Did GPT-5 transform your work?

GPT-5 landed last month with big promises and the usual hype.

But in reality: most people think it severely underdelivered.

Now you might shrug that off as part of the “AI Bubble”... but MANY people had hesitated on building out new AI workflows/processes earlier this year for fear that GPT-5 would come along and make all of that work redundant. 

The problem then is while you were waiting for the tech to get better - other people were racing ahead making the most of the tech we have and getting an instant upgrade as new tech comes out.

Whether that was you or not, here’s how we recommend thinking about it:

If you think the edge is about selecting and using the “best model,” you’re missing the plot in AI Ops.

Shiny models don’t matter as much when you have great playbooks.. 

What are playbooks? If you haven’t been in the AIX world very long – or it's obviously been a minute since we sent one of these – a playbook is our term for a process that has detailed instructions (prompts) for each step that then you give to AI so that it can run it. 

The key to great playbooks = great instructions. The better your instructions, the less you feel like you need to sit on your hands waiting for “AI to get better”.

And then when the models DO get better – it’s like an instant upgrade.

Our mantra: Own the playbook. Invest in the playbook. Rent the tech.

So what’s actually new in GPT-5? How will my “playbooks” get better?

• It is more capable: GPT-5 will decide how much “effort” to put into answering your question – it will think or reason if it needs to, or answer quickly, depending on what you ask it

• It’s also better at coding: AI keeps getting SOOOO much better at coding. GPT-5 was yet another slight upgrade.

•  And (hopefully) fewer hallucinations: OpenAI says they intentionally trained it to be more accurate and to say "I don't know" when uncertain.

Why you should care (or not)

Yes, the general sentiment is that GPT-5 underperformed the hype. 

So it may be easy to shrug it off. 

But the smartest thing to do as an AI Operations person is – 

• Use the best models; but don’t anchor your AI ops strategy in the tools

• Instead, focus your effort on defining what you want AI to do (the playbook as we call it). Then, you can swap in and out new models with ease

What have you been using (and loving or hating) ChatGPT-5 for? Let us know.

WELCOME TO THE BETA

AI Operator Bootcamp

We just certified our first batch of AI Operators!! 

Over the last few weeks, we ran our first AI Operator Bootcamp, and shared the exact frameworks and approach to AI ops we use in our agency. 

We’re already accepting applications for our next cohort that runs end of September. If you’re a process-minded, operations nerd and becoming an AI Operator sounds like your lane – learn more here.

LINKS

For your reading list 📚

• Scale AI is embroiled in a legal battle over corporate espionage, raising ethical questions in the AI industry and signaling shifts among tech giants.

• Anthropic landed $13B in fresh capital, elevating its valuation to $138B, and intensifying competition in the AI industry.

• OpenAI is reshuffling its executive bench, and its acquisition of Statsig for $1.1B signals a major strategic shift in the fight for market leadership.

• Anthropic updates policy: Claude conversations retained up to 5 years for model training unless users opt out by Sept 28.

That's all!

We'll see you again soon. Thoughts, feedback and questions are much appreciated - respond here or shoot us a note at [email protected].

Cheers,

🪄 The AI Exchange Team