AI Writer Pipeline
Prompt engineering → 60 specialized tools → 1.6M monthly generations. Built the meta-system that creates the tools.
About this system
AI Writer at Picsart serves 400K monthly users with 60 specialized writing tools. But building 60 tools manually doesn't scale. So I built a meta-system: a tool that creates tools.
The pipeline starts with identifying a use case (Instagram caption, product description, email subject). Then it generates the prompt template, input fields, output format, and examples. The result is a fully functional tool ready for production.
This approach means we can ship a new tool in hours, not days. And because it's systematic, the quality is consistent across all tools.
How to set it up
Define the tool framework
Standardize what every tool needs: prompt template, inputs, outputs, examples.
Build the tool generator
Create a meta-prompt that designs new tools from a use case description.
Set up the tool runtime
Build the execution layer that runs any tool with standard I/O.
Create feedback loops
Collect user ratings to identify underperforming tools for improvement.
Workflow
Identify use case
Find a writing task with clear input/output and high demand.
Generate tool spec
AI creates the full tool specification: prompt, inputs, outputs.
Test and refine
Run test generations, refine prompt for quality and consistency.
Deploy
Add to the tool library and monitor usage/feedback.
Variations
Social media tools
Captions, threads, bios for different platforms.
Marketing tools
Headlines, CTAs, product descriptions.
Professional tools
Emails, proposals, meeting notes.