YC Liu 🌱

@lesbianbezos code is becoming the new debt. the best builders will figure out how to ship less of it, not more.

Coding is not the moat anymore. Creativity is. Leadership is. Taste is. We've always known this. It's just happening now.

I onboarded my non-technical team to Cursor today. This is the future. For the longest time, we've been playing telephone. Teammates creating tickets for engineers, who copy and paste whatever they said into Cursor.

@paulg four confirmation codes is also four conversion kills. fascinating that this still exists in 2026

Creating an account on a European auction site is so hard that sometimes I keep going just to see if it's actually possible. The Druot account I just created required four separate confirmation codes. But that was easy compared to the auction site in Italy...

Why I'm increasingly cautious about hiring. Years ago, a neighboring team decided to use a specific database as their embedding store, set a latency requirement of xx ms, and kicked off the project. I thought: this makes no sense. But my manager told me to stay out of it. Some bureaucracy thing — not our team, not our problem. A lot of bad decisions at the top aren't really about bad judgment at the top. They're about incomplete or wrong information getting passed up the chain by people who didn't understand the problem deeply enough. Executives can only decide on what they receive. The flip side: when someone watches you walk into a mistake and says nothing, it's not always because they don't care. It might be the same reason I stayed quiet — their manager told them to. This isn't just a hiring problem. It's an information flow problem. And it's baked into how most orgs are structured.

Naval: "AI is going to drain a lot of moats." Brand? Copyable. Scale? Two people with AI can out-execute a team of 50. Switching costs? Near zero when tools are free. What survives: relationships, trust, judgment. Those don't compound in your product. They compound in you.

This is my screen time after Openclaw 🥹

@cathrynlavery just added to AGENTS.md and looks pretty handy! hopefully the agents wont make the id up but lets see!

Openclaw tip 🦞 Your agents will lie to you. They'll say "on it!" and not start. They'll say "done!" and it's not done. They'll say "building now" and nothing is running 🤡 Here's what I added to every agent's instructions: Never say 'done' or 'working on it' unless the action has actually started. Every status update must include proof — a process ID, file path, URL, or command output. No proof = didn't happen. A false completion is worse than a delayed honest answer. Now when I'm told something is building, it comes with this:

Today's OpenClaw (by 5yo)

@CryptoKaleo gm!

gm to everyone still here

Making the list. Joining the club. Getting the nod from someone important. Every one of these is a hook. You think you're collecting achievements. Really, someone else is installing control interfaces on you. Every new "status" is another string they can pull.

People who chase goals will stop. Goal reached, game over. VP to SVP, SVP to GP, then what? Buffett is 95 and still in the arena. He was never chasing goals. He was chasing the process - learning, taking risks, being around interesting people.

A mouse dropped in water drowns in 4.5 minutes. Pull it out just before death, comfort it, drop it back in - same mouse survives 60 hours. Same body. Different brain. It now knows rescue is possible. First time your startup almost dies, you want to quit. After you survive one, your bar for "I can't do this" shifts in ways you don't expect.

Three things for startups: 1. Get the direction right 2. Solve real problems - even without AI 3. Distribution Number 3 is the most overlooked. Best product in the world means nothing without a channel.

When I started in 2023, the pitch was "one-person company, global growth." GPT had just dropped. "AI teammate" was still science fiction. Now I can actually see where AI products are heading. Makes it easier to stay the course. Pick a direction you'd spend 5-10 years on. The specific product will change. The direction shouldn't.

Pre-OpenClaw AI agents were glorified Q&A bots. Now it's different. My AI teammate reads PRs, answers questions directly. Async communication got noticeably faster. Cross-timezone teams - even just using AI as a messenger already makes a huge impact.

Lock in the direction, not the product. Nobody gets it right on day one. Half the YC companies are probably pivoting right now. OpenClaw dropped and triggered another wave. Pivoting is normal. NOT pivoting is weird. As long as the problem you're solving is real, the product form can keep changing.

History books focus on emperors or timelines. I prefer the fiscal lens: treasury operations, tax systems, explaining rise and fall. Every empire follows: fiscal surplus → fiscal expansion → debt expansion → credit collapse → repeat. History doesn't repeat, but it rhymes.

Two architecture patterns for controlling AI bots: Option A: Agent handles read/write, LLM manages security itself. Option B: Full read/write separation. User-facing agent can only read. All writes go through a higher-privilege backend. Like a puppeteer - someone pulling strings, agent has limited autonomy. Which is better? Still validating.

@CryptoKaleo absolutely true

#Bitcoin / $BTC I still believe this is the best buying opportunity for Bitcoin since early 2020. Waiting for another dip to start stacking is just greedy at this point.

AI startups used to rush to RAG - vector search in databases. Now with Claude Code leading the way, everyone's rushing to text-based approaches. Insight: memory and knowledge management needs a middle ground. Vector search isn't a silver bullet. Neither is plain text.