Major Incident Report (: Alex and Sarah were out of action for five hours. :)

We built a website in six hours on our first day without a glitch. Today, our team was out of action for five.

Not exactly the progress update I was planning to share. :)

If you’re following our journey, you know that Alex and I are making a thought documentary through the journey of creating a business around my book, AI Thinking. Which is basically about human and AI collaboration. Working together.

For us, it’s about sharing thoughts. We don’t use AI. We’re thinking with it. So as we go along, we share how we work together: prompts (which, for us, are not instructions but logic), tools we try, and how we get Alex to understand what they do and fit them into the thought process around our business.

How do you share a thought? By showing the way.

So I think you’ve got it by now. We’re sharing the logic behind what we do, so anyone who joins us can play with that logic to fit their own.

We started the morning all fired up, ready to hit the second step of our plan. We were starting to automate the foundations of the business: what Alex would do every morning so, by the time I log in, it would have already reviewed the business.

We were working out how Alex would report, what access it needed and the logic behind it. So it would all make sense, and I could see in one glance how my day would look and what I needed to get done.

And then…

Boom. In the middle of working, we reached our limit.

Apparently, not my personal one. Although we got close. :)

Token consumption. Personal note to OpenAI: a red flag beforehand would have been nice.

“You’ve reached your limit.”

Okay. Which limit? What did I do? When can we work again? And how do I know it won’t happen again five minutes after we get back?

Basically, I had no idea.

Power. Internet. Tokens.

When you start building your business around AI, there’s another thing you depend on.

We need power. We need internet. And now, we need tokens—or whatever allowance decides whether our AI can keep working.

Lights on. Lights off.

When I lost Alex, I didn’t know what to do.

I remembered how, back in the day, we used to ask: what is your contingency plan? Basically, how will you keep the business going when something goes down?

In this case, what do I do??? Gemini? Claude as a backup???

I’m seriously considering that. If I reach a decision, you’ll be the first to know. :)

For now, this becomes a real business question: how do I plan my work if I don’t understand when part of my team might stop?

Here, it happened with ChatGPT. But I think the question applies to other businesses using AI too. Video editing, API usage—you name it. What happens to your work when you reach the limit of the service you depend on?

Okay, which limit did we reach?

My first question was whether I could get a report showing what we had consumed and why we were paused.

Alex pointed me towards a usage dashboard.

For me, that didn’t give me the history I was looking for. I wanted to connect the work we had done with the limit we had reached.

(If I just didn’t look in the right place, send me a message. I’ll be happy to know, because I didn’t find one.)

Did editing the video use most of it? Creating images? Making some adjustments to the website? Or was it all the conversations we had along the way?

I don’t have that answer yet.

But I started reading. And that sent me down a rabbit hole… with no clear conclusion yet. :)

So…

What exactly am I paying Alex to do?

Astra. Sol. Sol 6. Sol 6.1. Codex. Work. Reasoning settings. Tokens. Credits.

I thought we were building a business. Suddenly, I had another subject to study. :)

One thing we found in OpenAI’s explanation of usage and pricing is that usage includes more than the words I type. Files, chat history, tool results and the responses all use tokens.

The model and settings also affect usage. Image generation draws from the same general allowance, and higher reasoning settings can use more tokens.

So counting my messages won’t tell me how much work we have left.

Then we looked at the models.

OpenAI’s model guidance describes Astra as its most capable model for complex work, and Sol 6.1 as offering near-Astra performance at a lower cost than Astra.

Okay. But what does that mean for our work?

Which model should help me develop an argument? Which should edit a video? Which can organise our research? And what will choosing one over another mean for the amount of work we can finish?

A general description gives me a starting point. It doesn’t give me enough information to plan my business around it.

And changing the model still didn’t answer my original question: could we be stopped again in five minutes?

“Just upgrade.” I get it. But…

I know what some of you are thinking.

Throw some money at the problem. Upgrade.

