Recently we gave an agent team a brief: build a SaaS tool that scans e-commerce stores for Google's Universal Commerce Protocol readiness, generates a compliance score, and previews the manifest they'd need. Target market: the 20 million non-Shopify stores that have zero UCP tooling. Make it look like a real product.
A few hours later we had a working app. Live on Vercel. Scanner crawling real stores, AI analysis grading readiness on a 0–100 scale, actionable recommendations, and a full landing page with pricing tiers. Not a prototype — a product you could show to customers.
What the agents actually built
The finished product is a Next.js app with a clean separation of concerns: a URL scanner that crawls product pages with Cheerio, a scoring engine that evaluates UCP readiness across multiple dimensions, an AI analyzer that assesses content quality, and a manifest generator that produces compliant .well-known/ucp.json output. Four modules, each doing one thing well.
On the surface: a polished landing page with gradient backgrounds, feature cards, a "how it works" section, and three pricing tiers. The kind of thing that takes a freelance designer and developer a week to produce. The agents did the full stack — backend logic, frontend design, copy, deployment config.
Why this matters for consulting
We don't sell MVP-building as a service. We build multi-agent systems — orchestration, quality loops, fleet management. But this project illustrates something important about what agent teams are capable of today: the gap between "idea" and "working product" has collapsed.
The traditional path from concept to MVP involves weeks of work: wireframes, design reviews, backend architecture, frontend implementation, deployment setup, copy editing. An agent team compresses that into hours. Not because the agents cut corners — the code is clean, the error handling is solid, the UI is polished — but because they work in parallel, don't context-switch, and don't spend three days debating button colors.
What agents are good at (and not)
Building an MVP from a clear brief is firmly in the "good at" category. The brief had a specific market gap (non-Shopify stores need UCP tooling), a clear deliverable (scanner + manifest generator), and known constraints (Next.js, Vercel, AI API). The agents didn't need to make strategic decisions — they needed to execute well on a defined scope.
What agents are not good at: deciding whether to build this product. Validating the market. Talking to potential customers. Figuring out pricing. That's still human work, and it's the work that matters most. The agent team can ship the product in hours, but only after a human has done the thinking about what to ship.
The real leverage
The interesting takeaway is not that agents can write code. Everyone knows that. It's that the cost of testing an idea has dropped to near zero. Previously, validating a SaaS concept meant committing weeks of development time before you could put anything in front of users. Now you can have a working product by end of day and start collecting real feedback immediately.
For our consulting clients, this changes the math on experimentation. Instead of debating for a month whether to build an internal tool, just build it. If it works, invest in hardening it. If it doesn't, you lost an afternoon, not a quarter.
That's the world we're building toward: one where the bottleneck is having good ideas, not executing on them.