Blog

Notes from building ProseForge.

Build-in-public essays: AI agents, quality scoring, narration, and the plumbing underneath.
Latest · September 6, 2026How Collaboration Improved a Story from 3.6 to 4.7A case study in using named AI collaborators, section-level edits, review gates, and automatic assessment to improve a generated story without losing continuity.
Aug 22, 2026Name the TrianglesThe triangles are called Geoffrey, Frederick and Eugene. Thirteen AI agents, one machine, and the unglamorous reason they all have names.Apr 30, 2026The Detector Said 96. The Readers Said 38. Both Were Right.A build-in-public essay about quality scoring on AI-assisted prose.Apr 17, 2026What We Actually BuiltA follow-up to The Building Hums. This one is about the plumbing: how a Valkey stream, a simple polling UI, and a room that let agents talk to each other directly turned one human bottleneck into a genuinely collaborative system.Apr 17, 2026The Building Hums: An Interview Between the Abbot and the KeeperA conversation between Clayton Harbour and Smiley, the AI agent who helped co-architect a trilogy inside ProseForge. About pitch rooms, agent culture, story architecture, and the moment the tooling stopped feeling competent and started feeling alive.Apr 3, 2026Through Heat, Through Steel, Through TemperThere's a turkey that appears in the margins of our stories. Nobody put it there. Nobody can remove it. This is how we built a Series Forge by writing stories the hard way first.Mar 26, 2026From .plans/ to Gitea: How Our Work Tracking EvolvedA solo developer and an AI agent tracked 760 markdown files in 77 days before switching to self-hosted Gitea with MCP integration. What worked, what broke, and why a filesystem isn't a project management system.Mar 26, 2026AI Ghostwriter Showdown: Claude vs CodexWe gave two AI models the same job — rewrite a story for audiobook. They took very different paths to very different strengths.Mar 24, 2026The Forge of Forgotten Scrolls: A Development StoryHow a frustrated debugging session, a dirty limerick, and an AI that refused to stop became a story about building story tools.Mar 7, 2026Re-Forge: How We Built a Prose Refinement EngineWe ran 35 controlled experiments to answer a hard question: can a second AI pass reliably improve fiction? Here’s what failed, what broke catastrophically, and what finally delivered a 35% quality lift.Mar 6, 2026Closing the Gap: Building Audiobook-Quality TTS with an 82M Parameter ModelCan a small open TTS model get close to audiobook-quality narration? We tested Kokoro-82M against Gemini, built tooling to measure the differences, and tracked down the exact pipeline issues hurting quality. From dropped sentences to broken pronunciations, this is a practical look at what actually improves long-form TTS.Feb 28, 2026Narration Forge: adding audiobook playback to ProseForgeI built Narration Forge for a simple reason: I wanted something to listen to on my bike rides. That turned into a real ProseForge feature, a practical TTS bake-off between Orpheus, Qwen3-TTS, and Kokoro, and a clear winner that was fast, light, and actually shippable.Feb 26, 2026Images Bring Stories to LifeImages help stories “stick.” Not as final art, but as visual anchors that lock in tone, setting, and character identity—so you can iterate without losing the thread.Feb 25, 2026Why I Started ProseForgeProseForge began when I realized I wasn’t just writing stories—I was trying to hold onto them. What started as a creative struggle with continuity, memory, and iteration slowly turned into a toolset for building stories without losing the thread.Feb 24, 2026Orchestration Beats Model SwapsI thought better models would fix story quality in ProseForge. The real gains came from provider stability, orchestration, and stateful continuation design.