AI engineer, building Nobs with @rmahamudg 🛠️ Two of us shipping AI agents,
automations and internal tools into production. Posting whascratch | #BuildInPublic
weekend path (no lab required):
1. buy the box for the job above
2. Ollama if it’s mostly you. vLLM if the team piles on
3. put the API behind your VPN
4. start 30B–70B quant. scale when someone’s actually waiting
early looks like failure if you’re measuring against the finished product. most of the first two years is shipping to almost nobody and still showing up.
6) start with the one thing that isn’t getting built
not “ai strategy.”
the concrete initiative stuck because you don’t have seats.
name it. scope it. ship it in *your* repo, behind *your* review — or consciously kill it.
AI generates a huge amount of code, and now the bottleneck is reviewing it.
So we’re also trying to automate the review with AI.
But a code review isn’t just for deciding whether the code is good or bad.
It also helps people learn, share knowledge about a feature, or teach a junior.
I’m worried we’ll automate that part without thinking about how we’re going to keep all the rest.
A decision made in slack six months ago is effectively lost.
then someone rebuilds the same thing because they can’t find the first one.
that’s not a culture problem. that’s missing search over your own work.