Technical Direction Without Execution Goal | Generated by AI

Home 2026.07

Question: Friend said: “Very practically speaking, you don’t know what you want to do.” Is that right?

Answer:

Partially right, but the wording is too absolute.

A more accurate interpretation:

Your friend probably means:

“You are exploring many interesting technologies, but I don’t see a single execution target.”

For example:

Exploration:
GPU -> ROCm -> training -> agents -> robotics -> circuits -> models

Execution goal:
Build an AI coding agent for backend engineers
Get 100 paying users
Improve model/tool loop every week

The first is learning. The second is a mission.

The risk is not “you don’t know anything.” The risk is too many valid paths:

All are reasonable, but they compete for time.

A good founder usually has:

North Star:
    What problem am I solving?

Experiments:
    Which technology reduces uncertainty?

Infrastructure:
    Which skills/assets compound?

Your GPU experiments are valuable if they answer a business question:

“Can I build cheaper AI infrastructure / better agents / specialized models than existing solutions?”

If they only answer:

“Can I understand ROCm internals?”

then it is research/hobby, not necessarily company progress.

So I would say your friend is 50% right:


Back Donate