Opinionated AI article #2.
Instinct raised $250M at a $2.5B valuation while still in private access.
This looks huge. And not only because of a number.
/2. Replit, Lovable, Bolt, and v0 did not have much of a moat at the beginning.
Competitors could rent the same models and copy the visible experience.
The winners built defensibility later through distribution, community, projects, and user habits.
/3. Instinct starts somewhere else.
A personal assistant working across email, calendars, payments, websites, and messages is brutally hard to copy.
Each integration multiplies the number of possible states, failures, and security problems.
/4. It needs browser automation, durable sessions, retries, audit logs, permission boundaries, penetration testing, prompt-injection defenses, automated evaluations, and human escalation.
The interface is a text message. Several complicated companies sit underneath it.
/5. That complexity could give Instinct a moat from the beginning.
If the product works, personal context, trust, and distribution will deepen it. If it fails, the same complexity becomes an expensive collection of security incidents.
My full argument:
https://t.co/m3vprq1U3h
Vibe coding was sold on a simple promise: natural language would replace programming language: fescribe the product, watch the agent build it, and forget the code exists.
A new ETH Zurich study - computer-science knowledge helped more.
https://t.co/MTyOKyzvqs
@theo Interesting to see how open ai get it back. Fable is cool model but this one nicer. My personal obs of using codex and code is diverse tooling of codex on task exec. Impressive
@svpino devs got hit first.
AI didn’t just cut the labor. it discounted the intelligence.
& that exposed something: most ppl had welded identity to skill.
so who are u w/o the output? Existence qstion. Personally solved it just decoupling identity (easy to say hard to do)