When we shipped the Favorites page last week, the natural next question was: ok, I've saved an opportunity â now what? This update answers that. We're introducing a complete three-step AI build pipeline that takes you from "interesting market problem" to a phased launch roadmapâall within your workflow.
The AI build pipeline
Every saved opportunity now has a three-step AI workflow designed to progressively deepen your understanding and reduce build risk:
- Get Solution â a concrete product idea with a difficulty rating (Easy / Medium / Hard). This was already there, but it's now the first step in a pipeline. You get a high-level product direction, a suggested business model, and an honest assessment of how complex this would be to build solo.
- Get MVP â takes the solution and proposes a lean MVP: core features, target users, recommended tech stack, and a build effort estimate. This is where you move from "what to build" to "what's the smallest useful version?" The MVP spec includes feature prioritization (must-have, nice-to-have, post-launch) so you know where to cut scope under pressure.
- Get Plan â takes the MVP and generates a phased implementation roadmap. Phase 1, 2, 3 with timelines, deliverables, and success criteria for each. This is the roadmap you'd actually execute onânot vaporware, but a real sequence of work.
Each step feeds the next. You can stop at any stage â not every opportunity needs a full plan, and not every session has time for one. Someone validating a risky idea might just want the solution. Someone ready to start coding heads straight to the plan.
Community-powered AI caching
All three outputs are cached by problem. The first person to generate a solution, MVP, or plan for a given problem pays the AI cost. Everyone after recalls the same output instantly (marked ⥠recalled). Popular problems effectively get a community answer. This approach cuts average generation time by 60% for top-ranked opportunities while keeping costs predictableâyour Pro subscription covers unlimited generations within the cache.
Why this matters: If 50 founders are exploring the same market opportunity, only the first pays for AI. The other 49 get instant access and can spend time on evaluation instead of waiting for generation. It also means the most popular opportunities get continuously refinedâif you generate a plan for an opportunity and spot an issue, you can run it again and our system learns.
Download everything
Each panel now has a Download .md button. Solution, MVP, and plan each export as a clean markdown file named after the problem â ready to paste into Notion, Obsidian, Linear, or a README. You own your outputs. We don't lock ideas inside the platform.
A cleaner tier structure
We restructured the plans around what people are actually trying to accomplish:
| Tier | Price | What you get |
|---|---|---|
| Free Trial | 7 days | Full Pro access â taste the whole pipeline |
| Hobbyist | $19/mo | AI solutions, unlimited cards, favorites |
| Pro | $29/mo | Solutions + MVP proposals + phased plans |
| Premium | $39/mo | Pro + API access + team seats |
| Agency | $499/mo | Premium + white-label + priority support |
The trial now shows you the full Pro feature set so you know exactly what you're deciding between when it ends. Hobbyist is for people who want market intelligence without the build pipeline. Pro is for people who want to go from "interesting problem" to "I know what to ship."
A note on the build process
The MVP and plan generation uses a longer context window than the solution step â a phased plan for a real SaaS idea can easily run 4,000+ tokens. We found out the hard way (JSON parse errors from truncated responses) and bumped the limit accordingly.
The plan prompt asks Claude to return properly escaped markdown inside JSON. The solution is cached immediately on first generation so a truncation or network error on a retry doesn't cause a double-generation. We also added exponential backoff for failed generations to avoid hammering the API during outagesâretry after 2s, then 4s, then 8s.
What's next
Tier gating is live â MVP and plans are Pro+. The next thing we're watching is whether users stop at "Get Solution" or go all the way through the pipeline. That'll tell us whether the three-step workflow feels natural or whether we need to make the progression more obvious.
We're also experimenting with letting users save and refine outputsâright now they're view-only, but saving "my version" of a plan would let you mix what the AI suggested with your own domain knowledge.
As always â if something feels off, hit Support and let us know.