Fascinating concept; that a lot of applications are structured as "inboxes" which creates the psychological effect of "needing to action everything" even though there is no practical obligation https://t.co/Ao542LrJzk
Great piece on how the space you live in is far more than just "a space" and how living well in a space is closer to "sailing the building" than just existing https://t.co/LRcgPLBg5L
I've read 8,000 Y Combinator applications.
You would think the reviewers are thinking through a standard set of questions.
Is it a good idea?
Is it a large market?
Is it an experienced team?
Do they have traction?
During my time as a reviewer, we went back and tested which questions actually predicted success. The process was look at the best performing companies and then ask whether the application question would disqualify any of the biggest outcomes. If so, eliminate that question.
Is it a good idea? Well, Airbnb was a bad idea. Air bed and breakfast. Nobody thought it would work.
Is it a large market? Airbnb, Coinbase, Microsoft, Apple, Nvidia all started in tiny markets.
Experienced team? John and Patrick Collison had no experience in fintech when they built Stripe. Brian Chesky had never built a company. Brian Armstrong was just a product manager.
Do they have traction? We learned to invest in slope, not Y intercept. The initial traction doesn't tell you anything about the angle of ascent.
Sooooooooooooo...none of these standard questions worked.
The best question that predicted success was determination. Who is the most determined?
Even if they're not the smartest. Even if the market hasn't developed yet. Even if their initial product is wrong.
It outpeformed by a mile. And didn't disqualify the winners. AND was evaluatable.
It was so predictive that YC took that one question and found five different ways to ask it.
"How do you know the motor is running at full speed?"
"You hear a gear grind sometimes"
An imperfect but really useful model for thinking about how hard to push.
AI is messing with startup economics from a lot of different angles, e.g. cost to build product, size of product it's reasonable to build, type of value added etc. Interesting piece on what it means for B2B sales; https://t.co/fNT7ynU1KH
I've been gradually converted from "Event Sourcing is basically never the answer" to "Event sourcing MIGHT just be the answer sometimes", anyway this post is a super clear introduction to how it works; https://t.co/Er0Ftmj35U
I can't quite work out why there's so much effort going into trying not to believe that LLM's are good at writing high quality code? Engineering is still awesome, it's just a very different job to six months ago
It's a 153 page PDF but the Opus 4.5 System Card is really interesting; https://t.co/uPbHGDAHmj in particular multi agent combinations seem to be giving the types of bump we used to get from thinking tokens whereas the benefit of thinking tokens seems to be waning.
before you drag the "software engineering is over by next year" quote, consider Dario's "90% of Code will be made by AI in 3–6 Months" has rung true (for my own dev). No idea what next year looks like, but full-vibe-full-send could absolutely be the default (for better or worse)
got-oss:20b combined with open web UI is wildly good, like “I’ve stopped using Claude as my day to day bounce ideas off good” and then open chat UI’s OpenAPI spec approach to tool calling has me questioning why I’m using MCP. So that’s a thing.
My ElixirConfEU talk is live! I touch on the current AI dev landscape and live demo https://t.co/OoZmQToq8A.
We just left our punchcard era. For better or worse we're going to look back next year and think it's crazy we had to peck keys out by hand.
https://t.co/dAZ3ZNPoIs