Astray

Feb 2026

When I dug into startups, every thread eventually led me to Y Combinator, and from there to Paul Graham. He cofounded it, but that's not what people love him for most. It's the essays. So I started reading. (The OGs were reading them way before YC)

And the more essays I read, the more I noticed a phrase that kept coming back. Lead you astray. He uses it in Before the Startup, a talk he gave at Stanford in 2014: “Startups are so weird, that if you follow your instincts they will lead you astray.”

Once I noticed it, I started seeing it everywhere. It's basically a warning to founders about the instincts they should resist. A few stand out:

Trying to consciously think of startup ideas.

Sit down to brainstorm and you'll produce ideas that are bad but plausible enough to waste a year on. YC calls them sitcom ideas. The fix is to stop trying, work on what interests you, and let ideas surface.

Building something a lot of people want a little bit.

You have to choose: a lot of people who want it a little, or a few who want it a lot. Pick the second. Dig a well, not a hole. Narrow and deep beats wide and shallow. Narrow means fewer people. Deep means specific to them, with an important and urgent problem you actually solve.

Avoiding things that don't scale.

The best founders do the unscalable work themselves at the start: meeting users in person, onboarding them by hand. If your instinct is to automate before you've done that work, you're skipping the part that actually teaches you what to build.

Avoiding tedious, messy, or hard problems.

Graham calls this schlep blindness. Your unconscious filters out the most valuable problems because they involve unpleasant work, which is exactly why they're still unsolved.

Studying the game, not the players.

The expertise you need isn't in startups. It's in your users, usually because you are one. Reading more startup books is itself the trap.

The pattern is hard to unsee. The instincts that worked everywhere else stop working in startups. I have a sense that AI is adding new ones to the list. Four stand out. I'm no Paul Graham, so these are just first impressions.

1/ Using AI to speed up the past.

The instinct is to use AI to build the things we were already building, just faster. Better dashboards. Cleaner SaaS. More polished screens. It feels productive. You're shipping more than ever. The rule is binary. If you're using AI to build stuff that was possible before, stop. If you're using AI to build stuff that wasn't possible before, go.

2/ Asking users to start from zero.

The instinct is to ask users who they are. Forms. Placement tests. Onboarding flows. “Tell us about yourself.” But users live in tools that already know them. ChatGPT remembers what they're working on. Claude knows their codebase, their projects, the way they like to write. Asking them to start fresh feels like handing them a paper form.

3/ Treating AI as a feature, not a foundation.

The instinct is to bolt AI onto an existing product. A sparkle icon. A “summarize” button. A chatbot in the corner. But these are half-steps, and they get outrun by products built native to what's next. The best products in this wave aren't AI-enabled. They're AI-shaped from the start.

4/ Designing for clicks when you could design for prompts.

The instinct is to build buttons and menus. Dropdowns. Filters. Toolbars. Settings panels. I'm a product designer, so it was my job to do that. But you can just describe what you want. The interface is the conversation. Less navigation. More intent.

This is what I keep coming back to. The instincts you've trained for years are usually the ones to question now. PG figured this out about startups. I think it's just as true about what comes next.

Final edits done with like-you-talk, my own Claude skill based on Paul Graham's Write Like You Talk (2015) and Write Simply (2021).