Building an AI company: lessons from a $1.5B founder at 27

Executive overview

Most AI job-displacement narratives are too simple. Whether AI eliminates, reassigns, or creates roles depends entirely on which of three modes a company is in: growth amplification, quality improvement, or cost reduction.

Jesse Zhang co-founded Decagon, which automates customer conversations for Hertz, Duolingo, and Notion. In two years the team has grown to nearly 200 and is still hiring fast.

Non-technical founders have more leverage than ever — but analytical thinking and clear communication matter more than coding.

How AI is reshaping customer support roles

• About one third of companies using AI agents are actively downsizing outsourced call-centre agencies; one third are focused on quality; one third are in growth mode and not replacing anyone.
• Tier-one queries (password resets, room bookings) move to AI; humans shift to tier-two and tier-three complex interactions.
• New roles are emerging: "conversation architect" or "AI architect" — designing how AI should behave, written in plain English.
• Former CX managers and chatbot owners are upskilling into these roles through programmes like Decagon University.
• The nature of jobs changes with every technology shift; this is not unique to AI.

Skills that matter in an AI-first workplace

• Analytical thinking: ability to break a conversation into steps, trace AI reasoning, and identify where it went wrong.
• Communication: writing clear, non-contradictory instructions in natural language is how you program modern AI agents.
• Jobs writing copy or producing straight output are at highest risk; the role evolves into guiding the AI rather than disappearing entirely.
• AI tools used internally at Decagon: Cursor, Claude Code, Lovable (prototyping), Deep Research, ChatGPT for go-to-market research.

Building with AI without a technical background

• Being technical is not required, but it gives better intuition and faster decision-making.
• Non-technical founders now have access to tools (Lovable, n8n-style block builders) that were unavailable a few years ago.
• The audience for building tools has expanded from "super technical" to "semi-technical or not technical at all."
• If you have the time and interest, learning to code is still worth it.

How Decagon found and validated its market

• Start with customer conversations, not product — talk to large businesses first because they have scale and can co-iterate.
• Work closely with large enterprises to refine the product; only productize for smaller clients once the product is mature.
• Signal hierarchy in B2B: revenue is the purest signal; second is a named dollar commitment ("if you deliver this, it's worth X to us").
• Classic sales discovery maps directly onto early-stage founding: understand how customers make decisions, whose budget it affects, how they evaluate vendors.
• Don't over-index on other founders' playbooks — different strengths and circumstances mean different paths.

Advice for founders and graduates

• If you have conviction and energy, going straight into a company is valid — but expect it to be hard without prior intuition.
• If you choose to work first, join a post-product-market-fit startup so you can see what "good" looks like in people, customers, and product.
• Don't build anything until you've mapped what customers will actually pay for.
• Get explicit commitments early, not just enthusiasm.