Technical Direction Without Execution Goal | Generated by AI
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:
-
You know the direction, but maybe not the final destination.
- You are clearly interested in AI infrastructure, LLM training, GPU systems, agents, open-source models, hardware.
- You keep spending effort on things like MI50/ROCm, nanoGPT, SGLang, training experiments, GPU clusters.
-
The uncertainty is about the product/business outcome, not the technical path.
- “I want to understand GPU kernels, train models, build agents” is a direction.
- “I will build X product, for Y users, and make Z revenue” is a concrete goal.
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:
- Train models from scratch
- Build agent frameworks
- Do AI consulting
- Build hardware lab
- Study low-level GPU
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:
- Right: you need a sharper external objective.
- Wrong: you are not directionless; you have a strong technical direction. The missing piece is converting technical exploration into a specific outcome.