Y COMBINATOR · EXTRACTED
Building Faster with AI ft. Andrew Ng
Andrew Ng on the new bottleneck for startups, why concrete ideas beat vague ones, and how to ship 10x faster without breaking things.
Preview · 3 of 7 tactics
"As an executive, I'm judged on the speed and quality of my decisions. Both matter, but speed absolutely matters." — Andrew Ng
Andrew Ng founded Coursera, ran Google Brain and Baidu's AI group, and now runs AI Fund — a venture studio that builds an average of one startup per month. AI Fund cofounders are in the weeds writing code, talking to customers, and shipping product, which gives Ng an unusual operator's perspective on the AI tooling shifts. His Y Combinator AI Startup School talk is built around one thesis: execution speed is the strongest predictor of startup success, AI is the most significant speed multiplier in a generation, and most teams haven't adjusted their workflow to the new cost curve. The talk is unusually dense with specific tactics from his portfolio, several of which run counter to received Silicon Valley wisdom.
Make every idea concrete enough to build this afternoon
This is the central concept of the talk and the one most teams get wrong. Ng draws a hard line between vague ideas and concrete ideas. "Let's use AI to optimize healthcare assets" is vague — different engineers would do totally different things, and you can't ship it fast because there's nothing specific to ship. "Let's write software so hospitals let patients book MRI machine slots online to optimize usage" is concrete — engineers can start building it this afternoon. The deceptive trap, in Ng's framing, is that vague ideas get more kudos. Tell your friends you want to use AI to optimize healthcare assets and everyone says it sounds great. He puts it directly: "when you're vague you're almost always right. When you're concrete you may be right or wrong. Either way is fine. We can discover that much faster."
THE PLAY
Take your current top product idea and rewrite it twice. First version: keep it general enough that everyone you describe it to nods and says it sounds promising. Second version: make it specific enough that an engineer could start coding it today — name the exact user, the exact action, the exact interface, the exact integration. If your team has been working on the first version for more than two weeks without shipping, that's the symptom. The work is the rewrite.
Trust the expert gut over the data
This is the line in the talk that's most surprising coming from the head of an AI fund: data is often a slow mechanism for early-stage product decisions, and a subject matter expert with a good gut is often a better one. Ng says this works because someone who's spent years thinking about a problem — talking to users, watching competitors, building intuitions — develops a gut that functions as a high-quality, instant decision-maker. He cites his own Coursera origin: years of thinking about online education before launching, talking to users, refining his own intuitions about what would make a good edtech platform. He notes YC sometimes calls this "wandering the idea maze." After enough wandering, the gut is faster than the data and often as accurate. The corollary he flags: if every new piece of data causes you to pivot, you're probably starting from too weak a base of knowledge. Find someone who's thought about the problem longer.
THE PLAY
Audit your current decisions. Which ones are you waiting on data to make that you could honestly answer with your gut right now if you committed? For each one, set a hard deadline this week. If you have the expertise, decide on instinct. If you don't have the expertise, find someone who does — not someone who'll generate more data, someone who's been in the problem long enough to have a useful gut. The dangerous middle ground is collecting data slowly while pretending it's a decision.
Pursue one hypothesis doggedly, then pivot on a dime
Ng is explicit about something that runs against modern portfolio-of-bets thinking: a startup doesn't have resources to hedge across ten things at once. His framing — "pick one, go for it, and if data tells you to lose faith in that idea, pivot on a dime to pursue a totally different concrete idea." He describes AI Fund's actual experience as pursuing one thing doggedly with determination until the world tells them they were wrong, then changing direction and pursuing a totally different thing with equal determination and equal doggedness. The discipline isn't in keeping multiple options alive. It's in sustaining maximum commitment to one option while staying genuinely willing to abandon it when the evidence requires.
THE PLAY
Look at your current quarter's plan. Are you running one experiment with everyone's full effort, or are you running three experiments with a third of everyone's effort? If it's the second, consolidate. Pick the one with the strongest concrete hypothesis and put the entire team behind it. Set a clear falsification criterion at the start — what would convince you this isn't working. Then run hard until the data either supports the bet or hits the falsification line. If it hits, pivot fully. The middle path of half-committed experiments is the slowest path of all.
Subscribers Only
Unlock the Full Protocol
4 more tactics + Action Plan
TACTIC 04
Build twenty prototypes, ship insecure code
TACTIC 05
Build the portfolio of feedback tactics, fastest first
TACTIC 06
Treat tech stack as a two-way door now
TACTIC 07
Stack the AI building blocks combinatorically
Already subscribed? Log in
Newsletter
Get each new protocol the day it drops
One email per drop. No spam. Unsubscribe anytime.
Y COMBINATOR · EXTRACTED BY PODEX