Ideas & Opportunity

As a non-technical founder, how do I evaluate a tech-heavy trend like AI without getting fooled by hype?

A starting point

You don't need to write code to judge a trend, but you do need to separate what the technology can reliably do today from what a demo promises. Talk to two or three builders you trust and ask them what breaks in production, not what's possible in theory. As a starting point, focus on whether the trend removes a real cost or unlocks a real behaviour for a customer you understand, and let engineers judge the how.

Go deeper

Hand-picked from around the web, each with a note on why it earns your time.

3 resources 2 link-checked Listen Read

Listen

🎧 Podcast

✓ Link checked Free Intermediate

Why we picked it A demo that works once on stage and a system that holds up for real users are two very different things, and this conversation is about the gap between them. Tony Holdstock-Brown (who builds the plumbing that runs AI in production) walks through where multi-step AI actually delivers and where it quietly stalls: retries, edge cases, cost, and the un-glamorous reliability work. For a non-technical founder it is a useful ear on what "it works" really costs behind the demo.

Building Production Workflows for AI Applications

On AI + a16z by Tony Holdstock-Brown, Yoko Li and Derrick Harris ~45 min

The hard part of an AI product is usually not the model call, it is making the workflow around it reliable when real inputs get messy.
Ask any AI trend the operator's question: what happens on the tenth thousandth request, not the first perfect one.
Cost and reliability compound as you scale, so a capability that looks free in a demo can quietly become the whole business problem.

Open a16z.com →

Read

✍️ Essay

✓ Link checked Free Intermediate

Why we picked it This is the essay to read when you want to know whether a trend can carry a real company or is just a feature riding on someone else's model. Casado and Bornstein show why AI economics often look more like a services business than clean SaaS: lower margins, real infrastructure and human-in-the-loop costs, and moats that are shallower than the hype suggests. It is from 2020, but the core question (where does durable value actually accrue) is exactly the one to ask about any new AI wave.

The New Business of AI (and How It's Different From Traditional Software)

From Andreessen Horowitz (a16z) by Martin Casado and Matt Bornstein ~20 min read

AI companies often run at 50 to 60 percent gross margins, not the 60 to 80 percent of classic SaaS, because compute and human review are real recurring costs.
Model access is commoditizing, so defensibility comes from owned data, a narrow workflow, and real switching cost, not from the model itself.
Before betting on a trend, ask whether it supports a standalone business or is a feature a bigger platform will absorb.

Open a16z.com →

📄 Article

Free Beginner

Why we picked it This is written for the person deciding whether an AI idea is real, not for the engineer building it, so it stays in plain language a non-technical founder can act on. It names the exact red flags to distrust (jargon with no specifics, a founder who cannot point to a paying customer, a product that is just a thin layer on someone else's model) and the signals that actually hold up (owned data, a narrow real problem, real domain expertise). Treat it as a checklist for pressure-testing a trend, not a verdict on any one company.

Hype vs. Value: Assessing Early-Stage AI Startups for Real Potential

From Unite.AI by Igor Ryabenkiy, Nikolay Kirpichnikov, Sergei Bogdanov and Alexander Korchevsky ~8 min read

"We added AI" is not a business. Ask what data the startup owns and what specific, expensive problem it removes for a real buyer.
The signal that a trend is real is willingness to pay and fast iteration on real user feedback, not demo polish or a slick pitch.
Domain depth is the durable edge. A team that deeply understands one industry can defend a position that a generic wrapper cannot.

Open unite.ai →