Sponsored by
Hey folks,
I have not had an active ChatGPT subscription in 6 months. I've been a Claude guy, because I believe their models for development have been superior during this time and the other frontier assistants have been playing catch-up. Until now.
OpenAI shipped GPT-6 Astra, one of their co-founders went on record saying it might be the start of AGI, and my feed filled up with people doing things I had not seen a model do before. So I headed back to ChatGPT and re-subscribed on entry level at $30 a month for testing.
I have run one prompt in Codex. One Shot seems to be the rage for testing. In 21 minutes it built me a working retro ICQ Mac app, and it ate 93% of my 5 hour usage allowance doing it.
Let's dive in..
What I asked for
I don't have much history inside Codex. But I saw a cool use case of rebuilding Codex itself, and I wanted to give it a try.
So I asked it to build me a live Codex ICQ client from the 90s. A real Mac app, in Swift, that takes my actual Codex chats and projects and presents them as a 1990s instant messenger. Buddy list. Chat windows. The lot.
I set the model to high effort, 1 shot prompted it, and went to do something else.
21 minutes later..
My ICQ Mac app was running in 21 minutes.
In that time it wrote and compiled a working Swift application on my laptop and I did not touch it once. No follow up prompts, no "actually can you fix", no nudging it past a compiler error. I came back and it was built and launched on my MacBook, with an ICQ icon sitting in the tray.
Then I opened it and got the part I genuinely was not expecting.
It was wired to my real data from Codex. My projects, every conversation, sitting in the buddy list like old contacts. Click one and it opens in a chat window. It is fully interactive and it works properly, which is a low bar to describe and a high bar to clear from a standing start.
I have been around software long enough to know what I skipped. Scaffolding a Swift project. Working out how to read the data. Building the window, the list, the state handling. Then the loop where you build, it fails, you read the error, you fix it, you build again.
That loop is where the hours actually go, and it happened without me.
What it cost
That one prompt consumed 93% of my 5 hour usage limit. Which sounds brutal until you do the maths.
Codex runs on a rolling 5 hour window, shared between what you run on your own machine and what you run in the cloud. So one app cost me one window. In a working day you realistically get through 2 of those, maybe 3 if you start early and keep feeding it.
Call it 2 Mac apps a day, on the entry level plan, for $30 a month.
There is a weekly cap sitting on top of the 5 hour window, so you cannot run that flat out for 30 days and come out with 60 apps. But even a fraction of that number is a strange thing to be able to write down. A year ago the constraint on building a small thing was my weekend. Now it is a usage bar that refills while I have lunch.
The real constraint is not cost, it is choice. You do not casually fire off 4 Apps. You get roughly one good shot on GPT-6 Astra per window, so pick something worth spending it on, and let it build.
My take
Stop judging an AI on the quality of its answers.
Judge it on how long it can be left alone to complete a task.
The interesting thing about Astra is not that it is smarter. Every model is smarter than the last one, that is the least surprising sentence in technology.
What changed is the unit of work. I did not have a conversation with it. I gave it a job and walked away for 20 minutes. 6 months ago that would have been constant back and forth. Today it was none.
And the honest caveat, because it matters: an ICQ client is a toy. No legacy codebase, no other engineers, no compliance review, nobody's revenue riding on it. That is the easiest possible version of this job. One prompt on a greenfield app tells you a model can go the distance on a clean run. It tells you nothing about what happens on week 3 of a real project.
But I have been building software a long time, and the gap between "I would like an app that does this" and "there is an app on my machine that does this" has been the expensive gap for my entire career. It just got a lot narrower.
Slow billing doesn't just create friction — it delays cash and compounds across every deal you close.Slow billing doesn't just create friction — it delays cash and compounds across every deal you close.
Most SaaS finance teams assume the gap between contract signature and first invoice is a minor inconvenience. The Tabs Billing Lag Calculator puts a dollar figure on it.
Answer 5 quick questions about your contracts, invoicing process, error rate, and DSO, and the calculator benchmarks your billing lag against top SaaS companies — then shows you exactly what it's costing you.
Two minutes. One number that's hard to ignore.
Calculate your billing lag and see where you stand.
Next: the ICQ sound - Uh Oh.
The app is missing one thing and it is the only thing that matters. There is no ICQ sound when a message arrives, you know; Uh Oh..
Anyone who used ICQ knows exactly which sound I mean.
That is my next prompt, to bring back the nostalgia.
Cheers, Tim
