benchmarks (the “intelligence” ranks somewhere between super smart friend and verifiable genius).

Model variants:

  • o3-mini: Optimized for performance and cost efficiency.

  • o3-mini-high: Enhanced version with additional compute resources.

The model excels in:

  • Programming and coding tasks

  • Mathematical problem-solving

  • Scientific research

  • Complex reasoning challenges

  • Self-fact checking.

Pretty cool.

The o3 model also introduces several key advancements that significantly improves AI agent capabilities.

Which means if you have these models and agents run parts of your business, those functions just increased by 50+ IQ points.

Smarter co-workers, in the form of agents, with o3 for a brain.

Enhanced reasoning:

  • Uses "private chain of thought" approach for deeper logical analysis.

  • Implements simulated reasoning (SR) for more thoughtful responses.

  • Achieves superior performance in complex problem-solving tasks.

Better task and workflow management

  • Powers "Operator," OpenAI's new AI agent for real-world tasks.

  • Enables complex multi-step operations through web interfaces.

  • Improves operational efficiency through better reasoning capabilities.

  • Enables more sophisticated autonomous decision-making.

  • Facilitates complex task management and automation with better accuracy and speed.

Yeah, I’m telling you:

The o3 model will significantly upgrade the AI workflows and Agents I have running in my projects.

Someone will—it won’t be me but someone—put together a 1-5 person team that quickly gets to $10M and goes fast toward $100M.

You can be one person plus o3 brains inside your Agents, and get to $1M pretty easily.

The age of the autonomous business is here.

Everything is converging.

More on this soon.

Google Gemini: The (Still) Unfulfilled Promise

Then there’s Google’s Gemini.

Look, we need to talk about Google’s AI strategy.

Because for all the insane potential of their models (especially in multimodal reasoning), Gemini still feels like a promise that hasn’t quite been kept.

The original Gemini 1 launch? Overhyped and underwhelming. Google talked a big game about outpacing GPT-4, and then… it didn’t.

Gemini 1.5? Much better, but still playing catch-up in practical use cases.

Now, with Gemini Advanced, we’re finally seeing some traction.

It’s got impressive long-context capabilities and deep integration with Google’s ecosystem—but is that enough?

Some diehard Google fans believe this is the AI we should be using—especially if you’re already neck-deep in Google’s world (Docs, Gmail, Search, etc.).

Others argue that Google waited too long and let OpenAI take the lead.

Google has all the right pieces, but they keep fumbling the execution.

If Gemini gets better at task automation, it could be a serious contender. But right now? It’s still not the AI “co-pilot” we were promised.

Your best bet for business is to keep an eye on Gemini, don’t depend on it quite yet. Let the assistants battle it out while you master agents, take it one step at a time while not falling behind.

Perplexity: The Facts

If Operator and Gemini are fighting over the future, Perplexity is solving the problems people have right now.

Perplexity has always been about better information retrieval—a search engine, but smarter.

And their latest AI assistant isn’t trying to be an “agent” like Operator. It’s trying to be the best AI-powered researcher and assistant you’ve ever used.

Want a real-time summary of a topic for work? Perplexity does it faster than ChatGPT (useful for your job or business).

Need a source-backed answer instead of AI hallucinations? Perplexity cites its sources (also useful for business)

Looking for a personalized knowledge assistant that actually learns what you care about? That’s where Perplexity is doubling down (really useful for business)

Perplexity has capabilities within your phone that allow it to perform tasks based on your needs.

Perplexity can be a defining tool within your business. It’s centered around factual information and outputs specifically for your situation, not just for the prompt you input.

Perplexity also stands out in its deep search integration—it doesn’t just pull from a pre-trained model, it actively browses the web for live information, giving you the most current and relevant answers possible.

Researchers, students, and journalists love it because it doesn’t just generate content—it finds facts.

Which matters, but..

Marketers and businesses feel it’s still too focused on retrieval and lacks the creativity of GPT-based models.

This is the AI you use if you want an experience and answers tailored specifically to you.

But the loss in creativity can be a big hit if you’re a marketer or copywriter.

