The writers' room has always been a place of creative tension. Ideas clash, stories evolve, and the best work emerges from collaboration. When we started Katha, we asked ourselves: how can AI tools participate in this process without disrupting what makes it work?
The Problem with Most AI Writing Tools
Most AI writing tools fall into one of two traps. They either try to write for you—generating entire scenes or scripts that feel generic and hollow—or they offer so little that they might as well not exist.
The first approach misunderstands what writers actually need. Writers don't want AI to write their stories. They want tools that help them write better stories, faster. There's a fundamental difference.
The best tools disappear into the workflow. They become extensions of how you already think, not interruptions.
Our Approach: Agents, Not Assistants
We build agents. Not chatbots, not assistants—agents. The distinction matters.
An agent has a specific job. It understands its domain. It can take action within defined boundaries. When you ask a research agent to find historical details about 1920s speakeasies, it doesn't just search and summarize—it returns structured, vetted information that fits your story's needs.
Here's what this looks like in practice:
Research Agent Input:
"I need details about speakeasy security measures in 1920s Chicago"
Research Agent Output:
- Entry protocols (passwords, recognition systems)
- Physical security (lookouts, escape routes)
- Police bribery networks
- Source citations for each detail
- Flag: conflicting accounts found for X
The agent doesn't write your scene. It gives you the raw material to write it better.
The Three Principles
After building dozens of these tools, we've settled on three principles that guide our work:
1. Augment, Don't Replace
Every agent we build has a clear boundary. Research agents don't write prose. Dialogue agents don't make plot decisions. The human writer remains the creative authority.
2. Fit the Workflow
Writers don't change how they work for tools. Tools need to fit how writers already work. This means integrating with existing software, matching the pace of creative thinking, and never requiring writers to "learn" the AI.
3. Get Better Over Time
The agents we build learn from each production. Not in a creepy "training on your data" way—but in how they're configured, tuned, and refined based on what actually helps that specific writers' room.
What We're Building Now
Right now, we're focused on three types of agents:
Character Voice Consistency — An agent that reads your existing character work and flags when new dialogue doesn't match established voice patterns.
Research Acceleration — Agents that can dive deep into specific domains (historical periods, technical fields, cultural contexts) and return production-ready information.
Structure Analysis — Tools that help writers see the shape of their story—pacing, act breaks, tension arcs—without imposing a formula.
If you're running a writers' room and curious about what this could look like for your production, we'd love to talk. We're taking on a small number of partnerships this year.
Share this article
Written by
Katha Team