The Pragmatic Engineer just published their 2026 AI tooling survey. 906 developers responded, mostly engineers and engineering leaders with a median experience of 11-15 years. The results are clear.
Claude Code is the most-loved developer tool at 46%. Cursor sits at 19%. GitHub Copilot at 9%. If you include Claude's standalone chat product, the Claude ecosystem hits 57% combined.
I use Claude Code every day to build Sucana, so the top-line number did not surprise me. But one stat buried in the data stopped me cold.
75% of small startups (1-10 people) use Claude Code. At large enterprises (10,000+ employees), GitHub Copilot dominates at 56%.
That gap tells you everything about where software development is heading.
Small teams and solo founders pick the tool that gives them the most leverage. They do not have procurement processes, security review boards, or vendor compliance checklists. They try three tools in a week and keep the one that makes them fastest.
The survey confirms this. Companies that give developers freedom to experiment show higher Claude Code adoption. Companies that restrict tool choices show higher GitHub Copilot adoption. The correlation is direct.
Large companies pick the tool that fits their existing infrastructure. GitHub Copilot integrates with GitHub Enterprise, which integrates with their SSO, which connects to their audit trail. The decision is not about which tool is best. It is about which tool causes the least friction with IT.
This is not new. It happened with cloud computing (startups on AWS years before enterprises moved off on-prem), with Slack (startups adopted it while enterprises were still on Skype for Business), and with Vercel (startups deploying in seconds while enterprises managed Jenkins pipelines).
The pattern is always the same. Small teams adopt the better tool first. Enterprises follow three to five years later.
Before talking about which tools, consider how pervasive AI has become. The survey found that 95% of respondents use AI tools weekly or more frequently. Only 2.1% do not use AI at all.
And they are not just dabbling. 75% use AI for at least half their engineering work. 56% use AI for 70% or more of their work. These are not people experimenting with autocomplete. These are developers who have fundamentally changed how they build software.
Most developers also juggle multiple tools. 70% use 2-4 AI tools simultaneously. 15% use 5 or more. The days of picking one tool and sticking with it are over. Developers are assembling their own AI toolchains the same way they assemble their tech stacks.
The Real Shift: From Copilots to Agents
The developer love numbers are interesting, but they are not the most important finding in the survey. This is: 55% of developers now regularly use AI agents. Not copilots. Agents.
There is a meaningful difference.
A copilot sits next to you and suggests code as you type. You are still driving. You are still making every decision. The AI fills in boilerplate and autocompletes function bodies. GitHub Copilot pioneered this model in 2022.
An agent takes a task and executes it end to end. You describe what you want. The agent inspects your codebase, understands the architecture, generates the changes across multiple files, runs the tests, and opens a pull request. You review the output instead of writing the code.
We went from "AI helps me write code" to "AI ships features while I review."
The adoption varies by seniority. Staff+ engineers lead at 63.5% agent usage. Regular engineers are at 49.7%. Engineering managers at 46.1%. The most experienced developers are adopting agents fastest because they understand the leverage.
There is also a clear sentiment split. Among developers who use agents, 61% are excited about AI's impact on their work. Among those who do not use agents, only 36% are excited. Using agents does not just change your workflow. It changes how you feel about the future of your craft.
What This Looks Like in Practice
I build with AI agents every day. Not because it is trendy. Because three founders in three countries cannot build a SaaS product without them.
Sucana is built by me (Florida), Virgil (Bali), and Victor (Spain). We do not have a 10-person engineering team. We do not have dedicated frontend and backend developers. We have three people who each need to ship full-stack features independently.
Here is what a typical feature build looks like with Claude Code:
I open the terminal and describe what I want. "Add a multi-page analysis feature to GEOScore. When a user analyzes a URL, extract all internal links from that page and show them with Analyze buttons so the user can scan more pages and see an aggregate score."
Claude Code reads my codebase. It understands the types, the existing components, the API routes, the database schema. It generates changes across 4 files: the type definitions, the analyzer module, the report view component, and the homepage state management. Then I review, test, and push.
That feature would have taken a day or two with a traditional development approach. With Claude Code, it took about an hour including testing and iteration.
Multiply that by every feature, every bug fix, every refactor. Three founders shipping at the speed of a mid-sized team. That is the leverage AI agents provide.
The Numbers Behind the Shift
The survey has more data points that paint the full picture:
46% love Claude Code. The highest satisfaction of any AI developer tool. Released just in May 2025, it went from zero to the most-loved tool in under a year. Developers cite the ability to work across multiple files, understand project context, and handle complex multi-step tasks.
19% love Cursor. Strong second place, with 35% growth in the last 9 months. Cursor excels at the editor-integrated experience. I use it too, mainly as a smart editor with good autocomplete when I want to read and manually adjust code.
9% love GitHub Copilot. Down from its dominant position two years ago. Copilot was first to market and still has the most users at 46% adoption, but satisfaction has dropped as developers compare it to newer tools.
Claude Code love increases with seniority. Directors and above love Claude Code at twice the rate of junior developers. The more experience you have, the more you appreciate what an agent can do versus a copilot.
75% of small startups use Claude Code. Small teams optimize for output per person. When you are three people building a product, the difference between a copilot and an agent is the difference between shipping one feature a week and shipping three.
Codex appeared from nowhere. OpenAI's Codex reached 60% of Cursor's usage despite not existing 9 months ago. The AI developer tools market is moving incredibly fast.
What This Means for Solo Founders and Small Teams
If you are building a product with a small team, you are living in the best era of software development. The tools available to you today would have required a 20-person team five years ago.
A solo founder with AI agents can:
- Ship a full-stack web application in a weekend
- Add authentication, database integration, and API routes in hours
- Fix bugs across multiple files in minutes
- Refactor entire codebases without fear
The quality ceiling has gone up too. AI agents do not just write code faster. They write code that follows patterns, handles edge cases, and includes error handling that a tired developer at 2 AM might skip.
The solo founder with AI agents will compete with 50-person teams. That future is not coming. It is already here. I am living it.
The Enterprise Gap Will Close
Large enterprises will adopt AI agents eventually. They always do. But right now, while they are evaluating vendors and writing RFPs, small teams are shipping products at unprecedented speed.
If you are at a big company watching this from the inside, the frustration is real. You know Claude Code is better. Your procurement team does not care. They need SOC 2 compliance documentation and a signed enterprise agreement.
That gap is temporary. Anthropic and other AI companies are building enterprise features as fast as they can. But for the next two to three years, small teams have a structural advantage. They can use the best tools immediately while enterprises wait for the paperwork.
What I Would Do If I Were Starting Today
If I were starting a product today with no team and no funding, here is my stack:
- Claude Code for all coding. Terminal-based, context-aware, handles multi-file changes.
- Next.js for the framework. Server rendering, API routes, deployment on Vercel.
- Tailwind CSS for styling. AI agents are very good at generating Tailwind classes.
- Supabase for the database. Postgres with a generous free tier.
- Vercel for hosting. Push to deploy, no infrastructure to manage.
Total cost: $0 until you have users. Total team size: one person. Total output: competitive with funded startups.
The playing field has never been more level. The Pragmatic Engineer survey just put numbers on what builders already know.
What AI dev tool are you using? And has it changed in the last 6 months? Find me on Twitter/X or LinkedIn.
Source: The Pragmatic Engineer AI Tooling Survey 2026, 906 respondents, January-February 2026.
