LinearB launches Agentic Studio to provide context for AI coding agents

From metrics to actionable context

LinearB is shifting the focus of developer productivity platforms from simple data collection to active context delivery. Historically, these platforms provided engineering leaders with various metrics, but the new Agentic Studio aims to turn that data into something people and agents can actually use. By creating a knowledge graph of the software delivery lifecycle, the company allows users to interact with their data more naturally.

Engineering leaders can now ask questions in plain language to receive specific answers and suggested next steps regarding their teams. Instead of manually interpreting complex charts, they can query the system to find out where delivery is stuck or how their developers are performing. This shift is designed to help leaders provide evidence-based answers to boards and CFOs who are increasingly questioning the return on investment for AI spending.

The platform achieves this by normalizing data from various project management and Git systems into a single model. It matches engineer identities across different tools and links every change from the initial ticket to the final release. This unified layer provides the necessary background information that both human managers and autonomous AI agents need to make informed decisions.

Empowering AI coding agents

A major hurdle for AI coding agents is that they often start tasks without knowing the history or specific nuances of a codebase. LinearB addresses this by feeding the same context layer to agents through its Agentic Studio and MCP (Model Context Protocol) servers. This allows agents to understand their environment better, leading to smarter decisions and higher-quality code outputs during the development process.

One of the first tools available in early access is the 'effort skill,' which evaluates a task's complexity and risk before an agent begins work. It gauges business relevance by looking at code quality in a specific area, recent service incidents, and the history of ownership for that code. This ensures that the AI knows which tasks require careful, tested edits and which ones can be handled with less effort.

By matching effort to the task, the system can optimize for both quality and cost. Low-risk changes use fewer tokens and less effort, while complex or high-priority changes receive more intensive processing. This approach helps prevent agents from making risky mistakes in critical areas of the software while maintaining speed on routine updates. Leaders can also monitor how often agents are routed to different effort levels.

Solving structural bottlenecks

The platform is also being used to identify structural issues within the software factory that are not immediately obvious from raw numbers. For example, the company Kraken used LinearB's MCP server to investigate why their code reviews were taking so long. They discovered that the primary bottleneck was not the size of the changes, but rather reviewer availability and differences in time zones.

By identifying these specific friction points, Kraken was able to cut review time at the 90th percentile by more than half in just four months. This demonstrates how the platform moves engineering operations from anecdote-driven observations to evidence-based improvements. It allows teams to see exactly where the process is breaking down so they can apply targeted fixes to the workflow.

LinearB's 2026 Software Engineering Benchmarks Report highlights that these structural issues are becoming more complex as AI adoption grows. The report found that agent pull requests often wait more than five times longer for review than human ones. This suggests that while agents can write code quickly, the human-led review process remains a critical point of congestion that requires better data to manage.

The state of AI in engineering

Data from LinearB's recent benchmarks shows a significant divide in productivity based on AI usage. Developers who use AI on at least 75% of their coding days merged 96% more pull requests per week than they did a year earlier. In contrast, developers who do not use AI saw their merged pull requests decrease by 3.6% over the same period.

However, the report also notes that AI adoption is currently shallower than some might expect, with AI assisting only 54% of pull requests even at elite organizations. Furthermore, the cost of AI is rising quickly at the top end of the market. Token spend for organizations at the 90th percentile reached $35.20 per developer per coding day, a nearly 75% increase in just three months.

These trends indicate that while AI is a powerful tool for increasing output, it requires sophisticated management to be effective and affordable. LinearB aims to help organizations navigate these costs by providing the context needed to use AI more efficiently. The goal is to ensure that AI-assisted work is not just faster, but also safer and more aligned with business priorities.