THIS WEEK: PROFITEERING IN THE AI WARS

x🌟 Editor's Note

Thanks for reading H4CKER, where I share my experiments and discoveries from the frontiers of AI marketing. Each week I try new tools and approaches with AI and share the best of what I find. If it’s useful to you, please let me know to create more of it in the future. Find me @NathanABinford on X or at [email protected].

🚀 The AI Revolution Is A Mercenary’s Paradise

When the big AI labs feel market pressure they respond with new models, better fine-tuning, and cheaper tokens -if only for a time.

But that’s the thing, we’re all living on borrowed time right now. It’s well established fact that Anthropic, OpenAI, and the rest can’t keep burning cash like this forever and prices will have to go up or their costs will have to go down substantially. Maybe both.

Who knows? The future is hard to predict but the low consumer-facing cost of intelligence is astoundingly cheap compared to the cost of providing it and that just isn’t sustainable.

For now the competition for new users is too intense to monetize effectively and its forcing the labs to keep their costs low (while they get us good and hooked).

So now is the time to take advantage of all the low cost tokens we can to build ourselves a platform to sustain us when the costs eventually rise.

Right now we’ve got an epic battle for attention going on with Claude Code and OpenAI Codex. Each releases new finely-tuned models and compelling features regularly and each has given back tokens for one reason or another, to keep their fanatic fan bases happy.

But what’s the best of each model currently?

Well Claude Code is fast. And for simple problems it does very well. For complex problems, still pretty well but just the default model doesn’t compete with GPT-5 Thinking in my opinion. Still, very good for simple updates and prototyping (lots of edits) quickly.

Codex is slow, but methodical. It does well operating on its own for a long time without straying or making many errors. Great for long tasks and bug fixing.

By comparison Cursor’s Auto Agent is a crash test dummy. It’s not worth your time for any kind of complex problem solving. But it’s great for explaining your code, finding bugs (not necessarily fixing them), and managing git and other project hygiene.

Your mileage may vary depending on how you use each, but what I’m getting at is that they’re each best at something different which starts to make them look like a team of specialists.

And that’s the right way to think about AI as we settle into our new agentic way of thinking and working…

🥷 Choose Your Fighter

I’m switching back and forth between models for different takes, different solutions, and using them to power different agents.

And this is our next quantum leap as AI-first operators: Spinning up agents to do various kinds of work for us; each time selecting models and engineering prompts like we’re fine-tuning a race car engine.

The model you use for different agents drastically changes how you have to interact with it. I’m pretty hands off with GPT-5 but won’t leave the room when Claude is coding.

Claude is quick and reckless. GPT-5 is ponderous and predictable.

Each has its place though. I’m not going to ask Claude to do something I don’t understand. I would trust GPT-5 with it, probably.

But waiting for GPT-5 Thinking is like watching paint dry, so not a great experience when you’re doing front-end tweaks and other simple, repetitive types of coding.

No doubt you’ll have favorites, we all do. But it’s probably better to know them all well and be able to put them to work when there’s a compelling offer or advantage.

🤖 2025 Is The Year Of AI Agents After All

We didn’t reach the singularity (that we know of) and AGI arguably hasn’t been achieved either and AI’s endless naysayers are still crowing loudly about AI winter…

But the reality is that the most important AI prediction for 2025 did come true: this has been the breakout year for semi-autonomous AI agents.

All throughout 2024 we heard about the coming “agentic age” of AI and, like with most things, as this prediction actually became reality, it has gone largely unnoticed by the masses.

Yet you can now have multiple agents running on your computer at the same time, all working on different tasks, at a level that beats a junior employee by a stretch.

Check out The Laboratory section this week to see a multi-agent setup in action!

Anyone with the curiosity to watch a few videos on YouTube can set this up and if you have a few hundred dollars a month for tokens, you can be building state-of-the-art software in your spare time.

And agents are the engine that makes all of this possible.

🧑‍💻 Work Is Over, Now We Conduct Business Like An Orchestra

If we put all of this together it paints a fascinating picture of the future where a single entrepreneur can dream, prototype, build, deploy, market, and manage a pretty significant business on their own (or with minimal contract labor).

Employees are out. Agents are in.

It’s time to reframe our concept of delegation to include agentic AI, because an AI-first individual can outproduce teams of skilled human laborers.

That’s not a sci-fi future either —that’s right now.

Of course, we’ll need to adapt our thinking to thrive in this new environment. When working with agents they can be incredibly intuitive one moment, and dumb as rocks the next. They need guardrails (and supervision).

And replacing or improving one is an engineering problem, not a human resources one.

What got us here won’t get us any further.

We have to relearn how to delegate.

We have to find ways to break down our big picture into smaller, self-contained pieces that autonomous agents can easily handle without creating fatal conflicts that humans could easily overcome but break machine workflows.

Working with multiple agents at once is dizzying and overwhelming. It’s more like conducting and less like any form of traditional working. And it comes at you fast.

The only way to master this new AI-first work dynamic is to experiment and to grind through the frustrations of a very steep technical learning curve.

But the rewards are going to be enormous; simply because most people are not going to do it. And those of us that can will beat entire teams of traditional workers who can’t.

The future will not be evenly distributed.

⚙️ The Laboratory: Prompts & Automations

See how I set up a multi-agent working environment in Cursor with OpenAI Codex, Claude Code CLI, and Cursor’s built-in agent. This is AMAZING for coding, of course, but could be used for really any kind of agentic workflows.

You simply MUST try this yourself…

Here’s what I show in the video:

✅ Create A Multi-Agent Dashboard : Install extensions and configure Cursor to operate with 3 different agents!

✅  Tips On Managing Multiple Agents: Learn how I wrangle 3 agents working at once to avoid confusion and potential conflicts.

✅ My Agentic Division Of Labor : Learn which agent I use for which type of work, why, and what the unique demands of the underlying models.

Nathan Binford
AI & Marketing Strategist

I hope you enjoyed this newsletter. Please, tell me what you like and what you don’t, and how to make this newsletter more valuable to you.

And if you need help with AI, marketing, or automation, grab time on my calendar for a quick chat and I’ll do my best to help!

Thanks For Reading H4CKER!