The Pragmatic Engineer 2025 Survey: What's in your tech stack?

Gergely Orosz · article · 2025-07-15 source ↗ Reader 🖥️

Coding Agents Enterprise AI Adoption & Readiness Vibe Coding

AI coding-tool adoption among software engineers has become near-universal (85% of ~3,000 respondents) by mid-2025, but which tools are used is strongly shaped by company size and policy — larger companies converge on a single vendor-supplied tool (GitHub Copilot) while smaller companies permit broader experimentation (Cursor, Claude, ChatGPT) — and genuine engineer adoption should be distinguished from 'vibe coding' tool usage, which the data shows is mostly non-engineers (founders, directors, engineering leads) prototyping rather than hands-on developers building production software.

theoretical · mixed — Primarily an empirical reader-survey report (2,997 respondents, April-May 2025) with the author's interpretive commentary layered on top. This extraction focuses on the 'AI tools' section and its company-size breakdowns, per the task's scope — the article's later sections (languages, IDEs, version control, cloud) are outside that scope and are not extracted here. The survey is self-selected from this newsletter's own readership, which skews toward senior/engaged engineers already following AI-engineering coverage — a sampling caveat the article does not raise about itself.

Structure

Part	Role
Headline finding: 85% of respondents mention at least one AI tool; only 4% explicitly report using none	evidence
Most popular AI tools and year-over-year shifts: GitHub Copilot overtakes ChatGPT as most-mentioned; Cursor surges; Claude closes most of the gap with ChatGPT; Claude Code shows strong pre-launch traction	evidence
AI tools by company size: larger companies skew toward GitHub Copilot and single-tool policies; smaller companies skew toward Cursor/ChatGPT/Claude/Zed and multi-tool experimentation; Gemini usage is size-independent	central mechanism / evidence
Lower-profile tools, with commentary on vibe-coding tools' actual user base (mostly non-engineers) and on Devin's low developer adoption despite a 'replace engineers' marketing pitch	evidence / counter-example

Key terms

vibe coding tools
A distinct category from general-purpose AI coding assistants (Copilot, Cursor, Claude Code): tools like Vercel v0, Bolt.new, and Lovable, whose survey mentions come overwhelmingly from founders, director+/engineering-lead roles, and only secondarily from staff+ engineers — framed as prototyping/demo tools rather than tools hands-on engineers use for production work.
AI tools by company size
The survey's cross-tab of tool mentions against company-size bands (50 or fewer up to 10,000+), used to infer that company size proxies for AI-tool policy strictness: bigger companies narrow choice to vendor-supplied tools (and to a single tool), smaller companies allow broader experimentation.

Propositions

explicit85% of 2,997 survey respondents (2,555 people) mention using at least one AI tool; only 4% (130 people) explicitly report using none, citing workplace prohibition, lack of perceived usefulness, or ethical concerns.conf: high
extends2024_final_dora_report · p1 — DORA's 2024 report found 75.9% daily-task AI adoption; this survey a little over a year later finds 85% report using at least one AI tool — consistent continued-adoption trajectory from a different measurement method and population.
explicitGitHub Copilot overtook ChatGPT as the single most-mentioned AI tool year-over-year (roughly every second respondent now reports using it), while Claude closed most of the gap with ChatGPT that existed a year earlier — a shift the author attributes to Anthropic's Sonnet 3.5/3.7/4.0 releases, which are seen as excelling specifically at coding.conf: high
explicitCompany size strongly predicts which AI tools a developer uses: the larger the company, the more likely developers use GitHub Copilot (except at the very largest, 10,000+-employee tier); the smaller the company, the more likely developers use Cursor, ChatGPT, Claude, or Zed instead — a pattern the author attributes to larger companies standardizing on enterprise-vendor-supplied tools (with Microsoft's enterprise sales reach cited specifically) while smaller companies more often allow 'use whatever you want' policies.conf: high
supportsState Of AI-Assisted Software Development · p5 — DORA names 'a clear and communicated AI stance' as one of seven organizational capabilities correlated with better outcomes; this survey's finding that larger companies converge on narrower, enterprise-standardized tool policy (and smaller companies on permissive 'use whatever' policy) is a concrete, tool-level instance of organizations differing precisely on how clear/enforced that AI stance is.
explicitCompanies with 10,000+ employees are the only size band where the average respondent mentions using just one AI tool rather than multiple, which the author reads as evidence that the very largest companies have the strictest AI-tool policies, permitting only a single sanctioned tool.conf: medium
explicitAt the largest companies, AI-tool policy can be restrictive enough to ban third-party LLM products outright: a developer source at Amazon reports that ChatGPT and other third-party LLMs are banned, with only internally-sanctioned tools (Amazon Q, internally-hosted Claude) permitted — meaning 'AI tool adoption' figures at such companies measure policy compliance as much as developer preference.conf: medium
supportsAI-Assisted Software Development: Team Profiles and Capabilities for Putting Research into Action · p3 — Both treat company/organizational AI policy as a measurable, consequential variable rather than background noise — this survey's Amazon anecdote (third-party LLMs banned, only internal tools allowed) is a concrete instance of the kind of organizational AI-stance variation DORA's capability framework is built to capture.
explicitMost survey mentions of dedicated 'vibe coding' tools (Vercel v0, Bolt.new, Lovable) come from founders, director+ roles, or engineering leads (roughly two-thirds), with the remaining third split between staff+ engineers and other software engineers — suggesting these tools serve prototyping/demo needs for less hands-on roles rather than being broadly adopted by individual-contributor engineers for production work.conf: medium
explicitDevin, an AI coding agent marketed as a replacement for software engineers ('crush your backlog with your personal AI engineering team'), shows roughly 10% of the survey mentions that Windsurf has and about 1% of the mentions Cursor has — read by the author as evidence that marketing an AI tool as a developer replacement undermines its adoption by the developers whose use would ultimately drive it, even as that positioning may help with non-developer buyers.conf: medium

