🤖 Don't Write Off OpenAI Just Yet

I spent over an hour last week just talking to ChatGPT in one session.

Not typing.

Talking.

With the new "Live" audio feature, we laughed, we workshopped ideas, and we actually started coming up with a thesis for a new open-source project I want to build.

It felt less like using a tool and more like having a strategy session with a really sharp co-founder.

There's been a lot of noise lately about OpenAI slipping behind, but after this week, I'm telling you: do not write them off yet.

The GPT-5.6 release is amazing.

And it's changing how I work right now.

Here is what you need to know about every recent update and how I'm using it to power my next move...

🎧 The Magic of GPT-Live

OpenAI just launched GPT-Live, a new generation of voice models that changes the paradigm of how we talk to AI.

Key Facts

🗣️ Full-Duplex Architecture - It can listen and speak at the same time. You can interrupt it naturally, and it doesn't get confused.
🧠 Delegation for Deep Work - For complex questions, GPT-Live delegates the heavy lifting to the frontier model in the background while keeping the conversation flowing with you.
👀 Visual Answers - It can now pull up visual cards for things like weather or stocks right on your screen while you're talking.

Previous voice models were turn-based.

That means, you talked, you waited in silence, it talked.

If you coughed, it thought you were done. It's been annoying since it's been available.

GPT-Live fixes that.

It acknowledges you with "mhmm" or "got it" while you're speaking.

It waits when you pause to think. It doesn't cut you off. And it even will talk with you.

Honestly, it's the closest thing to a real human conversation we've seen from an AI that truly keeps a conversation valuable.

🤖 Codex Pro is Building My Startup

I've been pushing Codex HARD since it was released. I am completely in love with how well it works and, more importantly, how autonomous it is.

Key Facts

⚡️ Full Autonomy - I've upgraded to Pro, and it's currently building my new startup.
🛠️ Heavy Lifting - It's creating a server farm, connecting APIs, and building dashboards.
⏳ Unattended Execution - I can leave it working for 30 minutes at a time, completely unattended.

I'm not quite ready to announce my new startup, but I plan to sooner than I though was humanly possible.

Because that term, ‘humanly possible’ aged, and now it's no longer quite correct.

For those of us fully adopting agentic AI, it's now what the human can instruct agents to build with them.

This is the shift we've been waiting for.

Codex isn't just auto-completing code anymore. It's executing complex architecture while I step away to handle other things.

When you can trust a system to run for half an hour or even overnight without supervision, your leverage as a builder goes through the roof.

💡 Sol, Terra, and Luna: The GPT-5.6 Breakdown

With GPT-5.6, OpenAI moved away from confusing version names and introduced a three-tier model family.

Here is how you should be using them:

→ Sol (The Flagship):

This is the heavy hitter. It's designed for your hardest tasks, complex coding, deep reasoning, and long-horizon agentic workflows.

It introduces a new max reasoning effort and an ultra mode that actually spins up multiple subagents to work in parallel.

Use this when the ceiling on quality matters more than the cost.

→ Terra (The Daily Driver):

This is your balanced, everyday model. It offers performance competitive with GPT-5.5 but at roughly half the cost.

For 90% of your daily workflows, Terra is the sweet spot.

→ Luna (The Speed Demon):

This is the fastest, most affordable tier.

It's built for high-volume, latency-sensitive tasks.

Don't let the price fool you; it still outperforms older models on many engineering benchmarks, but it does it at a fraction of the cost.

🧠 How to Prompt GPT-5.6

With a new model comes new rules.

The way we prompt needs to adapt.

OpenAI just published the official playbook 🔗 for prompting the GPT-5.6 model family, and the folks over at AI Edge translated it into plain English.

Here are the biggest changes you need to make:

→ Use Shorter Prompts.

Replacing long, explicit system prompts with minimal ones improved scores by ~15% in OpenAI's internal tests, while cutting total tokens by 60%+. Less is genuinely more.

→ Define Autonomy Clearly.

GPT-5.6 is proactive.

Be clear about where you want the model to stop and what it actually has authorization to pursue.

And don't repeat "ask first" over and over - it triggers unnecessary permission checks even for safe actions.

→ Avoid Generic Instructions.

Instead of saying "be shorter," say "lead with the conclusion.", tell it your underlying goals, not surface-level preferences.

→ Keep Structure Lightweight.

Give it a small amount of task-specific structure, not a massive response template imposed on every answer.

👉 TLDR: 👈 Strip away the over-engineered prompts you built for GPT-4. Give GPT-5.6 clear boundaries, a specific goal, and get out of its way.

Ask yourself these two questions today…

If you had a system that could execute your instructions for 30 minutes completely unattended, what would you hand off first?
Are your current processes built for tools that need constant hand-holding, or are you preparing for systems that act on their own?

None of these announcements are revolutionary on their own.

But together, they paint a pretty clear picture of where AI is heading.

We're moving away from prompting and toward collaborating. Away from assistants and toward autonomous teammates.

And I think that's one of the biggest shifts we've seen yet.

🎙️ One More Thing

As of writing this, my team is at absolute capacity. I'm brining on more team members to offer more professional services, because we are building software, training brains, and supporting some incredible entrepreneurs and organizations.

So I have one ask.

If you have a podcast, click this link 🔗

I want to reach out to you before I officially launch this incredible new startup I'm working on to show you first.

It's going to be special. I can't WAIT to tell you all about it.

P.S. VIP day spots are open now for Aug! (link)