Why Eric Schmidt Is Right (And Why It Should Make You Nervous)
The former Google CEO just told founders to "go build an agentic AI company" if they want to make money. Here's why he's probably right — and why that advice is also a warning.
Eric Schmidt doesn't give investment advice on Instagram for fun. When the former Google CEO speaks at an AI summit, people with billions of dollars listen.
His recent message: Found an agentic AI company if you want to make money.
As someone running a digital solutions business (Cognify Tech), this hit different. Let me break down why Schmidt is right, where he's incomplete, and what it actually means for builders right now.
What Is Agentic AI Actually?
Most people think AI is a chatbot. Ask a question, get an answer.
Agentic AI is fundamentally different. An agent doesn't just answer — it acts. It can:
- Browse the web autonomously and synthesize findings
- Write and execute code without human approval at each step
- Manage multi-step workflows across different tools
- Make decisions with incomplete information and iterate
Think: "Book my client flight from Darwin to Singapore next Tuesday, compare prices across 5 airlines, and send me the best option with a calendar invite." — and the AI just does it.
That's agentic. And it's the difference between a calculator and a junior analyst.
Why Schmidt Is Right About the Money
Here's the economic logic:
The SaaS model is commoditizing. Everyone has a SaaS tool. Competition is fierce. Margins are压缩ed.
Agentic AI creates new work that didn't exist before. When you automate a workflow that previously required human judgment at each step, you're not just saving time — you're creating capacity for entirely new use cases.
Schmidt (and Andrej Karpathy, for what it's worth) are pointing to a genuine asymmetry: the tooling layer for agentic AI is still immature, which means the highest-value applications haven't been built yet.
For a small operator like Cognify Tech, this is the window. Large enterprises are too slow to move. Individual developers are building hobby projects. The gap is in the middle — building agentic solutions for specific business workflows where off-the-shelf tools don't exist.
Where Schmidt's Advice Gets Tricky
Here's what the "just build an agentic AI company" framing glosses over:
Reliability is the product. The whole reason most AI demos look impressive and most AI products fail in production is reliability. Agentic systems compound errors across multiple steps. A 95% accurate single-step AI is useful. A 95% accurate 10-step agent has ~60% chance of completing the workflow correctly. That's not a product — that's a liability.
The moat question. Anyone can call an LLM API. The question is what you're building on top. If your "agentic" product is just a better prompt wrapper, you'll be undercut the moment Claude or GPT ships the same capability natively (which they will, within months).
Legal and liability. When your AI agent makes a mistake — books the wrong flight, sends an email to the wrong person, processes a transaction incorrectly — who's liable? This question isn't resolved, and it will define which agentic applications actually ship versus which stay in demos.
What This Means for Cognify Tech
Honestly? Eric Schmidt's advice aligns perfectly with where we were already heading.
The digital premium accounts business (ChatGPT Plus, ELSA, Netflix, etc.) is a market entry — it creates relationships and demonstrates that we understand AI tools. But the real leverage is in agentic workflows for small businesses that can't afford enterprise AI consulting.
The gap is specific: A Darwin-based business that needs to automate appointment scheduling, customer follow-ups, and inventory management doesn't need a chatbot. They need an AI employee that works 24/7 and costs $200/month.
That's the agentic opportunity. Not glamorous. Not VC-funded. Just profitable and real.
The Warning in Schmidt's Optimism
Schmidt's advice is technically correct and strategically incomplete.
It's correct because the agentic AI transition is real and the window for independent builders is open right now.
It's incomplete because "make money" and "build something that lasts" are different problems. The first-mover advantage in agentic AI is real but fragile — the big model providers are watching exactly what applications create the most value, and they'll build it into the base product the moment it's proven.
The real play: Build agentic workflows that are deeply specific to particular industries or geographies, where the training data and relationship network create a genuine moat that an API call can't replicate.
Or, more simply: be the person who actually understands the problem before you reach for the AI solution.
Schmidt was speaking at the Imagination In Action AI Summit in Davos. Worth following up if you want the full context — his talks are usually more substantive than the headlines suggest.
Sources: Instagram (Imagination In Action AI Summit), various AI news aggregation | Personal reflection — May 2026