Build in Public | 7+ yrs of Engineering Exp | Data Lead | Day 10/x to quit 9-5
Building Ekome - Journal that helps you visualize your thoughts evolution.
Everyone wants AI coding agents to "ask the one clarifying question that matters."
I built it. Ran 120 trials.
It lost to asking nothing at all.
And the part that decides whether to ask cost more than just asking everything.
Everyone wants AI coding agents to "ask the one clarifying question that matters."
I built it. Ran 120 trials.
It lost to asking nothing at all.
And the part that decides whether to ask cost more than just asking everything.
The scary part isn’t that the AI boom is expensive.
It’s that revenue itself can become part of the financing mechanism.
At some point, you have to ask: are we measuring demand or measuring capital being recycled through the ecosystem?
AI infra is doomed...
This is the latest circular financing map from today’s Bloomberg article. There is no mystery why the revenue numbers are so good, even as losses expand, free cash flow turns negative, and off-balance sheet leverage explodes.
Also, the Bank of International Settlements (BIS) Annual Report released in late June is a sober look at all this.
I have spend the last week and a half going through all these filings post-earnings, and I am looking forward to tearing apart more 10Qs from Coreweave and Nebius soon.
From what I have seen it is worse than this.
@elonmusk will be industry leading bcz of so much tweet data, even gibrish data is going to make advantage over others, while distillation bound coming up.
Everyone wants AI coding agents to "ask the one clarifying question that matters."
I built it. Ran 120 trials.
It lost to asking nothing at all.
And the part that decides whether to ask cost more than just asking everything.
I actually can resonate with it recently i did a experiment where i measured the gate's cost, not just its accuracy. It asked 3.7x fewer questions than asking everything and cost 25% more, because running the check costs more than the questions it saves.
now given pathways route, i am actually aligned that with assumption awareness it's feasible for sure.
We've raised a $143M Series C at a $1.5B valuation.
In 2023 we spent our time arguing that AI code needed its own reviewer. That argument is over. The work now is moving into the change itself.
This is where we think Agentic Change Management comes in. https://t.co/TAeW1c1W9k
Pilot-then-freeze, yes. Should have done that.
Rolling adaptation during the run, no. If the threshold moves as trials accumulate, trial 1 and trial 100 came from different policies and the aggregate stops meaning anything.
Also neither touches the check cost. That's structural.
btw do you mind taking it further and see how it works.
would love to hear how it went.