The decision memo PDF: social-media-presence-2026-08-22.pdf.

Verdict

Do not build a personal brand on X, LinkedIn, or Substack. Keep a narrow, owned public surface. Use social as a pipe (paid ads, one-shot product notes, inbound replies), never as a self.

Do not send the Instacart-agent tweet. That is not building in public. It is publishing a half-formed secret into a litigated market.

The instinct — “seems like a losing move, don’t want to compete with the metric crowd and poison what I’m thinking, and share secrets” — is more right in 2026 than it would have been in 2020.

Why the posting playbook does not apply

Hormozi, Naval, Graham, Altman, and Chesky are a biased sample: they had a working machine, a product whose customers want to watch the founder, and status they could spend. Claims Genie asks for Gmail and files other people’s money. The live secret is a people-secret (platforms treat personal agents as bots; the consent rail is that they should not). There is no celebrity surplus. An owned surface already exists.

The posting-wins confound, in one line each:

• Hormozi built Gym Launch first (~$17M profit year one). Content is now the job.
• Naval was already AngelList. How to Get Rich is a late-career media asset. He also tells people to ignore status games.
• Graham owns the URL. The essays predate Twitter’s scoring system. He wrote the two best arguments against living on the feed.
• Altman inherited PG’s distribution, then ran the most-watched lab on earth.
• Chesky had Airbnb. In July 2026 his X account was hijacked to post AI-slop. Celebrity-on-X is an attack surface.

Packy’s 2020–22 “write in public → syndicate → fund” path existed because Twitter still had a human graph. He later wrote Burn the Playbooks: once a method is playbooked, AI and clipping farms eat it. a16z’s “It’s Time to Build” is about production. The last line is “what are you building?” not “what are you tweeting.”

Why silence fits

Thiel, Zero to One, Chapter 8 (Secrets), not Chapter 5. Every great business is a conspiracy. Share the secret with fellow conspirators — employees, users who must trust you, a few investors — not 40k drive-by accounts.

Early Stripe spent ~two years with production users and almost no public theater. The Collison installation, not thought leadership. Closest analog.

Early Palantir treated public narrative as a liability. A founder who is always posting is a diligence red flag in fiduciary rooms.

Graham, “How to Lose Time and Money” (July 2010): fake work that bypasses the alarm. Drafting tweets feels like work. It is the time equivalent of a bad investment. “We’ll increasingly be defined by what we say no to.”

Failure modes for this founder

Idea theft is overrated. Strategy leakage, audience capture, and legal speech are not.

• CFAA-adjacent. Amazon v. Perplexity (9th Cir. 4 Aug 2026) is the Instacart-agent fact pattern. Tweeting “here’s how an agent should authenticate as you at Instacart” is a prosecutor-friendly summary.
• Audience capture. C. Thi Nguyen: simplified scores take over practical reasoning. The metric crowd will not reward “quiet fiduciary that files $18 claims accurately.”
• DoNotPay lesson. FTC order 11 Feb 2025, $193k. Public AI-legal-finance claims are a known enforcement pattern. Discovery will include the founder’s public posts. PROMISES.md already forbids typical-outcome dollar claims. A personal feed of legal theories is the same class of speech, unfiltered.
• 2026 slop. Eight AI bots wrote ~50% of visible Community Notes on X (May 2026). Founder Twitter is now a recognizable genre of slop.

The tweet, graded: not a finished public fact; does not recruit, fundraise, or demo; is the company secret; is CFAA adjacent; attracts the wrong users. Do not send it.

What to keep

1. Product as the only loud surface.
2. Talk to 20 users.
3. Private memos (this format).
4. research.digitalsurfacelabs.com as a slow archive. ≤4 pieces/year. Gate exploits, user data, legal advice.
5. One long essay per year, owned domain, only when it has a job.
6. Company account that is not him. Counts, not typical-dollar promises. Can be quiet for months.
7. Dormant LinkedIn. Profile hygiene. No carousel habit.
8. No Substack.

Default is no. A post must recruit, fundraise, demo, correct a false claim, or support users at scale — and must not be a half-formed secret, a legal theory, user data, or a likes-driven pellet.

Instead of the tweet: the Instacart idea is now a private briefing. If it is a product bet, schedule the work. If it is a question, ask two users whether they would let an agent log into Instacart as them with an inspectable consent receipt. That conversation is the company. The tweet is a pellet.