Go from idea to product with AI vibe coding
What this webinar covers
Andreas Melvær and Line Hjartarson — both multi-award-winning AI-assisted product designers at SmplCo — joined Barclays Eagle Labs to walk founders through how to actually take an idea, vibe-code it into a working product, and avoid the traps that turn a fast prototype into a pile of regrets six months in.
Key takeaways
1. Get the vocabulary right: prototype vs proof of concept vs MVP
These three words get used interchangeably. They are not the same thing.
- A prototype is a clickable user interface for testing a concept with stakeholders. Usually no real code under the hood.
- A proof of concept has real code, lives on the internet, and tests whether the idea works technically — sometimes at scale. Not for public use.
- An MVP — minimum viable product — is the smallest version of your product someone would actually pay for.
The "actually pay" part is what most teams skip. If nobody is paying, it's a prototype with extra steps.
2. The right number of users to test with is "diverse, not many"
Founders ask Andreas this constantly: how many users should I test with? His answer: it depends, but less is almost always more. Three to ten of the right people, with diverse perspectives, will give you better signal than fifty people drawn from your existing echo chamber.
The pitfall to avoid: only testing with people who already love you.
3. The vibe-coding workflow that actually ships
Line walked through the stack she and Andreas use for fast, shippable AI-assisted builds:
- Figma for design and the source of truth — works as a canvas you can collaborate inside, plus its built-in AI-coding capability for prototyping
- Claude (Code) for planning and building — Claude's planning mode lets you reason through architecture in steps before generating code
- GitHub for version control once you have something worth keeping
- Gemini Nano Banana for image and illustration generation
- Resend / Stripe / standard SaaS for the boring-but-essential parts: emails, payments, auth
The pattern: prototype in design tools first, plan in Claude with full context, then code with version control. Not the other way round.
4. AI tools need context, or they hallucinate plans
A lot of founders open Claude or Cursor and just start typing. Line was emphatic: AI tools need context to plan well. That means feeding the model documents, prompts, and constraints before asking it to design the system. The plan you get out is only as good as the input you put in.
This is why she often uses one AI session to generate the planning artefacts, then feeds those into another AI session to do the actual build. Layered prompting, not single-shot prompting.
5. Don't ship a prototype that looks like duct tape
When you're vibe-coding, the front-end can look polished from day one. There is no excuse in 2026 for handing a stakeholder a prototype with placeholder JPEGs, lorem ipsum, and three different button styles. That used to be acceptable. It isn't any more.
Investors, customers, and partners now expect the surface to look real even when the back-end is held together with binder clips. The good news: AI tooling makes the polished surface fast and cheap.
Useful links
- The full AI Integration Playbook — the framework Andreas and Mike use with founders
- Build Your Own Internal Tools guide — for the "AI for your business" angle
- 25%-off discovery offer for Eagle Labs members
Want a second pair of eyes on your prototype?
If you're partway through a vibe-coded build and want a sanity check before pouring more time in, book a call — happy to look at it without a pitch.