36% of AI-generated code changes are now accepted without manual review. Six months ago that number was 7%. That's from Cursor's Spring 2026 Developer Habits Report, out today. Worth a read if you care about where software development is actually heading. A few things stood out to me and how I think about them for the teams I work with: The cost story is underappreciated. Developer output doubled YoY (3.6K to 8.6K lines/dev/week), but model costs vary by nearly 9x across model families. If your developers are defaulting to the most capable model for every request, you're burning budget on tasks that don't need it. Model agnosticism with a smart harness is how you get the output gains without the runaway spend. Better code, faster and cheaper isn't a slogan, it's an optimization problem. AI is moving past the "write code" stage. Automated acceptance is up 5x. Agent sessions are getting deeper. The teams treating AI as a single-stage writing tool are missing most of the value. The real opportunity is removing bottlenecks across the full SDLC: planning, review, testing, deployment. That's where the compounding happens. The power user gap should concern enterprise buyers. P99 users are producing 46x more output than the median active user. Add the 9x cost variance, and without governance controls, AI spend becomes nearly impossible to predict or audit. Governance isn't a brake on productivity. It's how you scale it without losing control of cost or security. The report has a lot more. Curious what others are seeing on the governance and cost control side as these tools mature. Jason T. Yttre Prem Iyer Gregg Burke Chris Amaefule Biz Eshetu Mike Kinsella Joe Gaitley Chris Pagano

We just published Cursor's first Developer Habits Report! One number that stood out to me: lines of code per developer per week went from 3,600 to 8,600 in 18 months. That's not a rounding error. What's interesting isn't just the volume. It's where the growth is coming from. Mega PRs (1,000+ lines) nearly doubled as a share of all merged PRs. The top 1% of power users are pulling further ahead, fast. If you lead an engineering org, worth 10 minutes of your time: https://lnkd.in/gW_Qd6xi

👀 Introducing Cursor's Developer Habits Report - based on the most comprehensive dataset of AI coding in the world, across ALL MODEL FAMILIES. Check it. https://lnkd.in/ez3SEJfh

Two things that struck me most in the report: 1. The huge spread in model costs. In what other realm are people willing to pay 7-9x for the same outcome? 2. The power user gap. If you wanted one stat to explain how AI will affect the industry and society, that p99 v p50 multiple might be it.

We just published our Developer Habits Report, and one of the most interesting findings is that the top 5% of users produce over two-thirds of the total code generated on the platform. The adoption of agents is rapidly growing, but it's still centered around early adopters and power users, with most companies only just starting to scratch the surface of what's possible. If you're interested in trying out Cursor for your team, we're offering $500 in credits - get started at https://lnkd.in/eyfbMF23

Cursor’s Spring 2026 Developer Habits Report is a fascinating look at how quickly software development is changing in the age of AI agents. The headline is pretty clear: developers are moving faster. Cursor reports that lines added per developer per week rose from about 3.6K in January 2025 to 8.6K by mid-May 2026. PRs are also getting bigger, and “mega PRs” are becoming more common. But the more interesting takeaway, at least to me, is that AI is not just speeding up the old workflow. It's changing the workflow itself. Agent sessions are getting deeper. Models are reading much more context before they write. And since the start of 2026, the share of agent-generated changes reaching commits without a separate manual diff acceptance step has increased more than 5x. That points to a shift from AI as an assistant toward AI as part of the software delivery system. The report also shows a widening gap between average users and power users. P99 developers produce far more AI-generated code and merge far more PRs than the median user. That suggests the advantage may not come from simply having access to AI tools. It comes from learning how to use them well: how to provide context, structure workflows, review outputs, and build trust without giving up control. AI coding is no longer just about autocomplete or faster syntax. It's becoming a new operating model for software delivery, one where context, workflow design, governance, and developer enablement matter as much as the model itself.

Cursor's first Developer Habits Report is full of exciting numbers. One isn't. Lines per PR are up ~2.5x year over year, and "mega PRs" (1,000+ lines) are now ~14% of merges. Meanwhile, changes accepted without manual review jumped from 7% to 36% since January. Bigger diffs, less review. That's exactly where bugs hide. A 50-line PR gets three comments; a 1,500-line PR gets a LGTM. AI helps us write more. I'm not convinced it's helping us review more. My bet: the senior folks end up with more work later on, cleaning up the AI slop that slipped through. What are your thoughts?

CRESS Principles for Context Engineering - C is for "Current" Imagine you’re trying to deliver groceries in a busy city using a map that was published in 1971. You’ll find yourself looking for houses, apartment blocks, entire neighbourhoods that didn’t exist when the map was drawn. This is what it’s like when an AI coding assistant or agent is trying to work on a code base using an out-of-date picture of the structure. With every change to the code, the map gets a little bit more out-of-date. With every change, the context gets a little bit more misleading. (Link to whole blog post in comments)

Vibe coding is changing how developers build software, and this article explores the tools behind it. Understand how to work with AI to generate and refine code. https://lnkd.in/gB-HdK7m