Unique Research · Insight · November 10, 2023 · Adapted from a16z Connect/Enterprise — Marc Andreessen & Martin Casado · Original: a16z.com/building-products-with-generative-ai/
At a16z’s annual Connect/Enterprise, generative AI and LLMs dominated. Below is an editorial digest of how founders should view genAI from a product lens.
Preface — we don’t know the trajectory yet
GenAI’s future is wide open — but from product and UX, we cannot yet predict the path: which interfaces win; which tasks LLMs will own vs. transform; how new models redefine what we expect of computers. The moment rhymes with the rise of the internet and smartphones. Early apps excite and hint at second-order uses that could remake personal and professional life — still largely a mystery. Marc Andreessen and Martin Casado dig into how startups should inspect genAI as product opportunity.
GenAI as a computing paradigm shift
Architecture shifts recur; each births a new compute platform. They open new apps and experiences — and expose incumbent fragility. Giants with users, brands, assets, and headcount still must re-examine business and tech stacks. Past parallels: mainframe software ecosystems vanished in the Unix/client-server shift; a generation of enterprise apps vanished with web and cloud; even infrastructure-layer cloud left many computer makers behind.
For startups this is a double opportunity: invent on the new architecture, and rise while large firms are unstable. We cannot be 100% sure of the genAI scenario — but have strong reason to believe it is a powerful opening. Computers are unlocking truly new capabilities — LLM-class abilities once seemed impossible within a lifetime. Every prior machine, mainframe to smartphone, was a deterministic von Neumann programmed system. LLMs introduce probabilistic computing — the first probabilistic computers: same input twice, possibly different outputs. That “hallucination” is also “creativity” — challenging accuracy/reliability while opening a world where computers are not only tools but somewhat creative.
First principles for a new platform
New platforms (internet then; LLMs now) release energy by enabling what old platforms could not. They are stronger in some dimensions and limited in others. You would not want an LLM as a calculator — it may get 2+2 after many tries, inefficiently. Yet it can write love poems to enterprise software in Shakespearean style. Founders must drop prior assumptions, reason from first principles, explore unique value, and confront new problems. Open, experimental mindset is the key.
A golden age for entrepreneurial thinking
Adobe shows incumbent adaptation: AI in Photoshop/Illustrator (smart prompts, generative images), copyright issues addressed so users can freely use AI images. AI integrated into the whole workflow — yet traditional photo-edit tools might still go obsolete if AI is simply faster. That gap is the founder opportunity: new ways to handle photos and generate art without manual edit — technically feasible and commercially valuable. We still don’t know how best to use these systems or how they hit existing software categories. If the thesis holds, this revolution brings an entrepreneurial golden age unseen since the mid-1990s — requiring courage, invention, and adaptation.
AI’s future: the fight for control
Early Microsoft pushed “Blackbird” against HTML and owned MSN (originally a dialogue network). Strategy: extract the “vig” — a cut of every transaction (target ~3% of internet commerce). Without pushback they might have succeeded. In AI today the question remains open. One camp wants the vig — control of the ecosystem via a platform and plugin-like APIs, lobbying Washington for regulatory blessing as lawful monopoly/cartel. Another camp wants AI open-source, open, free, downloadable, runnable — with no one dictating allowed use. Which future wins is unresolved — and must be faced together.
— END —