75% Faster Plans, 2x Daily Applies: How Cursor Rebuilt Its Terraform Workflow with Masterpoint

At Masterpoint, we focused on Terraform and IaC improvements so our client Cursor could focus on making their users extraordinarily productive.

Sidebar: Key Value Delivered

• Dramatically faster plans: Plan time dropped significantly. Originally plan time had spikes of 10+ minutes; after our work, plan time was roughly 2 minutes.
• Improved delivery speed: Engineers shipped 157% more PRs per week after the engagement, and have increased shipping cadence since.
• More frequent, safer deploys: Weekly infrastructure PRs grew from 194 to 500+, and fewer plans were abandoned as the apply-to-plan ratio improved from 55% to 64%.
• Safer, more contained changes: The median Terraform run now touches 5 resources instead of 54 — a 91% smaller blast radius, so engineers can reason about exactly what a change will affect before applying it.

The Problem: Infrastructure That Couldn't Keep Up

By late 2025, Cursor's Terraform setup had become one of the biggest friction points in the engineering org. The company behind the Cursor IDE experienced issues familiar to fast-growing startups: infrastructure code written quickly, then scaled by copy-pasta rather than through careful design.

Those early Terraform patterns were replicated at machine speed as engineers dogfooded Cursor. They were copied and instantiated hundreds of times without the architectural guardrails needed to keep the system manageable. The result was a classic terralith: critical infrastructure was managed as a single monolithic production workspace with over 7,000 resources in one state file.

And because that workspace was so large, it was too slow.

"We had one giant workspace that took way too long to plan. It was killing us.” — Travis McPeak, Security Lead, Cursor

But the problem wasn't just the lack of speed. It was a lack of engineering confidence in the IaC system. Issues included:

• The AWS console was used to make changes during incidents and these fixes were never rolled into TF, and so were later inadvertently reverted.
• Managing ECS deployments through Terraform caused constant drift in the production terralith, flooding PR diffs with hundreds of unrelated changes.

Over time, the team stopped trusting the system, let alone reading the diffs.

"We would have engineers click apply on prod workspaces that had 120 ECS service changes. The ECS services would change all the time because of drift, so the team became desensitized to large Terraform plan changes. — Travis McPeak

This confusion caused downtime, such as a network firewall change that caused a 10 minute outage.