Linear AI in 2026: Linear Agent, Triage Intelligence and the Context Linear Cannot See

TL;DR

Linear AI is the set of AI features built into Linear. Linear Agent is a built-in assistant on every plan, including Free, that answers questions about the workspace and can create and update issues. Triage Intelligence suggests teams, assignees, labels and duplicates on the Business and Enterprise plans, and Code Intelligence lets the agent answer questions about connected GitHub repositories. Loops and coding sessions run on prepaid AI credits, and an MCP server lets Claude, Cursor and other tools read and write Linear. All of it works from what is inside Linear, plus connected code and any Slack thread where someone mentions @Linear. The reasons behind the work usually live in Slack, on customer calls or in email. Sentra, the managed organization memory for teams and AI agents, reads Linear alongside Slack, email, meetings and GitHub, keeps each fact with its source and the newest version on top, and serves that memory to people in Slack and to agents over MCP.

What Linear AI includes today

Every feature and plan below was checked on 19 October 2026 against Linear's pricing page, its documentation for Linear Agent, Triage Intelligence, Code Intelligence, AI credits and the MCP server, and the changelog.

On Business, Linear Asks also turns requests sent in Slack or by email into issues.

What Linear AI costs

Linear Free costs nothing, allows unlimited members, two teams and 250 issues, and includes Linear Agent and the agent platform. Basic costs $10 per user per month billed yearly. Business costs $16 per user per month billed yearly and adds Triage Intelligence, Loops, Code Intelligence, Insights and Linear Asks. Enterprise is priced on request with annual billing only. These prices come from linear.app/pricing, checked 19 October 2026.

Coding sessions and Loops draw on prepaid AI credits rather than the seat price. According to Linear's AI credits documentation, a coding session is billed for model tokens at the provider's published rates with no markup, sandbox time costs $0.25 per 20-minute block, and a typical Loop run without a coding session costs $0.07 to $0.20. The minimum top-up is $10 and purchased funds expire 12 months after purchase. For a 20 person team on Business, seats come to 20 × $16 × 12 = $3,840 a year before any credits.

Recent Linear AI updates, July to October 2026

• 20 July 2026: Loops launched as recurring agent workflows for teams, with $20 per seat in promotional credits.
• 13 August 2026: Enterprise-managed authorization for the MCP server through Okta.
• 20 August 2026: Coding sessions gained configurable environments, browser use and per-token pricing with spend limits.
• 24 September 2026: Adaptive model routing, environment secrets and open-weight models for coding sessions.
• 8 October 2026: Stacked pull requests in Linear Diffs and an AI credit usage explorer with spend limits.

What Linear AI cannot see

Linear Agent works within each user's Linear permissions. It reads the workspace's teams, projects, cycles, initiatives, issues, comments, activity history and documents, plus connected GitHub code when Code Intelligence is on. In Slack it acts on the conversation where someone mentions @Linear, and admins can add MCP servers such as Granola, Notion and Figma to reach further. That covers the work well. It covers far less of why the work looks the way it does.

• Decisions made in Slack channels. A product call made in a channel last Tuesday reaches Linear only if someone mentions @Linear in that thread or copies the outcome into an issue. Linear Asks captures the original request as an issue, but the discussion that followed in other threads stays in Slack.
• Commitments from meetings. A date promised to a customer on a call lives in the transcript. It reaches the issue only if someone writes it down or an admin connects a meeting tool over MCP.
• Customer email. Escalations and the thread where a customer explained what they need sit in inboxes Linear does not read.
• Which version is current. When a decision changes outside Linear, the old version can survive in an issue description, a Slack thread and a doc at the same time, and nothing in Linear knows which one replaced the others.

The gap matters most for agents. A coding agent assigned an issue reads the issue, its comments and the repository. If the constraint that shaped the issue was agreed on a customer call, the agent never sees it, and it can build exactly what the ticket says and still miss what the customer needed.

How teams give agents around Linear company context

Teams usually try one of four approaches, and they differ mostly in who keeps the context current.

1. Paste the context into the issue. This costs nothing, and it goes stale the moment the decision changes somewhere else.

2. Connect more MCP servers. Adding Slack, a meeting tool and a docs tool to Linear Agent or to a coding agent gives it more places to search. Each server is searched separately on every run, and the agent has to work out on its own which of three conflicting answers is current.

3. Build a retrieval pipeline. An internal index over Slack, docs and Linear is flexible, but the team owns the connectors, the permission checks and the freshness of every source for as long as it runs.

4. Add a shared memory layer. A managed memory reads the tools continuously, keeps one current version of each fact with its source, and serves the same answer to every person and agent that asks.

Ways to add AI to a Linear workflow, compared

How Sentra works with Linear

Sentra connects to Linear, GitHub, Slack, email, calendar and meetings and turns what flows through them into memory: decisions, owners, commitments and the reasons behind them, each kept with the message, meeting or issue it came from. When a decision changes, the newest version sits on top, so a person or an agent gets the current answer instead of three competing ones. Permissions follow the source tool. A fact from a private Slack channel is visible only to people who can open that channel, and Slack direct messages are not used to answer questions.

People ask Sentra in Slack, for example why a single sign-on issue was moved out of the current cycle, and the answer cites the customer call and the Slack thread as well as the Linear comment. Agents read the same memory over MCP. A coding agent that picks up a Linear issue can ask what was decided about the feature and get the constraint from the customer call before it writes any code. Linear stays the system of record for the work, and Sentra holds the context around it.

Frequently Asked Questions