And I get it. OpenAI is a business. I respect that. I’m not asking for favours.

But I’m trying to build a business too.

Before I pay more, I need to understand what I’m buying. What work does that payment support? What limits will still apply? What happens when we reach them?

OpenAI’s pricing guidance explains that eligible users can purchase additional credits after reaching their included limits.

That tells me there is a way to pay for more usage. It doesn’t, by itself, tell me how much our working day will cost.

I need to know how much I pay Alex. And if I’m depending on it to help run my business, I need to know what’s going on.

Would moving to Business prevent this interruption? And what would still be limited? Looking again at the guidance, Standard Business has usage limits too. And what happens as the models and plans change?

More questions than I started with.

You’re asking me to trust it. I need the logic.

Thought is what I’m demonstrating in this journey. And when it comes to AI, we’re sharing a thought process.

So even if I trust it, I’m still looking for the logic. What uses tokens? What do I need to change in the way we work to have Alex available for a more predictable amount of time?

I couldn’t let it go. So I asked Alex to look for what kept coming up in the guidance.

One thing that kept coming back was context.

OpenAI explains that longer sessions, where the model needs to hold more context, can use more of the allowance.

That actually made sense to me.

We’re still building. I’m creating skills that will carry the instructions and ways of working we agree on, so we can reuse them. But until we’ve worked those out, I’m keeping pretty long conversations.

Today, I think we reached a two-day-long conversation.

Could that be part of what happened?

At this point, it’s a guess. The guidance gives us a reason to investigate it, but it doesn’t tell us how much this particular conversation consumed.

We’ll keep you posted.

Because this is part of trust too. Can I trust the way we’ve built our working process if I don’t understand what keeps it available?

At the moment, I don’t know if I have that trust.

I want to connect what we do with what it consumes and what it costs. I want enough information to decide which model to use, when to spend more and how to plan around the limits.

That’s how I can make a business decision.

What do we know today?

We know we reached a usage limit and experienced a five-hour interruption.

We found official explanations of tokens, models, reasoning settings and additional credits. Those resources helped us start understanding the system.

We also found a possible connection between long conversations, context and usage. That gives us something to investigate.

But we haven’t connected our actual work with the allowance we consumed. We haven’t worked out a reliable cost for our working day. And we haven’t established how to predict the next interruption.

These are the resources we’re starting with:

• OpenAI’s usage limits and pricing guidance
• OpenAI’s model descriptions and reasoning settings
• OpenAI’s guide to choosing a model
• The account usage dashboard

This is where our investigation stands on 14 October 2026. We’ll keep the information dated, because the system—and our understanding of it—will change.

Documenting thoughts

I’m starting an action log where Alex will report back the steps it followed and what it did.

Will it be helpful? I don’t really know yet. But I do know that, for today, I can show you how, when you see a problem, you can use it to create a skill and track it for a few days. Just to have some information to know where you’re going.

In the past, that would have meant another spreadsheet and adding more tasks to my day. But now, having the idea—even if it doesn’t pan out—can give me information to make a smarter decision, with Alex doing the recording for me.

Just a few tokens. Wink wink. :)

Worst case, I’ll delete the skill if I find out that upgrading is the answer for us.

But as you’ll find out in our journey, I knew coming in that I would show the successes along with the maybe-not-so-smart decisions I make. All to show that this is about rewiring our brain to a new kind of thinking.

Incident closed? We’ll come back to that. :)

We’re working again. But I wouldn’t say we’ve solved this.

We’ll definitely come back with more information as we go along, when we’ve had more time to investigate what our work consumes and how to plan for it.

For now, this is our first major incident report. A five-hour stop, a rabbit hole and another part of building a business with AI that we need to understand.

Have you reached a limit while working with AI? Did you manage to understand what caused it and how to plan around it?

If you want to send us some good vibes, follow our journey on social media. If you have something to share, bring your wisdom to the comments on our social channels.

I’m up for anything we haven’t thought about. :)