This looks great! Sharing with the Harvard in Tech London community.

P.S. Feedback on our new site welcome. I'm still iterating!

As a product-focused founder formerly limited to product management and never being able to effectively help developers create production-quality code, being able to pair with Claude is transformative.

Full post 👇
blog.nseldeib.com/p/pairing-wi...

From Prompts to (Almost) Production: Choosing the Right AI Tools for a Startup Website Refresh: blog.nseldeib.com/p/from-promp...

Immediately called out the cheek cuteness when my partner started using your app. The details are (pun intended) delightfully cheeky 🍑

Tony Stark isn’t the same Iron Man without JARVIS!

What channels have worked best for your user research? Always looking to improve our approach.

#UserResearch #StartupLife #ProductDevelopment

This is the exact process we're using as we refine how we describe our software simulation tools at @codeyam.bsky.social . Hope it saves you some headaches!

I've shared our complete playbook including:

- How to define who you actually need to talk to
- Where to find these people without expensive tools
- Our actual screening survey and email templates
- Tips for catching people who try to game your screening (it happens more than you'd think!)

Update: this meetup was a blast!

Pictured: @nathenharvey.bsky.social facilitating an epic live coffee discussion on AI, DORA for the cloud, velocity vs quality, measurability of AI’s impact on software development and more!

This is why we're building CodeYam - to deconstruct software projects into individual functions, demonstrate their behavior through simulations, and make it easier to validate, discuss, and modify complex code with AI assistance.

This is why tooling is crucial. The right tools dramatically simplify collaboration by letting you isolate specific components of complex systems for focused discussion.

There are two primary approaches to this challenge: iteration and tools.

Iteration provides tremendous value by breaking tasks into smaller components and validating results at each stage.
But this approach comes at a cost - you must invest time validating AI's output, which undermines efficiency.

When you describe a task to AI, it makes assumptions based on its training data. Ask it to write JS tests? It defaults to Jest, even when explicitly told not to.

This pattern repeats across all communication. It's impossible to provide sufficiently detailed specifications for complex ideas.