sethserver.com Subscribe
MVP Meaning: Build a Learning Machine, Not a Tiny Version of Your Product

MVP Meaning: Build a Learning Machine, Not a Tiny Version of Your Product

By Seth Black • Updated: March 18, 2026

Startups · 5 min read

Like this kind of writing? Get one email a week with notes on startups, AI, and the occasional strong opinion about Python: subscribe to the newsletter.

You can tell a lot about a startup by how it talks about its MVP. If the first sentence is "It's basically a full platform," you're about to spend six months building a very expensive demo that nobody asked for.

An MVP is not a tiny version of your final product. It's a learning machine. The job is to test the scariest assumptions fast, with the least code and the fewest people possible.

I watched a company die because we couldn't do "least." A VC offered to recap us with $1MM if we could get spend under $250k/year. We had one employee, a shared office, and basically no bills. Fifteen minutes later, somehow "spend" was still "too high," and the company went from life support to dead. Fastest speedrun from hired to laid off. The lesson wasn't budgeting. It was ego. MVPs die from ego all the time.

The MVP is "just enough" to prove or disprove reality

A good MVP answers one question:

Will a real user do the thing that makes this business work?

Not "Will they like the UI?" Not "Will TechCrunch clap?" It's the money question, the retention question, the behavior question.

Dropbox's famous MVP wasn't even a product. It was a video. That's the whole point. They tested demand before building the whole sync engine.

If your MVP needs microservices, you're not building an MVP. You're building a shrine.

Feature triage: cut it in half, then do it again

I do this:

  1. Write every feature you want.
  2. Cut the list in half.
  3. Cut it in half again.
  4. Keep the one feature that solves the core pain.

Amazon started with books. Not because books are magical. Because starting with "everything" is how you end up shipping nothing.

Every extra feature adds:
- more code paths
- more bugs
- more UI decisions
- more edge cases
- more support load

And the worst part: it gives you more places to hide from the hard truth that users don't care yet.

Feature creep is like that scene in an '80s movie where someone keeps adding gadgets to a plan that already worked. Suddenly you've got a grappling hook, night vision goggles, and a trained falcon. All you needed was a key.

"Smoke and mirrors" is fine. Lying is not.

Some of the best MVPs are basically organized cheating.

Zappos didn't start with warehouses and a supply chain. They posted shoe listings, then bought shoes locally and shipped them. That's not fraud. That's validation.

For software, the classic move is the Wizard of Oz MVP: the user thinks it's automated, but it's a human behind the curtain. Works great for AI features.

I've done this with ML systems: start with a manual labeling pipeline, then automate the parts that hurt the most. It also forces you to learn what "accuracy" even means to users, which is usually not the same thing as your offline metric.

Just don't do what my old sports-betting-adjacent company did. We had traction until the CEO decided he wanted to game the system. I was told to nerf the algorithm so he and his buddy could win. Usage fell to zero almost immediately. If one person can rig your MVP, your MVP is measuring corruption, not product-market fit.

Feedback: watch behavior, not opinions

Users will tell you they want "more customization" and "AI insights." Then they won't click the button you already built.

Collect feedback in multiple ways:
- short in-app questions (one question, not a survey novel)
- a plain email reply channel
- 15-minute user calls where you watch them use the product
- basic event tracking (what they click, where they drop)

Split these up:
- usability problems ("I can't find it")
- value problems ("I found it and don't care")
- trust problems ("I don't believe it works")

If you don't separate those, you'll ship Digg v4 energy: a big "improvement" that users hate, and then you're doing emergency rollbacks while pretending it's fine.

Pivoting is not quitting. It's steering.

YouTube started as video dating. Slack started as an internal tool. Instagram started as a check-in app and then stripped everything except photos.

A pivot is: keep the part that works, kill the rest. It's feature triage at the company level.

Also: pivoting from physical product to SaaS is usually not a pivot. It's a different company. I watched one team try it in 2019 after their IoT device flopped in retail. Hardware took them ~18 months per iteration and burned hundreds of thousands before they learned anything. Their "quick SaaS pivot" shipped in 10 weeks-and users roasted it in public within 48 hours. Different timelines, different cash needs, different expectations. Hardware is slow and expensive. SaaS is fast and brutally honest.

Scale later. Prepare for it anyway.

If the MVP works, you'll need to harden things:
- refactor the "prototype spaghetti"
- add security basics early (auth, rate limiting, audit logs)
- move off the laptop-hosted database before it becomes folklore

Scaling too fast can wreck you (hi, Groupon). Scaling too slow can also wreck you. The move is to scale the winning loop, not the whole org chart.

What happens after the MVP ships is where most teams discover their technical debt. You'll need to triage security issues, add tests, and fix error handling before the cracks become chasms.

MVP is ready when:

Everything else is decoration. Decoration burns runway. That's it.

-Sethers

Share this post
Newsletter

One email, once a week.

Notes on databases, systems, and the occasional strong opinion about Python. No spam, unsubscribe anytime.

Seth Black
Written by

Seth Black

Engineer and founder based in Texas. Writes about databases, AI, and running things in production. Embeds in small teams as lead engineer.

More from Startups

View all →
Startups

What Actually Goes in a Consulting Contract

Sep 17, 2026
Startups

Productizing a Service: From Hours to Licenses

Jul 26, 2026
Startups

The MVP That Is Not a Product

Jul 11, 2026