The new product development cycle: morning idea, evening feedback

How Gamma and Anthropic validate product ideas in hours, not weeks — and why your make vs. buy decisions just changed forever.

Lenny Rachitsky recently shared this innovation flow from his conversation with Grant Lee, CEO of Gamma:

Team member has an idea in the morning
They build a prototype that day (with AI tools)
That afternoon, they test with ~20 real people using a moderated UX testing tool
Get rich user feedback by end of day
Tweak/kill/build based on results

👉 From idea to validated insight in a single day.

This isn't an isolated example. Anthropic's team follows a similar pattern: idea → prototype → internal launch → watch behavior → data-driven prioritization. They skip the spec document entirely and let real usage guide decisions.

The common pattern is that generative AI enables two critical capabilities:

Faster data-to-insights: Analyze feedback at scale, immediately
Rapid prototyping: Build functional prototypes without a development team

But here's another shift happening right now: General AI is eating the rest

ChatGPT accounts for 69% of AI tool traffic. Claude, Gemini, Copilot split the remaining 31%. The specialized tools (Jasper, Copy.ai, etc.) are being absorbed into the general platforms.

The strategic question: Why buy a specialized AI tool when Claude or ChatGPT can do it seamlessly. Think XP onsite customer on steroids.

For product teams, this means:

Stop buying tools - buy outcomes: faster cycle time, better feedback analysis, quicker prototyping
Start with a general AI and add specialized tools only where ROI is proven
(Re-)consider make vs. buy decisions - make has become much easier with AI

From our practice: employees building their own tools 🔨

We're seeing this play out with one of our clients:

The setup: An employee is developing a user recruiting flow — with coaching support but building it themselves. Tools: Cursor and Streamlit. Time invested: 2 weeks, part-time (3 hours/week). Result: A working App that saves approximately 10 hours per week.

The shift: Instead of configuring out-of-the-box tools, they're creating custom workflows with code. And they're not a developer.

This is "vibe coding" for internal tools — and it follows the same product principles we use for customer-facing products:

Discover: What's the actual job to be done? Where's the time savings or quality improvement?
Design & Test: Does it work in real workflow conditions?
Iterate: Refine based on actual usage

The connection to AI-accelerated product innovation cycle: Optimized internal processes directly improve cycle time from idea to validation. These two themes belong together — faster innovation cycles require better internal tooling and vice versa.

The open question: Where does the code run? More broadly: What's the product operations model when many employees create their own tools? And what happens to security concerns and the architecture of the internal tool stack?

How might we create the right balance between exploring and exploiting new ways of using AI in our workflows

The question isn't whether to let people build — it's how to provide the infrastructure that makes a thousand individual projects feel like one coherent system.

Practical stuff: Claude Skills vs. MCPs (and why it matters) ⚡

We've been running experiments with Claude's Model Context Protocol (MCP) and the newer Skills feature. The results are instructive.

MCPs are powerful but problematic:

They flood context windows (think CO2 footprint here) with unfiltered data (expensive)
Technical instability — still very early stage, sometime Notion page update works, sometime it doesn't
Architecture problem: It's not clear what actually gets included into the context and context selection isn't accurate enough

Skills as an alternative:

Very precise — designed for specific tasks
With Claude Code, you can build custom skills (no coding required for many use cases)
Solves the "where does it run?" question → it runs in Claude
Potential to replace multiple Claude Projects with static contexts by single, focused skills that can be called within any Chat context

Example use case: Preparing for workshops used to require shuffling context between multiple projects (emails, Notion pages, transcripts, calendar). With Skills, the context flows naturally to the task at hand.

Our expectation: A massive marketplace of skills will emerge. This fits the consolidation pattern — one platform (Claude), extended with specific capabilities only where needed.

Workshop hacks with AI: three use cases 🎙️

For anyone running meetings or customer workshops, we've discovered some practical patterns using Whisper Memos, Claude, and Miro.

Use Case 1: Team Contract Canvas

Old process: Everyone writes post-its. New process:

After 1-2-4-all participants speak 🎤 their contributions instead of writing post-its or typing them into Miro
Whisper Memo transcribes in the background
Simple prompt generates post-its via Google Sheets → Miro

Why it works:

Less staring at screens
More eye contact and natural conversation
People open up more when speaking vs. typing
Result still digitally available

Use Case 2: Lean Coffee closure format

The flow:

Collect "ahaa moments" on board by speaking 🎤 them out (not writing post-its)
Whisper Memo creates transcript
Prompt generates table with three categories: Actions, Insights, Questions
Claude creates a Lean Coffee moderator from the table
Optionally include dot voting results and participant expectations for prioritization

Result:

Moderator presents themes step by step
Visual backlog of the full discussion
If Whisper Memo ran throughout: automatic summary with conversation content

The feeling: Much more lightweight and natural (after some tricky technical setup).

Use Case 3: Culture check from meeting recordings

We're experimenting with using Gemini for analyzing UX test recordings. Why Gemini over Claude?

Gemini handles up to 40 minutes of UX testing footage (sufficient for most sessions, and extending as context windows grow)
Better precision than the built-in analysis options from moderated interview platforms we evaluated
Open data interfaces and agent-based synthesis mean insights connect to your broader product context — exactly the pattern from our opening theme. Real interviews still matter, but AI extracts more value from them.

Running regular UX tests and spending a lot of your time analysing the video recordings? We are happy to test our solution with your data.

Book a discovery call here

The next AI Prototyping Dojo — cohort starts in November

We still have spots available, and we have a discount code for newsletter subscribers:

Why this is valuable

Two participants built a branded prototyping tool — fully customized to their style. Not a simple e-commerce site or landing page, but a complex, interactive web-based tool. Now they can prototype ideas interactively and use them directly in stakeholder calls and team sessions.

Time savings: A few prompts instead of 1-4 days of developer time. Functional UX and design prototypes without writing code.

"Great workshop! I liked that it was very hands-on. There is some knowledge necessary about all the tools I feel like. But for me personally it was great"

— Bastiaan Korte, Growth Marketer & Digital Designer

New expanded focus:

We're planning to integrate our experience with internal tool prototyping into the agenda. Tools covered besides Lovable will be Claude Code, Cursor, Streamlit, Replit.

Relevant for: Anyone with internal tool needs, process optimization challenges, or questions about workflow automation.

Format:

Open backlog — no question goes unanswered
Latest approaches to using AI for software development and prototyping
Hands-on: you'll build real prototypes during the sessions

Enjoy 10% Off with code AIPIONEER10

Join the next cohort in November

Still unsure if this fits your specific situation? Book a 20-min discovery call → We'll look at whether the Dojo answers your questions.

P.S. 30-Day Risk-Free Learning Promise 🛡️ Not satisfied? 100% refund after Session 1, 60% after Session 2. No risk, only learning.

Was this helpful for you? Please let us know and any updates or wishes are always welcome from your side.

Happy prototyping