Code is nearly free. Judgment is not.
Finding, spiking and killing ideas with an agent fleet, and the one-person company that comes out of it
For most of my career the expensive part of a venture was building it. A prototype meant a team, a quarter, and a budget line someone had to defend. That constraint set the pace of everything: which ideas got tried, how many, and how long a bad one survived before anyone could prove it was bad.
That constraint is gone. A working version of most software ideas now takes days, built by agents that run overnight while I sleep. What did not get cheaper is knowing what to build, what the test is, and when to stop. A techventurist is a technologist who treats that judgment as the job and runs ventures as a loop instead of a bet. This pillar is the loop, written from the ideas I have spiked, sold, or scrapped on this site and the others I run.
The Techventurist loop
Five stages, one rule: the date is set before the spike goes live
- 1
Signal
A cost curve just moved. Something that took a team last year takes an agent this year. The signal is the gap between the old price and the new one, not the technology itself.
- 2
Thesis
One claim, written so it can fail. "People searching for X will pay for Y" is a thesis. "AI will change the industry" is a mood.
- 3
Spike
A working version in days, built with agents, live on a real URL. Not a deck, not a waitlist. Real data with real edge cases, because fake data never breaks the way customers do.
- 4
Sell
Proof of demand before any scale work: search traffic, a paid pilot, an inbound request. Traffic that costs nothing to earn is the cheapest proof there is.
- 5
Scale or Scrap
A date is set before the spike goes live. On that date the numbers decide. Most spikes get scrapped, and the write-up of why is the asset that carries into the next one.
What this pillar covers
Four threads, one argument: more spikes, shorter lives, better write-ups
The Techventurist Loop
Signal, thesis, spike, sell, scale or scrap. How each stage runs when the building is done by agents and the scarce input is judgment about what to build and when to stop.
Finding Ideas Worth Testing
Where startup and AI business ideas actually come from: cost curves that moved, searches nobody answers well, and workflows still priced as if people did them. Plus the ones I rejected and why.
A Venture Studio, One Person Wide
What venture studios and venture builders do, and how much of it a single technologist with an agent fleet can now run: sourcing, spiking, and killing ideas on a schedule.
Fleet of One
The one-person company in practice. AI employees, an agentic workforce, and the supervision rhythm that keeps it honest. Sam Altman bet on a one-person billion-dollar company; this is what the ordinary version looks like.
The venture log will run the same way the training and biohacking pillars do: on logged data. Each spike gets a start date, a kill date, and the number that decided it. Nothing here is backfilled or estimated. The log goes live when the first spike under this pillar reaches its date.
Running the same kind of loop?
If you are deciding what to build next, or whether an agent fleet can carry a venture on its own, that is the work I do as a fractional CTO and AI strategy advisor.