Welcome back to Watch Me AI
If you’re new here, great to have you.
Watch Me AI is where I share practical ways to use AI in your work and life. Here are a couple of past issues to get you started: It's time to switch to Claude and "How can I make more money?"
OpenAI released GPT-5.5 recently, and I’ve been deep in it through Codex.
Codex is OpenAI’s coding agent and very similar to Claude Code. You give it access to a folder of files, tell it what you want to build or fix, and it works through the codebase for you.
This is not a tool most people need.
That is actually the point of this newsletter.
Whenever a new frontier model launches, the internet immediately becomes very dramatic. Everyone wants to know: Is this the one? Do I need to switch? Am I falling behind? Is Claude over? Is ChatGPT back? Do I need to spend my weekend migrating my entire AI life?
My honest answer: probably not.
Most people can carry on with their lives.
But if you are doing hard, technical, multi-step work, especially building digital products, GPT-5.5 is worth paying attention to.
Most people do not need GPT-5.5
Let me say this carefully, because I do not mean it in a patronizing way.
For most everyday AI use cases, you probably do not need the most powerful model available.
If you are using AI to draft emails, summarize meetings, brainstorm content ideas, plan a trip, organize your week, write a LinkedIn post, come up with dinner ideas, or think through a business problem, the top models are already very good.
Claude Opus is good. ChatGPT's prior models are good. Gemini is good.
At this point, for a lot of everyday work, the difference between the best models is less important than how much context you’ve given them, how well you know how to work with them, and whether you actually use them consistently.
This is not because GPT-5.5 is unimpressive.
It is because the baseline has gotten so high.
You do not need a Formula 1 car to drive to Trader Joe’s.
And most people are driving to Trader Joe’s.
Where I actually feel the difference
That said, I am not only using AI for everyday tasks.
I am a non-technical person who somehow keeps finding herself building digital products.
Client tools. Internal systems. App ideas, etc.
And in that world, I can feel the difference.
This is where GPT-5.5 through Codex has been impressive.
Not in a “wow, that answer was 7% better” kind of way.
More like: the work is moving faster, I am intervening less, and it is catching more issues before I have to point them out.
That is the kind of model improvement that matters.
What I’m liking about it
Here’s what feels different so far.
It has better momentum.
Codex is faster than Claude Code. This is partly due to speed (it's significantly faster on the same tasks) and partly due to autonomy.
Claude Code is very powerful, but it constantly pauses to ask for permission. I understand why. It is a good safety behavior. But in practice, it means I have to babysit.
Codex gets through more work before handing it back to me. That makes it feel much more useful for the way I actually work: give it a task, go do something else, come back to progress.
I often tell it "I'm going to sleep now, keep working" and it will work for 2-3 hours while I sleep without getting stuck. I've heard other peoples' go for 8+ hours.
The context compaction is a big deal.
With Claude Code or Opus, I can see when I'm approaching the context limit, and this is very stressful. It means I'll have to ask it to write a handoff brief, open a new chat, paste the brief, and then spend time reorienting it.
I always lose something in that process.
Codex compacts the conversation and keeps going. For long product builds, that removes the “oh no, this chat reaching its context limit” stress and improves quality.
It catches more issues.
This is the biggest quality difference I’m noticing. On a complicated client product, it is catching more bugs, flow issues, and “if we change this, that other thing breaks” problems before I have to point them out.
That is where model quality becomes visible.
Not in one perfect answer, but in fewer corrections over time.
Why is it better than Claude Code?
I've heard that Sam Altman's strategy of investing heavily in data centers (i.e. capacity) is paying off relative to competitors. Apparently Anthropic just doesn't have the compute power that OpenAI has, which means its products can't keep up.
I have not fact checked this. And I wouldn't be surprised if the next Claude model leapfrogs 5.5. The pace of all these companies is insane.
So, should you switch back to ChatGPT?
Here is my current answer: probably not.
If you are using Claude and you love it for writing, thinking, planning, strategy, or general business work, you do not need to abandon it.
For those use cases, model choice matters less than context, habit, and using the tool consistently.
But if you are using AI to build, debug, operate across files, clean up messy outputs, or carry a complicated project forward over hours or days, GPT-5.5 in Codex is worth testing.
That is where I’m feeling it.
Not in one perfect answer. In the accumulation of fewer interruptions, better continuity, faster progress, and less babysitting.
If this is the kind of thing you’re trying to sort out for yourself or your business — which AI tools actually matter, which ones you can ignore, and how to build workflows that save real time — that is exactly what I cover in the AI Business Playbook.
It is not about chasing every new model.
It is about knowing enough to use the right tool for the right job, and building practical AI systems into your work.
Get the AI Business Playbook here →
Until next time,
Mollie
