Directions Brain

2026-01-30-alonhuri-linkedin-full-stack-ai

5 min read

Alon Huri — Sell hammers or open a carpentry shop? Full-stack AI startups (January 2026)

Source: raw/articles/2026-01-30-alonhuri-linkedin-full-stack-ai.md

Note: LinkedIn post by alon-huri (Managing Partner at team8); ~4 months before the 2026-05-30 capture; date approximated to 2026-01-30 (Friday) per Alon’s stated weekly Friday cadence. Hebrew original; load-bearing claims translated to English here. The companion post to 2026-03-06-alonhuri-linkedin-saas-is-dead-ai-native-agency — same thesis, different angle, written ~5 weeks earlier.

Summary

Alon picks up Jared Friedman’s (YC) framing: most founders try to sell AI tools to existing companies (e.g. selling to lawyers); the Full-stack AI founder does the opposite — don’t sell software to dinosaurs, make the dinosaurs irrelevant. The Full-stack AI startup begins as a human-staffed service company (with Growth-First customer acquisition and a Proprietary Data Loop), then progressively replaces 80%+ of the human labor with AI to reach 5–10x cost-and-speed advantages over the incumbent firm. Two structural moats: (1) the proprietary data loop generated only by doing the work; (2) the systematic distribution engine. The challenge framing he offers: where does AI not just imitate a human, but provide capabilities a human cannot supply at all? — that’s where category-creating Full-stack AI companies live.

Key claims

  • Most founders pick the wrong starting point. Most try to sell AI tools to existing professional firms — lawyers, accountants, agents. Per Jared Friedman of YC, the Full-stack approach inverts this: don’t sell software to the dinosaurs, replace the dinosaurs.
  • Full-stack AI definition. In software, “full-stack” means owning the whole technical chain. In AI, full-stack is business: don’t build a product that helps someone else do their work — own the whole service chain and do the work yourself. “Don’t develop software for mortgage advisors. Be the mortgage advisor.”
  • Two moats once you’re the service provider.
    1. Technological — the data loop. When you actually perform the service, you generate proprietary data that no third-party software vendor sees. This data closes a loop that sharpens the model’s processes, decisions, and quality over time. (Alon notes he wrote on this separately a few months prior.)
    2. Business — the distribution engine. How do you bring customers? Not by luck. You build a systematic acquisition machine that doesn’t depend on a star sales rep — same Growth-First insistence as in 2026-05-16-alonhuri-linkedin-ai-native-growth-hacking and 2026-03-06-alonhuri-linkedin-saas-is-dead-ai-native-agency.
  • The uncomfortable truth: start as a services company. “Almost always, this kind of company has to start as a services company. A literal Agency, fully owning the result.” Initial work is mostly performed by humans — not because the tech doesn’t exist, but because you have to learn the work end-to-end: where the value is, where the mistakes are, where the gold is hiding. “Look the customer in the white of the eye.”
  • The migration goal. Don’t stay a services company. Migrate to ≥80% of the work performed by AI.
  • The bar of disruption. “We are not looking for 30% efficiency improvement — that’s nice for a tooling product. We are looking for 5x or 10x more efficient, cheaper, and faster. Only that kind of leap justifies starting a new company that replaces an existing industry.”
  • The category-creating question. “Where does AI not just imitate a human, but provide capabilities a human cannot supply at all? It’s much more than availability or politeness. New categories appear when the customer gets something that doesn’t exist today, simply because no human can supply it at this price and speed.”
  • Bottom line. “A Full-stack AI company is a company that starts as a human service, builds proprietary data and a strong distribution engine, and grows until it can deliver the same service almost without human touch and 10x better than the market.”

Notable quotes

  • “לא מעט סטארטאפים מנסים למכור פטישים משוכללים לנגרים. חברת Full-stack AI מחליטה לפתוח נגרייה ולמכור ארונות.” — Quite a few startups try to sell sophisticated hammers to carpenters. A Full-stack AI company decides to open a carpentry shop and sell cabinets.
  • “אל תמכרו תוכנה לדינוזאורים, תהפכו את הדינוזאורים למיותרים.” — Don’t sell software to the dinosaurs, make the dinosaurs irrelevant. (Citing Jared Friedman, YC.)
  • “אנחנו לא מחפשים שיפור של 30% ביעילות… אנחנו מחפשים פי 5 או פי 10 יותר יעיל, זול ומהיר. רק קפיצה כזו מצדיקה הקמת חברה חדשה שמחליפה תעשייה קיימת.” — We are not looking for 30% efficiency improvement… we are looking for 5x or 10x more efficient, cheaper, and faster. Only that kind of leap justifies a new company that replaces an existing industry.
  • “איפה AI לא רק מחקה בן אדם, אלא נותן יכולות שבן אדם לא יכול לספק בכלל?” — Where does AI not just imitate a human, but provide capabilities a human cannot supply at all?

Why this matters for the team

  • The companion to 2026-03-06-alonhuri-linkedin-saas-is-dead-ai-native-agency. That post named 15 vertical professions; this earlier post is the strategic framing for why the Agency model exists. They should be read together; both anchor full-stack-ai-vertical-services.
  • Reframes the team’s “build a Brain for X vertical” question. Where the team’s current direction asks “which vertical’s workflows do we own as a software product,” Alon’s question forces a sharper version: “which vertical do we replace — and is the Brain the back-office of our own services company in that vertical?”
  • Most aggressive cut of the existing thesis. This is the most ambitious version of vertical-use-case-led-brain: not “Wonderful for X” but “be X.” Different capital-intensity profile, different go-to-market, different exit shape (services-firm-with-AI-margins is a less-trodden M&A path than software).
  • Skepticism caveat. Same as all Alon sources — team8 partner; thesis maps to Team8 portfolio bets; treat as advocacy, not neutral analysis.

Mentioned in

Graph View

Backlinks