If OpenAI and Google don’t figure out a better way to validate their outputs, Perplexity could easily take over if it solves this issue.

But also remember: Trust is the bottom line in any business, don’t sacrifice the bottom line in yours. Although the creativity might not be 100%, the facts are.

When I say to delegate work, I’m not saying you shouldn’t check outputs, and Perplexity could be the perfect tool for the job.

Astral: The AI Marketer

And the last of the assistants, Astral.

Right now, it’s waitlist only but I’d suggest you sign up for that so you can try it out. I don’t get paid for this mention, it’s not an endorsement, I just want you to get involved quickly with things like this.

Unlike the others, Astral isn’t trying to be everything at once.

It’s positioning itself as a team of “marketing agents that interact with the browser to automate repetitive tasks.”

I’ve said it for a few years now, but I’ll say it again:

The real power of AI models and assistants is when they’re specialized for a particular task or group of tasks—like Astral.

ChatGPT-4o is an excellent general purpose model.

But it’s not perfect for marketing or copywriting, and so on—unless you give it custom instructions inside a GPT or Project.

We will see more and more specialized Assistants and Agents (same thing, different name) in the next few months.

If you had some time and elbow grease, you could put together your own email marketing team of Agents—or a team of Agents for any marketing and sales function.

We’re very close to the possibility of stringing together much of a whole business with Agent teams.

In an upcoming issue, I’ll cover this in detail (the dawn of the autonomous business).

DeepSeek R1: The New Contender

If you don’t live under a rock, you've probably heard the buzz about DeepSeek R1.

(It’s not an agent or assisant—it’s a reasoning model, like OpenAI’s o1 model—but it can be the “brain” of your agents and agentic workflows).

It's not every day that a new player emerges and shakes up the status quo, but that's precisely what's happening.

For the good of the AI race, I’m glad Deepseek is contending with OpenAI.

On January 20, 2025, DeepSeek unveiled R1, quickly surpassing ChatGPT as the top free app on the U.S. App Store.

DeepSeekR1 was developed using fewer resources and for less money.

Also, it’s completely free to use, which is where all the hype is at.

Although DeepSeek R1 got a lot of hype claiming it’s the next big AI player, skeptics quickly came out and now it’s thought to steal a lot of information, especially on how it was built.

Speculation of it poaching OpenAI’s o1 model reasoning has been the main topic behind R1 in the past few days, along with why it has been released from China.

Many are afraid it’s stealing information and reporting back to China, but honestly, how is it any different than TikTok?

The same claims have been made about TikTok (which is why it was banned).

Alright, enough with the skeptical banter, you want to know if it’s going to be better for your business or not:

The model handles tasks requiring logical inference and problem-solving with notable proficiency. It’s similar if not slightly better than o1.

It provides accurate and relevant information, making it a reliable assistant for various tasks, but it's not without its limitations.

Currently, DeepSeekR1 lacks some advanced features found in competitors, such as voice mode and image generation. This is where the limited resources bottleneck the model.

There are also instances where the model avoids certain topics, likely due to content moderation protocols, just like ChatGPT used to when it first came out.

Although DeepSeek R1's hype is dying, it has done its job in pushing AI forward.

All so-called “AI” automations, workflows, agentic flows, and agents make use of various LLMs to “think” and “do” things.

DeepSeek R1 is an excellent “reasoning assistant” that can be in charge of analyzing, planning, and strategic work for agents and their workflows.

We’re entering an era where your choice of AI assistant is as important as your choice of smartphone, laptop, or productivity stack.

(They go together now).

It’s no longer about AI vs. no AI.

It’s AI vs. Better AI

So ask yourself:

What do you actually need? An executor? A researcher? An advisor? A marketer? A bookkeeper for your business?

Because the way you integrate AI into your busines now will shape how effective you are in the years to come.

The AI assistant wars have begun. Choose wisely.

In an upcoming issue, I will show you the vision, roadmap, and clear view of how you can transform your business into a self-driving business (a large part of it, anyway).

The dawn of the autonomous business is quickly coming.

Talk soon,
Sam Woods
The Editor