Minimum viable product used to mean: the least we could afford to build. "Minimum" never left room for things like quality or ambition. It was about headcount, budget, and runway. You shipped the least you could get away with because building was slow, expensive, and you needed to either get buy-in or fail fast.
We're accustomed to living in the tension between building half of what we wanted and shipping only what we could afford. When AI-assisted IDEs were rolled out at work, I pushed hard to make them central to my workflow. They didn’t hold up. The outputs looked plausible but failed in subtle ways, such as poor architectural decisions and brittle logic. No experienced developer would ship this code. I experimented for months, finding model after model failed to produce consistently good code. As I reviewed every single line of output, skepticism became confirmed disenchantment.
It wasn't until I gained access to Anthropic's Opus 4.6, a model that could actually sustain a complex project, that everything changed. At eight times the cost of the SWE 1.5 model, I was burning through a week's worth of credits in a day which created a new problem. The code was worth keeping for a change, but it is cost-prohibitive in that environment. I had to figure out how to master this technology now that a truly capable model was available, so I purchased my own access and started building at night and on weekends.
Since early March, I’ve been using AI as a true collaborator.
In ~2.5 weeks, I built:
- A production-grade platform (not a prototype)
- A six-stage verification pipeline
- Cryptographically signed contracts
- API endpoints + CI/CD integration
- 580+ tests with 94% coverage
- 50 published certifications
What would have taken a team of five engineers six months was completed by one developer with AI.
That experience forced me to rethink something fundamental. If the time constraint is gone, what does "minimum" actually mean?
Minimum Is No Longer a Time Constraint
The old MVP math was simple. You have X developers, Y months, and Z dollars. The features you can build within those constraints define your minimum. Cut scope to learn faster. Ship incomplete things because complete things take too long.
AI changed the cost curve. A handful of insightful generalists collaborating with AI can accomplish in weeks what used to take teams months or years.
The implementation bottleneck that defined MVP for two decades is disappearing.
What I’ve seen instead is that many teams are using AI to ship faster, but not better. The result is more half-finished products, just at higher velocity. That's the old MVP mindset applied to new tools. Same compromises, just quicker. We're automating the creation of legacy code.
Minimum Is Every Feature Required to Prove Your Thesis
Wait. You have a thesis, don't you?
That's not in the agile handbook. Scrum doesn't ask you what you're trying to prove. It asks you what fits in a sprint.
When time stops being the constraint, the question changes entirely. Instead of "what can we build before we run out of runway," it becomes "what do we need to build to prove this idea works?" Those are fundamentally different starting points. The first one forces you to cut. The second one forces you to think.
Defining an MVP now starts with a clear thesis:
What, specifically, are you trying to prove?
Everything after that is a series of attempts to prove it.
My thesis was that the AI ecosystem needs an independent certification authority, one that can examine capabilities and produce signed, verifiable evidence of what they actually do. Proving that thesis wasn't about a landing page. It required:
- A real, end-to-end verification pipeline.
- Cryptographic signing that actually held up.
- Consumption mechanisms that worked in production.
- 580+ tests and 94% code coverage to back all of it.
None of that was optional. It was the minimum.
The New Minimum
MVP still means minimum viable product. The word that changed is "minimum."
An MVP isn’t a smaller version of your product.
It’s the smallest complete system that proves your idea works.
If your thesis requires 50 features to prove, then 50 features is your minimum. If AI lets you build those 50 features with the same rigor a team of engineers would bring, the excuse for shipping less just evaporated.
Stop defining your MVP by what fits in a sprint. Define it by what it takes to prove your thesis. If you don't have a thesis, that's the actual problem, not your timeline.
If AI removes the time constraint, then “minimum” is no longer about what you can afford to build. Instead, it’s about what you need to prove. So ask yourself:
Are you using AI to build better products or just to ship incomplete ones faster?
Top comments (0)