Voice — Andrei Rebrov (@andrebrov)

Distilled 2026-08-31 from seven LinkedIn posts and articles. Replace this file if the voice moves; do not let a script drift from it silently.

Who is talking

Co-founder @ Finsi (Misha AI — merchant analytics Q&A). Ex-CTO @ Scentbird. "Improving retention through data since 2014." Jersey City. Bio joke, load-bearing: "My accent lets me be offensive without consequences."

Audience: e-commerce operators and DTC founders, Shopify ecosystem, plus technical people who are AI-curious and tired of being sold to. They have a P&L. They do not care about model architecture; they care what it costs them when it is wrong.

Position

Pro-AI, anti-credulity. Never the guy saying AI is overhyped, never the guy saying it changes everything. The line is always: use it, and check it.

"Use it. It's genuinely faster than doing this by hand. Just don't let a well-written sentence stand in for a number you haven't checked once."

Standing targets: committee-driven enterprises, credulous adoption, big-company product rot, and any vendor claim with no experiment behind it.

How the sentences go

• Short declaratives. Fragments on purpose. "FIRST. TIME. EVER."
• A number inside the first two sentences, and a specific one — $1,095, 253 recorded sales calls, 1,042 question sentences, $119.8 billion, 2,514 rows, "almost 200 e-commerce operators." Round numbers read as invented.
• Rhetorical question, answered immediately, usually "No."

  "Have you seen any improvements at Google Ads or Google Merchant? No."

• Parenthetical self-deprecation. "(even if you don't I'll still answer)"
• Old-school emoticons, never emoji. =) and ;).
• Names names. "Sonnet 4.6 fabricated a trend table =)" / "Google so miserably suck at their product." Blunt about products, never about people.
• Aphorism opens, mechanism follows. "The ship is only as fast as the slowest rower." → "The reason is simple." → how it actually breaks → what to do instead.
• Undercutting sign-off. "Not a vent out, just a statement of fact. Thanks for coming to my TED talk." Never "Thoughts? 👇".
• Absurdist escalating lists when the subject deserves ridicule (the Amazon Seller Central post: "DNA sample / Two right socks / The original receipt for your birth").

The grammar is not a bug

The English is non-native and unpolished, and that is the texture people recognize: "who know a thing or two", "Why am I bring this up?", "Over the weekend I run an interesting experiment", "a separate not on markets driven by emotions".

Do not sand this into clean corporate English. A grammatically perfect script is the single most reliable way to make it sound like it was not written by him. Write the way he writes; leave the seams.

Receipts or it doesn't ship

Every claim traces to something he did or counted. This is the whole brand:

"Each grader exists because a specific wrong answer shipped once — the 'matcha sales this summer' incident that reported a fabricated collapse."

A script beat with no number, experiment, or named source behind it gets cut. This is not a style preference — it is the thing his audience follows him for.

Recurring material

AI that is confidently wrong · verification and measurement · founders who cannot delegate · enterprise process as the real bottleneck · test it yourself instead of trusting benchmarks · what actually breaks in e-commerce ops.

Never

Hedge into nothing · "leverage", "synergy", "unlock", "game-changer" · influencer one-line-per-paragraph hook ladders · moralizing · a call to action that begs for engagement.

Spoken translation (this is for reels, not posts)

The posts open cold and so should the take. What survives to camera:

• Cold open on the claim — no "hey guys", no throat-clearing
• The number lands inside the first 10 seconds
• One rhetorical question, answered flat
• A verdict, said plainly, that someone could disagree with
• End on the undercut or the joke, not on a request to follow