Schedule risk for Linear users

Linear earned its reputation honestly. It is arguably the most refined product-engineering tool on the market: instantaneous interactions, a thoughtful issue and cycle model, and a roadmap view that product teams actually keep up to date. Triage, sub-issues, projects, initiatives, and the Linear Agent combine into a workflow that software teams adopt because it removes friction rather than adding process.

If your work is shipping software in cycles, Linear is a strong default and CritPath does not try to replace it. The distinction is one of category: Linear answers "what are we working on and in what order," and answers it beautifully. It does not answer "what is the probability we finish by the committed date, and which task is driving the risk" — because it was never built to.

Linear's roadmap is a presentation layer, not a model. A roadmap bar reflects a target window someone typed in; it is not derived from a dependency network, it does not roll up task uncertainty, and it does not move on its own when an upstream task slips. There is no critical path calculation, no float, and no simulation — so there is no P50, P80, or P90 finish date to put in front of a board or an investor.

The Linear Agent is genuinely useful, but it is context-aware over the work graph, not over a schedule. It can summarize, route, and update issues; it cannot tell you which task is on your critical chain, how much buffer you have consumed, or what a gate decision does to your launch date — because none of those quantities exist in Linear. That is the difference between an AI that reasons about your backlog and an AI that reasons about your risk.

CritPath AI is built around the math Linear leaves out. Its Monte Carlo engine runs PERT-Beta three-point distributions with risk-event injection, criticality index, tornado sensitivity, optional duration correlation, and P50/P80/P90 finish dates — the probabilistic backbone a serious program plan needs. Under that sits a full Critical Path Method (four dependency types, lag, float, near-critical analysis), so the dates are derived from the real network rather than typed onto a bar.

On top, CritPath layers the methods R&D programs actually run on: Critical Chain Project Management with Theory of Constraints and Drum-Buffer-Rope (drum, project and feeding buffers, live fever chart), decision gates with retroactive rescheduling that re-cascade the plan when a Go/No-Go/Pivot lands, WSJF and Cost of Delay, and Earned Value Management. All of it carries AACE RP 132R-23 Level 4 risk-driven scheduling with an append-only audit log, runs in a browser with multi-tenant orgs and 2FA, and imports your existing plan via XLSX/CSV.

And the AI copilot is the inverse of Linear's: built on Claude and Gemini and grounded in your actual dependency graph, it reasons over the live schedule, not the backlog. It can tell you which task is driving your P80 slip, how much project buffer you have burned, and what a gate decision does three steps downstream — a copilot that computes risk rather than flagging anomalies.

The table compares the two on the capabilities that decide whether you can defend a date, not just track work. Linear's pricing is its public Basic and Business tiers; figures drift, so treat them as approximate.

This is not an either/or for most teams. If your engineers ship in Linear, keep them there — it is excellent at execution, and CritPath has no interest in becoming your issue tracker. Use CritPath AI for the layer above: the program-level plan where uncertain tasks, constrained scientists and instruments, and gate decisions need probabilistic dates and buffers a roadmap bar cannot express.

The practical workflow is to run delivery in Linear and run schedule risk in CritPath, importing your milestone plan via XLSX/CSV. You keep Linear's velocity and the team's muscle memory, and you gain a defensible P80 date, a fever chart your sponsor can read in seconds, and an AI copilot that explains why a date moved — all for $10/user/month, with AI usage billed separately by metered token cost.