Roadmap

Goals

Q4 2026 objectives

Objective 1: Finish the UI redesign Carlos MarchalCarlos Marchal

Motivation: Every tab in the core tracing product should help users find issues with their AI agents and features. The redesign will make these tabs more useful and provide a consistent experience across the product.

What we'll ship:

• UI redesign – redesign all tabs in the core tracing product around finding issues with AI agents and features, with consistent navigation and interactions
• Matching MCP tools and skills – add tools and skills that mirror the new UI pages, so coding agents can find the same high-priority issues with AI agents and features through MCP

Objective 2: Roll out offline evals Radu RaiceaRadu Raicea

Motivation: Offline evals shipped internally in Q3. We now need to roll them out to external users, iterate on their feedback, and reach general availability.

What we'll ship:

• External rollout and iteration – roll out offline evals to external users and improve the experience based on their feedback
• Billing and data retention – add billing support and apply data retention policies to offline evaluation data
• General availability – bring offline evals to general availability after the external rollout and iteration

Objective 3: Multimodal general availability Carlos MarchalCarlos Marchal

Motivation: Multimodal ingestion and trace views are available. We need to make them generally available with clear limits and pricing, and support large text content too.

What we'll ship:

• General availability – finalize the cap and pricing for multimodal content
• Large text content – extend multimodal storage and ingestion support to large text inputs and outputs

Objective 4: Query and loading optimization

Motivation: Slow queries and loading times make investigations frustrating. We need improvements in both the UI and the underlying queries.

What we'll ship:

• UI optimizations – improve loading performance in the UI Carlos MarchalCarlos Marchal
• Query performance investigation – investigate lower-level speedups for slow queries Radu RaiceaRadu Raicea

Objective 5: Useful clusters Radu RaiceaRadu Raicea

Motivation: Clusters need to surface useful, actionable findings, and their cost needs to make sense for the product.

What we'll ship:

• Opt-in clusters – make clustering opt-in
• Customizable summarization prompts – give teams control over how their traces are summarized for clustering
• Richer cluster centroids – enrich centroids with evaluation results and show how they evolve over time
• Pricing exploration – explore pricing that makes clustering sustainable

Objective 6: Online evals improvements Bernat TorresBernat Torres

Motivation: Teams need evaluations that go beyond boolean results, work with multimodal content, and make the results easy to understand.

What we'll ship:

• More output types – support categorical, multi-categorical, and numerical evaluation results
• Multimodal evaluations – evaluate multimodal content
• Results reporting – improve the UI for reporting evaluation results

Objective 7: Use case / agent primitive Bernat TorresBernat Torres

Motivation: Teams need to distinguish between the different use cases and agents in their product to understand their behavior.

What we'll ship:

• Use case / agent primitive – introduce a way to organize AI Observability data by use case or agent

Handbook

Who we're building for

Product Engineers and Full Stack Developers who are building:

• AI-native products (agents, assistants, copilots, specialized hardware)
• AI-adjacent products (LLMs integrated into existing products)

AI Observability is a good fit if:

• They need to monitor traces, spans, token costs, latency, and analyze usage of AI features
• They're using trace summaries to debug and evals to make product decisions
• They care about questions like: “how does interacting with LLM features correlate with retention, usage, or revenue?”
• They're already PostHog users (or should be) using product analytics and session replay to combine qualitative context with quantitative data
• They want to start getting value right away, without needing extensive setup and configuration

Who else might want to use AI Observability at their org:

• Application Ops / SRE to monitor production AI systems for errors, prompt injection, jailbreaks, or other security issues
• Product Managers to understand user sentiment, usage and make decisions about their AI roadmap
• Customer Success / Support Teams to improve documentation or investigate user issues

Who we're NOT building for (right now)

AI Researchers and Machine Learning (ML) Engineers doing:

• Deep foundation model work
• Complex benchmarking and evaluation
• Advanced experimentation requiring specialized tooling

These folks are running CI/CD pipelines and building full QA automation frameworks with performance benchmarks. If they try us and churn, that's fine. We haven't built the tools they need (yet).