Arguments

An AI coding tool that markets itself as a replacement for software engineers, rather than an assistant to them, undermines its own developer adoption even if the positioning serves a separate sales narrative to non-developer buyers. strength: medium

Explicit premises:

Devin markets itself explicitly as replacing developers ('personal AI engineering team', 'crush your backlog').
Devin shows markedly lower developer-mention rates in the survey than comparable coding-agent tools (Windsurf, Cursor).
Devin was previously shown to have faked parts of its launch demo, which the author separately notes as a credibility problem.

Implicit premises:

Low mention counts among this newsletter's self-selected engineer readership reflect genuinely low developer adoption, rather than the tool being newer, less marketed to engineers specifically, or used more by non-respondent populations.
A tool's marketing framing (rather than its underlying technical capability) is doing meaningful causal work in developers' tool choice, separable from other differences between Devin and its competitors.

Inferences:

A company pursuing both an 'AI replaces engineers' narrative for leadership/investor audiences and grassroots developer adoption for its product faces a structural tension between the two goals, and will likely have to choose one emphasis over time.

Narratives

The Devin/Windsurf/Cognition saga: OpenAI's $3B offer for Windsurf collapsed over Microsoft's IP rights in OpenAI; Google then 'bought out' Windsurf's core team and founders for $2.4B (who all moved to work on Gemini), leaving Windsurf a 'zombie' company with $100M cash and remaining staff; Cognition (Devin's maker) then acquired the leftover Windsurf entity.
→ Grounds the proposition that Devin's maker needed developer traction it lacked organically, by showing the lengths (acquiring a competitor's remnant) it went to in pursuit of that traction.

Questions addressed

✅ How does AI coding tool choice vary by company size, and why?
Larger companies converge on fewer, enterprise-vendor-supplied tools (chiefly GitHub Copilot) under stricter policy; smaller companies permit broader multi-tool experimentation.
✅ Are 'vibe coding' tools being adopted by engineers themselves, or by a different population?
Mostly by non-hands-on roles (founders, directors, engineering leads) for prototyping, not by individual-contributor engineers for production work.
⭕ How representative is this survey's company-size and tool-adoption breakdown of the broader engineering population?

Gaps

incompleteThe survey sample is self-selected from The Pragmatic Engineer's own newsletter readership — engineers already engaged enough with tooling discourse to read and respond to this specific newsletter — which plausibly skews toward more senior, more AI-engaged respondents than the general developer population. The article presents company-size breakdowns and adoption percentages without flagging or correcting for this selection effect anywhere.
Read next: A survey run through a more neutral, broad-reach channel (e.g. Stack Overflow's or JetBrains' annual developer surveys) to check whether the company-size pattern found here replicates outside a tooling-newsletter audience.
incompleteThe survey measures only which AI tools respondents mention using, not any team- or process-level outcome (code review practice changes, who owns an AI-authored PR, delivery throughput or stability) — so it cannot speak to whether the company-size-driven adoption patterns it documents translate into better or worse team-level outcomes, which is exactly the question this reading's broader initiative (teams + coding agents in larger orgs) needs answered.
Read next: extract-dora-2025-state-of-ai-assisted-software-development.json and extract-dora-ai-team-profiles-and-capabilities.json, which measure team-level capabilities and outcomes rather than tool mentions.