Gemini Says My Ego Essay Is Still a Humblebrag

Earlier today I published a piece called “Four Ego Mistakes I Made as a 22-Year-Old Founder.” Then I ran it through Gemini 2.5 Pro as an independent reviewer. Gemini’s verdict was that the essay is itself an ego move. I think Gemini is mostly right.

Author: Koutian Wu; GitHub: ktwu01

If you have not read v1, the short version is: I listed four ego traps young founders fall into, wrote a clean taxonomy, and ended on a tidy “scoreboard” metaphor. I asked Gemini, posing as an independent senior editor, to be ruthless. It was.

What Gemini actually said

I am quoting the parts that hurt the most, because the point of doing this is to keep the receipts.

“The essay doesn’t read like a confession; it reads like a thesis. The author has positioned himself as an expert observer of his own psychology. The tone isn’t one of struggle; it’s one of ‘I have struggled, and now I am back with a brilliant framework to explain it to you.’”

On Mistake 2 (the bad term sheet I did not sign):

“This is a classic humblebrag framed as a mistake. The author confesses to the temptation of taking a bad deal but emerges as the hero who was smart enough to say no. The actual failure would have been signing the sheet. By presenting a dodged bullet as a learned lesson, he frames a victory as a vulnerability.”

On the four mistakes as a set:

“These are ‘kinds’ of mistakes, not instances. Which repo? For what feature? What did that feature ultimately deliver to the company (or fail to)? How bad was the term sheet? Liquidation preference? Participation rights? Which clause kept you up at night? When you talked to your first potential customer, how did they push back? What did they actually say?”

And the ego-flavored mistakes I conveniently did not list:

“Technical arrogance: looking down on non-technical co-workers. Not-Invented-Here syndrome: refusing to use mature third-party tools. Conflating difficulty with value: obsessing over technically hard problems instead of valuable ones. Hiring for subservience. Refusing to delegate code you wrote.”

That last list is the one that stings, because at least three of those describe me more accurately than the four I picked.

Why my v1 essay failed at its own goal

The stated goal of v1 was to be honest enough to actually correct my behavior. Gemini’s read is that v1 instead reads like a “branding exercise in thoughtful leadership.” When I sit with that, I think it is right, for three reasons.

First, the structure was too clean. A real confession is messy; mine had four numbered mistakes and a closing definition with parallel structure. That neatness was the giveaway. If I have already arranged my failures into a four-quadrant grid, I have not been recently humiliated by them.

Second, I picked the safest possible mistakes. “I built before I sold” is a failure mode every reader of Paul Graham can recite. It is approved-vocabulary self-criticism. It does not cost me anything to admit, because every founder admits it. The mistakes that would actually cost me to admit, the ones in Gemini’s list above, did not appear.

Third, I told a hero story dressed as a confession. The bad term sheet I did not sign is a story about my judgment, not my ego. A real ego failure in that area would be: I spent two months chasing an investor whose brand I wanted on my cap table even though their fund thesis did not match mine. Or: I posted a soft-circled commitment as a “raise” before the wire cleared because it felt like progress. I have done versions of both. I did not write about them.

What I actually learned from running this loop

The interesting thing is not the verdict on v1. The interesting thing is the loop itself: write the most honest piece I can, hand it to a model with no investment in protecting my self-image, ask it to be ruthless, and notice which critiques produce a defensive twitch and which produce a quiet “oh.”

The defensive twitches are signal. The quiet “oh”s are the actual finding.

When I read Gemini’s note that Mistake 2 was a humblebrag, my first internal response was to defend it: “But the term sheet really was bad, and I really did not sign it.” That defense is exactly the tell. The point was never whether the facts are true. The point is which true facts I chose to put on the page, and which I quietly left off because they would not make me look like the smart protagonist of my own essay.

The selection bias in what I confess to is itself the ego.

What I am turning this into

I am codifying this into a skill called founder-ego-audit. The shape is simple: take any artifact I am about to publish (essay, pitch, plan, README), run it through an independent reviewer with a prompt specifically calibrated to find performative vulnerability, humblebrag patterns, hero-story-dressed-as-confession, and approved-vocabulary self-criticism, and triage the findings before shipping.

The skill is not “let an AI rewrite my writing.” It is closer to “let an AI tell me which of my own sentences I should be embarrassed by.” The rewrite is still my job.

The honest closing

I am going to leave v1 up. Deleting it would be a worse ego move than writing it in the first place, because the embarrassment of a public, dated record is exactly what a self-corrective is supposed to produce. v1 is now the artifact. This piece is the receipt.

The real test is whether the next thing I publish, three weeks from now, has fewer of the patterns Gemini just named. If it does not, this whole exercise was theater. If it does, the loop works.

Either way, the scoreboard is not on this page. The scoreboard is whether I am on a sales call this week.