~/talk

Build Something Agents Want, Not More AI Slop

slides:none on file

Argues human taste must filter agent output and shows concrete tools and numbers: an open‑source recorder tapes.dev that saves traces to SQLite with observation.md and state so sessions become reusable data; a Pokémon speedrun experiment that spins up ten agents inspired by AlphaEvolve and reaches Professor Oak in ~3 seconds after context capture; a workflow that surfaces 77,000 sessions, converts solutions to one‑page skills, and runs a fine‑tuned 4B QWin model alongside parallel sweep agents to automate only the well‑vetted work.

AI-written summary · about

  1. 0:00Introduction and AI slop
  2. 4:41Volume problem and learning
  3. 10:12City infrastructure and Goodhart's law
  4. 18:35Building agents what they want
  5. 25:59Tapes and agent workflows
  6. 31:54Skills and specialized fine-tuning
  7. 33:41Taste and content strategy
~/search
type to search 6,518 conferences, 143,918 speakers and 192,133 talks
~/help
cmdctrl k
search every conference, speaker and series
/
focus the filter search
h
home
c
conferences
s
speakers
e
series
t
light / dark — the status bar button does the same
← →
on a talk: 5 seconds back / forward
0–9
on a talk: jump to 0%–90% of the recording
?
this help
esc
close this, then clear every filter — or just the search, while you are typing in it

Ignored while you are typing. Pages without a filter rail have no search to focus.