@garrytan I have been loving gstack so far. One thing that could be imporved is the way it creates and review plans. Right now it just adds everything to TODO.md but I much prefer how superpowers does it with dedicated plan files which makes it easier to run through review and handoff.
Having AI do most of my job for me, while also having the existential fear of it replacing me, while also having more inbound requests to join companies than ever before is fucking strange.
AI has entered the self acceleration loop.
@DarioAmodei says he has engineers telling him they don't write code, they let the models write the code. We have AI writing the next generation of AI systems.
We are at the begin of the acceleration curve.
@lessin Wrong, product lead growth and saas is booming.
Narrative, trust and community have always been important.
The only thing thatβs changed is more competition and more customers.
People typically fall into two camps as it relates to using AI to create products.
1. You stay permanently high level. You are a vibe coder you ship fast but don't understand deeply what you are building typically leading to the wrong thing unknowingly.
2. You stay permanently low level, insisting on understanding every line and your throughput hits a hard ceiling.
Neither of these approaches work anymore you need to do both and understand when to utilize what.
This intuition comes from understanding what you are trying to build and why. Zooming into technical details when its warranted and out to high level agentic system design that ladders up to high level product vision and strategy.
This is the new "full stack".
This just in, execution is now cheap. Agile is dead.
Timidity is the new risk to hedge against.
Not building enough things towards a larger vision that is really transformative for the customer.
@ThePrimeagen on the @claudeai vs @opencode drama
"You don't start a land war in Asia and you never go head to head with open source on the line with a vegan" π
AI is forcing everyone to replace the comfort of control with the discipline of systems. Most companies wont be able to do it, the ones that do are going to leapfrog the competition by orders of magnitude.
If all of your #ai experiments are random sandbox tasks (e.g. "summarize this doc") you are doing it wrong...
AI is about compounding workflow gains. It's a path-design problem yielding exponential benefits based on chosen workflows.
Hint: map key junctions in your org's information flow.