• It’s not just about output. The value lies in how well the result meets your industry’s unique criteria.
• Your data is your moat. Train on it.
• Models will improve. What wasn’t good/cheap enough yesterday could be transformative by the time you get your product to market.
When applied to a dev tool like Storybook:
• We can auto-generate stories by pasting component code today.
• Our open-source data format lets AI tools like Copilot, ChatGPT, and Claude train on millions of example stories.
But if we wanted to productize a solution around story gen, we'd need to ensure the stories adhered to our user's unique project requirements, coding preferences, AND validate against errors or duplication.
In our case the generation is the most straightforward part of the toolchain.
The community put that together a year ago (h/t Kaelig at Netlify). To make a solution that's fit for purpose, we'd need to build the guardrails and fit it into customer's workflows.
Most companies weren’t built with AI in mind, yet now they’re navigating this transformation. These are some of the notes I’ve picked up along the way.
Comments
• Your data is your moat. Train on it.
• Models will improve. What wasn’t good/cheap enough yesterday could be transformative by the time you get your product to market.
Build vs Buy.
Definitely BUY the generation piece.
Use your unique perspective to BUILD the quality validation and refinement instead.
• We can auto-generate stories by pasting component code today.
• Our open-source data format lets AI tools like Copilot, ChatGPT, and Claude train on millions of example stories.
The community put that together a year ago (h/t Kaelig at Netlify). To make a solution that's fit for purpose, we'd need to build the guardrails and fit it into customer's workflows.