Sam Garvin

Strategy & Ops @ Luxor | Building @ BenchOS

[Sam Garvin @SGarvin20]

Use TCR every day and am finding this is also so true anecdotally. The ability of Fable to one shot something not only *feels* different, but you can also clearly see it in “real world” evals like this. A bunch of cool analysis a tool like TCR can unveil about model capabilities!

[Tenki Cloud @TenkiCloud]

Here's a quick test we did: Opus 4.8 vs Fable 5
This is what Tenki Code Reviewer found:

Opus-written PR:
3 high-severity bugs across 22 files and 1,615 lines of code

Fable-written PR:
zero issues across 22 files and 1,134 lines of code

Same repository, same prompts.
Small sample, but this is worth paying attention to.

[Sam Garvin @SGarvin20]

This is the first thing I've shipped that points in this direction. It won't be the last. If you coach, or know someone carrying a program alone → https://t.co/WypyEQDIt1

[Sam Garvin @SGarvin20]

The most underrated metric in product onboarding is "time to wow."

Not time to value. Not activation rate.

The raw seconds between "I signed up" and "oh — this is different."

Miss that moment and the best product in the world still churns more than you think

[BenchOS @GetBenchOS]

Onboarding to BenchOS used to mean filling out forms about your program.

So we taught Coach Prime to just... ask you instead. 🎙️
Early look at voice onboarding 👇 (sound on)

[Sam Garvin @SGarvin20]

The human keeps judgment + direction + oversight. The agent carries the load. The human defines point B, the agent takes the mental load of getting from point A to point B.

[BenchOS @GetBenchOS]

The BenchOS pilot launches today 🏀

An AI coaching staff for basketball programs that don't have one. Five specialists, one coordinator, every output grounded in your program.

Now onboarding pilot coaches for free during the beta

https://t.co/CHnKdEQlfI