How do you diffuse AI into the real world? — Varun Shenoy, Long Lake

How do you diffuse AI into the real world? — Varun Shenoy, Long Lake

More

Descriptions:

Long Lake does not sell software. It buys the company. Thirty five services businesses so far across property management, architecture and HR, plus a 6.3 billion dollar take private of the world’s largest corporate travel platform. Varun Shenoy is blunt about why that matters: when the AI does not work, there is no customer to blame, because they are the operator and the owner. His framing for the whole problem is diffusion. Electricity was demoed in the 1880s and Ford’s electrified assembly line arrived in 1924, because it was never enough to have the technology, you had to rip out the motors and retrain everyone.

What he wants is the services equivalent of an asynchronous coding agent, and he is candid that nobody has it. Engineers happily launch ten jobs and accept that the seventh finishes before the third. Nobody clears an inbox ten emails at a time. Meanwhile the valuable tasks are absent from the internet entirely: closing books when receipts are missing, scoping a building from a blueprint, coordinating vendors to fix a roof. That work does come with ground truth, though, which is what makes their evals real. Did the roof actually get repaired. His closing argument is that none of this can be codesigned over Zoom, which is why the job involves showing up, running a stand at somebody’s trade conference, and asking questions on a mountain bike.

Speaker info:
– https://x.com/varunshenoy_
– https://www.linkedin.com/in/varunshenoy
– https://varunshenoy.com

Timestamps:
0:00 – Everyone has seen the demo, nothing has changed
1:30 – Electricity, Ford, and how long diffusion takes
2:36 – Who Long Lake is, and why they buy the businesses
4:54 – The ladder from copilot to coworker
6:02 – Earning the right to more autonomy
8:16 – Why engineers parallelize and nobody else does
9:24 – Representing knowledge work as code
10:32 – The valuable tasks that are not on the internet
11:42 – Traces, ground truth, and evals that mean something
14:00 – Continual learning and enablement as one loop
15:09 – The elephant: getting the first usage at all
16:19 – Codesigning in person, not over Zoom

1 Item

Channels