906 Engineers Just Told the Pragmatic Engineer What They Actually Use. 46% Said Claude Code. Here's What the Adoption Curve Tells You About the Next 12 Months.

In early 2026, The Pragmatic Engineer surveyed 906 software engineers with a median of 11-15 years of experience, mostly in the US and Europe. The results are worth sitting with.

95% use AI tools at least weekly. 75% use AI for at least half of their software engineering work. 56% say AI handles 70% or more of their engineering. 55% now use AI agents, not just autocomplete.

When asked which tool they love most: 46% said Claude Code. 19% said Cursor. 9% said GitHub Copilot.

Claude Code didn't exist 12 months before this survey. It went from zero to the most-loved tool among experienced engineers faster than any developer tool I can remember. Faster than Cursor's rise. Faster than GitHub Copilot's early adoption. Faster than VS Code dethroning Sublime Text.

That adoption speed has implications. Let me work through them.

The survey demographics matter

This isn't a survey of developers in general. The median experience is 11-15 years. These are engineers who've used many tools across many cycles. When a cohort with that level of experience converges on a new tool at 46% preference in under a year, it's not hype. Experienced engineers are notoriously skeptical of new tools precisely because they've seen so many hyped tools fail to deliver.

The fact that experienced engineers are the early adopter cohort for Claude Code (rather than juniors) is significant. Senior engineers set tooling standards for teams. When a senior engineer loves a tool, it propagates down and across: to the developers they hire, the teams they advise, the companies they found.

The 55% agent adoption figure is the other number worth pulling out. These aren't developers who use an AI for autocomplete or occasional code review. They're running workflows where the AI takes multi-step actions (reading files, writing code, running tests, iterating based on results) without constant human interruption. That's a materially different working style, and it means the "AI as assistant" framing is already obsolete for more than half of experienced developers.

What 8 months to #1 actually means

GitHub Copilot launched in June 2021. By early 2026, it was at 9% preference in this survey, after nearly five years, Microsoft distribution, GitHub integration, and enterprise sales contracts.

Cursor launched in 2022. It hit 19% preference after 4+ years of iteration, $60M in funding, and a viral developer community.

Claude Code launched in May 2025 and hit 46% preference in a survey conducted in February 2026, nine months later.

The obvious question is whether this is a temporary surge or a durable platform position. I think it's durable, and here's the argument: the preference questions asked which tool developers love, not which one they use for corporate compliance. Claude Code users at 46% preference are there because it's producing materially better output for their actual work, not because Anthropic has an enterprise sales team enforcing it.

When adoption follows genuine quality preference in an experienced professional cohort, it tends to lock in. The tool becomes the default in new projects. It becomes what you reach for when you're stuck. It becomes what you recommend when someone asks. Network effects that have nothing to do with marketing.

What this means if you're building developer tools

If you're building anything for developers (an IDE extension, a deployment tool, a monitoring product, a framework, a CLI utility), the distribution surface just consolidated.

Claude Code has a marketplace. It accepts MCP extensions. It integrates with tools through a published API. The developers who are most engaged, most senior, and most likely to pay for premium developer tooling are clustered around it.

For a solo operator building developer tools, the calculus used to be: build for VS Code (dominant IDE), consider GitHub integration (dominant code host), maybe JetBrains for enterprise. The calculus now includes: build for Claude Code's extension model, support MCP so your tool is discoverable inside the agent workflows that 55% of these developers run.

I've been working through this for my own projects. Tools that integrate cleanly with Claude Code's agent mode (that expose their functionality as callable MCP tools rather than requiring a browser UI or a separate CLI session) have a different distribution surface than tools that don't. The developers who'd most likely use and pay for what you're building are running agent workflows. If your tool doesn't show up in those workflows, you're invisible to them.

The two-to-four tools pattern

One more detail from the survey: most developers are using two to four AI tools simultaneously, not just one. Claude Code at 46% doesn't mean 54% use nothing else. The same engineer might use Claude Code for heavy coding sessions, Cursor for quick edits in an existing codebase, and a separate model API for testing pipelines.

This multi-tool reality means the distribution opportunity isn't exclusive. Being good at one thing that integrates with Claude Code's workflows doesn't require being a Claude Code competitor. The agents these developers run regularly call external tools. Being one of those tools is the opportunity.

The pattern I'm watching: tools that make Claude Code better (better context retrieval, better deployment hooks, better test output parsing) are gaining faster than tools that try to compete with it on coding quality.

The honest counter-take

Tools reach #1 and then decline. VS Code overtook Sublime Text, and Sublime Text is still around. Cursor rose fast and is still at 19% despite Claude Code's surge. Developer tool preferences are stickier than consumer preferences but not permanent.

There's a world where Anthropic fumbles the pricing, or OpenAI ships something substantially better in the next 12 months, or a new category of tool (AI-native IDE built ground-up for agent workflows rather than adapted from a chat interface) displaces the current generation. The survey data tells you where the platform is in May 2026, not where it'll be in May 2028.

What I'm confident about: the structural shift from autocomplete to agents is durable regardless of which specific tool wins. 55% agent adoption among experienced developers in early 2026 is a leading indicator. That number doesn't go down. The developers who've adopted agent workflows aren't going back to manual code completion. The opportunity for tools that fit into agent workflows is large and growing.

Build for where the developers are, with an eye toward where the workflow is heading, not where it was two years ago.

Author

Lukas

@lukcombinator

Sources

• AI Tooling for Software Engineers in 2026 | The Pragmatic Engineer
• Pragmatic Engineer Survey: 95% of Devs Use AI Weekly, Claude Code Tops the List | AI:PRODUCTIVITY
• Claude Code Usage Statistics 2026 | SERP Sculpt
• AI Weekly: Claude Code Dominates, MCP Goes Mainstream — Week of March 5, 2026 | DEV Community