Working in ClaudeBuild a product end-to-end

Build a product end-to-end

From a sentence to a live, branded, publishable AI product — what to say, what Claude does, what you get.

The connector makes building a Productised product a one-conversation activity. You describe what you want; Claude plans, builds, generates, themes, and previews. Here's exactly how the flow works under the hood — and how to brief Claude for the best output.


The four-call build flow

Every product Claude builds goes through the same canonical sequence. You don't have to know about it — Claude orchestrates the calls. But knowing the shape helps you brief Claude better and intervene at the right point.

plan_product — Claude analyses your brief

Claude calls plan_product with your description. The tool returns a recommendation: outcome type (standard, advisor, design, collect_only), funnel stage (tofu, mofu, bofu), the right product flow, and a rationale.

You'll see this as a visual Blueprint card in chat — outcome, funnel, flow, and reasoning. If anything looks off, you correct it in the same message.

create_product — empty shell created

Claude calls create_product to reserve a product_id in your workspace. The product exists but has no conversation flow, no system prompt, no questions, no canvas yet. This is the step where you appear in the dashboard.

generate_product — the AI build

Claude calls generate_product with the enriched brief. Behind the scenes Productised's generation pipeline runs: collection goals (your intake questions), system prompts, decision logic, the canvas workflow, and a default theme — all tuned to the outcome type and funnel stage you chose.

This is the heaviest step (~30 seconds). You'll see a build-progress card with each phase completing.

apply_result_template — design output

For design outcomes, Claude calls apply_result_template with one of five branded presets (Score, Dimensions, Radar, Persona, Deep Insights). This sets the page HTML and configures the AI agent to output the exact field names the page expects.

For standard outcomes the agent streams a personalised document. For advisor, the AI continues chatting after collection. No extra step needed.

The result: a complete product with a live demo URL, in one Claude turn.


What to put in your brief

The richer the brief, the better the output. Good descriptions usually contain four things:

A good brief

"AI Readiness Scorecard for B2B marketing leaders at growth-stage SaaS companies. Six questions across strategy, data, skills, content, tools, and culture. Score each pillar 0–10 with a personalised insight per pillar, then a top strength and critical gap. Deliver a branded results page with the score gauge, dimension breakdown, and a CTA to book a 30-min strategy call."

This gets you a complete product with no follow-ups needed.

A weaker brief

"Build me an AI readiness quiz"

Claude will ask follow-up questions to fill in the missing pieces — audience, scoring shape, CTA — which is fine, but you'll spend a few back-and-forths getting there.


Outcome types — picking the right one

Every product has one of four outcome types. Claude infers it from your brief, but if you want a specific shape, name it.


After the build

Every successful build returns:

  • A product_id Claude can use in follow-up calls
  • A demo URL (?demo=true) — bypasses gates and visibility so you can preview immediately
  • A live URL — your public, shareable product link
  • A summary of what was created — question count, node types, outcome type

From here, you keep iterating in the same conversation. Common follow-ups:

  • "Open the conversation flow and change question 3 to a 1–10 scale instead of multiple choice"
  • "Make the system prompt more direct and outcome-focused"
  • "Apply the dark serif theme and set the CTA to my Calendly"
  • "Switch the result template from Dimensions to Radar"

Each one is a real, persistent change. The demo URL refreshes immediately.

Designing result pages in Claude → Branding and themes →