By September 2026, more than 40% of GitHub pull requests started with a coding agent’s first draft — up from 17% in March. The problem: those agent PRs wait five times longer for review than human-authored ones. Not because reviewers are slow. Because the agents have no idea what your team actually cares about. LinearB launched Agentic Studio today to fix that.
What Agentic Studio Actually Is #
Agentic Studio is a unified context layer that sits over your entire engineering delivery footprint — Git repositories, project trackers, sprint history, code ownership maps, rework rates, incident logs — and exposes it all through the Model Context Protocol (MCP). Any coding agent that speaks MCP can now query this context before making a single commit.
The architecture is straightforward: LinearB normalizes data from your Git provider and project management tools into one model, links identities across those systems, and traces every change from ticket to production release. That data becomes queryable by your agents at runtime — so instead of writing code in a vacuum, your agent knows which services have had recent incidents, who owns what, and what your team’s rework patterns look like.
The Effort Skill: Right-Sizing Agent Work #
The first agent skill shipping with Agentic Studio is the one you actually need. The Effort skill decides how much effort a coding task warrants before the agent touches the code. It grades each task Low, Medium, or High based on business relevance, complexity, and risk, cross-referenced against LinearB’s delivery data: rework rates, incident history, ownership patterns. Low-risk changes get lighter edits and fewer tokens burned. High-risk or high-priority tasks get thorough, carefully tested work. The skill is agent-agnostic — it works with Claude Code, Cursor, Copilot, or anything else you’re running.
claude mcp add --transport http linearb https://mcp.linearb.io/mcp
The skill also reports every effort decision back to your LinearB org as a custom metric, so engineering leaders can track adoption and grades per developer and per repo. It is currently in early access — request it through your LinearB account dashboard.
Kraken’s PR Review Problem Was Not What Anyone Expected #
Kraken, the cryptocurrency exchange, used LinearB’s MCP integration to diagnose why its code reviews were lagging. Everyone assumed the culprit was oversized pull requests — the kind of bloated diffs that agents tend to produce. The data said otherwise. The real bottleneck was timezone misalignment: reviewers and authors working opposite shifts, creating hours of idle wait between each round-trip. Kraken reorganized its review workflows based on that finding and cut review time at the 90th percentile by more than half in four months. “That’s the structural shift,” said Nik Sudan, engineering operations lead at Kraken. “It moves leaders from anecdote-driven to evidence-driven.”
Context Engineering Is the Actual Bottleneck Now #
The agent coding wave is not coming — it arrived. LinearB’s own 2026 benchmarks from 8.1 million pull requests show that heavy AI users merged 96% more PRs than lighter users. But agent PRs are 2.6 times larger and wait five times longer for review, and they merge less often without human ownership. The gap is not intelligence — agents can write functional code. The gap is organizational context. Agents write code without knowing your team’s norms, current sprint priorities, or which service had an outage last week. That gap is expensive, and it scales with every agent you add.
LinearB’s move is to make that context programmable. MCP has become the standard protocol for agent-to-tool communication, and plugging delivery data into it is the logical next step. Whether it becomes infrastructure everyone runs is an open question — but the problem it solves is real and getting more expensive by the quarter.
How to Get Started #
Agentic Studio is available to all LinearB customers today, no upgrade required. The MCP endpoint is https://mcp.linearb.io/mcp. OAuth is the recommended auth method; API keys are supported for clients that do not handle OAuth. Full setup is in LinearB’s MCP server documentation. The Effort skill is in early access — request it through your account dashboard.