Moda

The Moda blog: notes on building, debugging, and improving production AI agents, including agent harnesses, intent discovery, behavioral failures, and the continual learning loop.

Silent agent failures never file tickets -- they hide in production traces. Here's the unsupervised pipeline that clusters 150 million trace segments into named behaviors and failure modes.

The company builds behavioral cohorts from conversation data, including at-risk, power, and stuck-early users, and offers a CLI callable from Claude Code, Cursor, or Codex.

Octolane's AI CRM team had more customer sessions than anyone could read. With Moda they found where users got confused, which workflows broke, and shipped five fixes in their first few weeks.

Moda's pricing for AI agent observability and continual learning is customized per client and scales with production trace volume and pipeline usage.

The page compares Moda with Raindrop, LangSmith, Langfuse, Braintrust, Helicone, AgentOps, Arize, and Trajectory, and discusses differences among adjacent observability, eval, and post-training tools.

Braintrust offers a proprietary trace database, beta auto-clustering, and an AI assistant that analyzes production traces to surface failure patterns and generate scorers and datasets. The page contrasts these tools with Moda's harness-layer approach.

Helicone is a Rust-based AI gateway with request-level observability. Mintlify acquired it in March 2026; its standalone product entered maintenance mode, Experiments was deprecated in September 2025, and its OSS gateway continues to ship.

Moda vs Langfuse is a comparison of layers: Langfuse for traces, prompts, and evals; Moda for production agent monitoring above them. Keep Langfuse; add Moda to catch silent failures on every run.

Moda is a continual learning layer for AI agents that uses production traces to improve agent harnesses and surface emergent intents and behavioral failures.

Moda offers demo bookings and email contact with its team.

Moda uses production traces to generate improvements to agent harnesses, including prompts, tools, skills, evals, and memory. It is designed for teams running an agent in production.