I’ve posted a lot in here recently over the last couple of days and I’ve probably moaned a lot about the new review feature and the slowness of the Auto model, but wanted to say thank you for the free credits.
I’m taking a few months off on savings to build my own SAAS and upskill a bit and the $40 really helps me out as I can’;t really afford to drop $100’s on expensive models.
This has allowed me to use Composer-1 a lot more and churn out a lot more integrations with sites I want to scrape.
Thanks a lot guys it means a lot, I hope you manage to have a well deserved weekend off.
My advice: use top models like Sonnet, Gemini, or GPT for planning. Switch to Grok Code Fast or Gemini 3 Flash for execution. This helps me save more tokens. For big tasks, I break them down into smaller pieces to make work easier. Just my two cents.
thanks for the tip, I’m not really a huge fan of writing plans but I’ll give it a go.
I do a lot of full-stack development and have a “new-feature” or “new-integration” template checklist which I feed it which to some extent negates the need to plan.
To me the bottleneck is not so much clarity but the speed at which Auto greps/reads files etc. When you have to change the code in about 10 discrete places its makes it unbearably slow.
In another thread a staff member said its a known issue which I hope they’ll improve.
In my opinion, you should try adopting a ‘Plan Mode.’ As developers, we naturally follow this process, starting with requirement gathering, PRDs, and RFCs/TRDs to document what we are changing, why we are changing it, and how.
This structured approach makes it much easier for us to supervise AI and minimize hallucinations. This blog post might provide some useful references on why planning is so effective, especially since smarter models tend to be slower and require a more iterative approach
