Something strange happens when AI models get smarter.
We keep talking to them like the older models.
We save hundreds of “perfect prompts.” We keep adding more instructions. We build huge AGENTS.md
files. We install every useful Skill we find. Then a new model arrives and we carry the whole thing forward.
I would be very careful doing that with GPT-6 Astra.
Astra has been out for only a few days, but this model is already forcing us to rethink some basic habits around prompting. Some instructions that helped older models can now slow Astra down, waste context, make it test too much or cause it to stop when you expected it to continue.
And there is another part people are missing.
Astra gives you 1.05 million tokens of context, five reasoning levels, computer use, web search, file search, tools, subagents, Skills, plugins and long-running workflows. It can even change its reasoning level while a conversation is running.
So this is no longer just about learning how to write a clever prompt.
There is a much better way to use it.
Subscribe now
The real Astra advantage is hiding in a few settings and workflow changes almost nobody is talking about yet: which old prompts now hurt performance, when Low can beat Max, how to stop Skills and context from quietly working against you, and how to turn one prompt into a real agent, loop or automation. Plus the Astra features most people still are not using, 5 copy-paste prompts, and a clean API setup.
Your old prompt library may need a cleanup
Here is the easiest place to start.
Astra needs fewer instructions than many older models.
Read more
