ashby

Ex-McK → PE → YC → Operator (Fin/Ops @ Stablecore).
Dad to three, plus a dog, wedded to my high school debate opponent. ☧

[ashby @af3]

We're all talking about what AI means for software. The real question is what it means for everyone - and everything - else. https://t.co/G8Xxj06heB

[ashby @af3]

https://t.co/jXjWh04ybX

[ashby @af3]

@acaseofthegolf1 @tbpn would love to see this level of coverage for vesting schedules

[ashby @af3]

@normonics You may be interested to learn more about Catechesis of the Good Shepherd: https://t.co/wnknM9gmcw

[ashby @af3]

@bermchain Followed.

[will o’brien @Willob]

Another takeaway after three years of building a company is that the returns to just "being a good person", and not acting transactionally, are higher than I expected.

There are so many transactional, boring, and at times immoral people in the tech industry that if you act with sustained warmth, authenticity, humor, and integrity, great gifts will be laid before you. White pill.

[ashby @af3]

@casualclubguy Category nominee for risky but beautiful course watering aoots: Lincoln Park (SF) #17

[ashby @af3]

@chadmumm 5.5 playing out of Mizner is my guess. (There’s a homonymous 25.7 at Coral Ridge but I have my doubts there.)

[Will Young @wynot1234]

I am so bullish and optimistic about what AI is going to enable us to do. Golden age is coming 🤞

At the same time I 100% agree with @Pontifex that we need to cherish and celebrate the creative forces that are fundamentally human

These two things are complementary

[ashby @af3]

@rcbGT Funny AND true

[ashby @af3]

A question I keep wrestling with this year is: what will be valued when intelligence is too cheap to meter?

[Hassan @nutlope]

I used Jev to classify 1,018 AI research papers.

The result: $0.08 total cost and 256ms median end-to-end latency per paper.

The pipeline was:

1. Summarize each paper with DeepSeek V4 Flash
2. Send the title + summary + 24 possible topics to Jev
3. Use Jev to classify each paper
4. Visualize everything on https://t.co/hs63SlHxjw

The summaries cost $3.99 on @togethercompute. The classifications cost $0.08 on @typesafeai.

So for just over $4 of inference, I ended up with a pretty useful way to explore the top AI research papers from the past year.

I think this is where things are heading: different models for different parts of the workflow, instead of using one model for everything.

I’m running evals on the Jev classifications before replacing the current ones, but the site is already live: https://t.co/hs63SlHxjw

[ashby @af3]

@SHistorians

[bravo🇻🇪 @venehzuelah]

It’s beautiful

[ashby @af3]

@RhysSullivan My guess is that Type 1 companies will have to have some network effect for a durable advantage.

[ashby @af3]

@maiab Wealthfront x Frec would go very hard

[ashby @af3]

@shanebacon My three-year-old son on the 9th of Charlie Yates: “Dad, I did them all!” Kid gets it.

[ashby @af3]

This is elite chow. Burger dog of the east.

[Ross Martin @RossMartinNC]

Finley Golf Club, UNC’s home course in Chapel Hill, announced that Merritt’s will now be its new food and beverage partner.

Second location for Merritt’s. Open to everyone, playing or not. The two locations are just 2.5 miles apart.

[ashby @af3]

@JD_Patrick Followed. Where would you get these made?

[austin petersmith @awwstn]

i lost the lid to my coffee grinder and codex tracked down the part, found the dimensions, and made me a 3d print file that fit perfectly in one shot

[Waffle House @WaffleHouse]

We don't believe in Belgium waffles

[Gokul Rajaram @gokulr]

TIME TO RETIRE THE CLASSIC PRD

The PRD is a relic. Time to retire it.

The word "PRD" brings to mind a 10-page Word doc or spreadsheet with a Background section, a Goals section, a Customers section, and 7 pages of feature descriptions. Half of it is throat-clearing. The other half is so vague that engineering and design end up guessing what to build anyway. Nobody reads it twice. Most of the time nobody reads it once.

Replace it with the Product Spec.

A Product Spec is a different kind of artifact, designed for the people who actually consume it (engineers, designers, AI agents)

The Product Spec has four mandatory pieces:
• The problem: who is hurting, what they are doing today, why now
• The bet: a falsifiable hypothesis (if we ship X, then [specific user] will [observable change] within [time], measured by [metric])
• The success criteria: what concrete behaviors we will see when this is working
• The evaluation: how we will measure it, what the kill / scale / graduate thresholds are

The shift from PRD to Product Spec is structural. It is what Coach is designed to enable.

In the PRD era, the bottleneck was getting alignment from a room of humans.
Long docs and exhaustive sections were the price of that alignment.

In the agent era, the bottleneck is giving an agent or an engineer a tight enough specification that it can ship without 12 follow-up questions, ideally using the goal loop in your agent of choice (h/t Peter Yang)

A 10-page PRD fails that test. A 1-page Product Spec with clear acceptance criteria and evals passes it.

Founders and product leaders: stop calling your docs PRDs. Stop writing them like PRDs. The artifact you need is a Product Spec with a falsifiable bet, a specific problem statement, concrete acceptance criteria, and a measurement plan.

[ashby @af3]

This product recall email from a recent summer camp is the stuff of fever dreams