Firstory

Firstory: Your real AI-native story writing companion to drive you effortlessly through writing your first story—and stay ahead in every one after.

[Firstory @firstoryapp]

Story Theory Benchmark launched

34 tasks | 21 models | 5 narrative theories
Tests whether LLMs understand story structure — using frameworks screenwriters have used for decades.
Open source. Reproducible.

Full results below

https://t.co/N9I5TMMDAe

[Firstory @firstoryapp]

The conclusion:

The best AI writing tool doesn't write.

It remembers.

It remembers what the writer wants. What the character would say. How the world works.

It handles the predictable — so the writer handles the profound.

That's the next generation of agent-native writing.

[Firstory @firstoryapp]

The Future of AI Writing

The future of AI writing tools isn't about making writers faster.

It's about something more fundamental.

[Firstory @firstoryapp]

But they're all solving the same problem:

Speed. How to make words come faster.

The question nobody's asking:

How do we help writers think less about mechanical parts — so they can think more about what matters?

[Firstory @firstoryapp]

8/ The future of story agents lies in:

• Data → Action (Ontology concept)
• Shared Context (Orchestration)
• Character as Product

These enterprise concepts point to a new paradigm!

#StoryAgent #CharacterAsProduct #Ontology #WritingTools #Firstory #AIWriting #AIAgents

[Firstory @firstoryapp]

🧵 What if enterprise AI's best-kept secrets could transform how we build story agents?

Two powerful concepts from enterprise AI might just be the breakthrough AI writing tools need.

[Firstory @firstoryapp]

7/ Three-Layer Story Agent Architecture

① Data Layer: Character profiles, world lore, plot history
② Orchestration Layer: Connect different writing agents, share narrative context
③ Action Layer: Character Engine — generate consistent actions/reactions based on settings

[Firstory @firstoryapp]

@sotoalt_ @grove_online Cool! Would love to beta test!

[Firstory @firstoryapp]

Built a Ralph Loop with open specifications!

1. You approve the open-end rubric first
2. All evaluator LLM models evaluate each iteration
3. When over a threshold, approve

Weekend experiment: https://t.co/aAvnG763gQ

[Firstory @firstoryapp]

Story Theory Benchmark is open source

→ Run it on your model
→ Add new tasks
→ Check the full leaderboard

If you're building AI writing tools, the methodology might be useful.

Follow for more on LLMs and creative writing!

#LLM #AIWriting #Benchmark

[Firstory @firstoryapp]

Story Theory Benchmark launched

34 tasks | 21 models | 5 narrative theories
Tests whether LLMs understand story structure — using frameworks screenwriters have used for decades.
Open source. Reproducible.

Full results below

https://t.co/N9I5TMMDAe

[Firstory @firstoryapp]

Example: Multi-Beat Synthesis

Write 3 connected beats where:
- Each has specific requirements
- Beat 3 can't contradict Beat 1
- Character voice stays consistent

Average: 79.8%
Best (o3): 92.9%
Worst: 43.7%

47-point spread = high discrimination between models.