AI Adoption Score vs Span

Span (span.app) measures AI coding tools in engineering: how widely they're used, what they cost and how much of your code AI writes, from your repositories, your AI tools' admin data and a recorder on developers' machines. AI Adoption Score asks people how well they use AI and what holds back the ones who don't, in engineering and every other team. They answer different questions, so they work well together.

Last checked: October 17, 2026

Which one fits you

Two questions decide it.

Side by side

Scores out of 5: 4–5 strong 3 partly 1–2 weak or missing

Question	AI Adoption Score	Span
What it measures	Score: 5 out of 5.How people adopt AI: a level for every person and team across usage, skills, impact, culture and vision (opens in a new tab)	Score: 4 out of 5.AI coding tools' use and cost, AI-written code, and the throughput, review cycles and rework of AI-assisted work; plus delivery metrics such as DORA and cycle time^{2, 6, 7}
Who it covers	Score: 5 out of 5.Everyone: the 360° has one questionnaire for engineering and delivery and one for business roles, on the same scale	Score: 2 out of 5.Engineering: priced per code contributor, anyone who commits code or is tracked in your code tools; people who only view dashboards are free¹¹
AI tools covered	Score: 5 out of 5.Any: Copilot, ChatGPT, Claude, Gemini, Cursor, and tools people pay for themselves	Score: 4 out of 5.AI coding tools: Claude Code, Cursor, GitHub Copilot, Codex, Augment Code and VS Code, plus coding agents such as Devin, found in pull requests³
Where the data comes from	Score: 3 out of 5.Surveys and short quizzes that people answer	Score: 4 out of 5.Your source control, issue tracker and HR system, your AI tools' admin data and telemetry, a recorder on developers' machines, and surveys in Slack^{1, 2, 3}
Measured AI coding use	Score: 3 out of 5.AI Coding Usage: sessions, repositories, tools and connectors from Claude Code and Codex, for each engineer who connects their Mac (opens in a new tab)	Score: 4 out of 5.Adoption, use and spend per tool, team and person, and AI-written code found in pull requests, even before a tool is connected^{3, 6, 7}
What you connect	Score: 5 out of 5.Nothing for the surveys: no repositories, no tools, no admin access. For AI Coding Usage, each engineer pastes one line on their Mac	Score: 2 out of 5.Source control first, then your issue tracker, HR system and AI tools; Slack for surveys; for agent sessions, its recorder goes on developers' machines through device management^{1, 2, 3, 4}
Why people don't use AI	Score: 5 out of 5.18 barriers, each with who can fix it: IT, leadership, managers or training (opens in a new tab)	Score: 3 out of 5.Developer surveys on sentiment, friction and AI readiness, with themes by team, manager, level or tenure^{2, 8}
AI skills	Score: 4 out of 5.Who can do what with 125 AI tools and models, and quizzes that score n8n and Claude skills (opens in a new tab)	Score: 3 out of 5.Signals from agent sessions, such as how clear prompts are, and the developers who use AI most^{6, 10}
Attitudes	Score: 5 out of 5.AI anxiety on a validated research scale, plus culture and vision in the 360° (opens in a new tab)	Score: 3 out of 5.Developer sentiment and AI readiness, in its surveys for engineers⁸
Levels	Score: 5 out of 5.L1 Observer to L5 Native for every person, team and the organization	Score: 3 out of 5.Four stages for the engineering organization: AI Adoption, Strain, Proficiency and Mastery; developers grouped by how much AI they use^{6, 9}
Benchmarks	Score: 3 out of 5.Inside your company: every team against the others and the whole organization, and each round against earlier ones; no cross-company benchmark yet (opens in a new tab)	Score: 4 out of 5.Benchmarks from 103 engineering teams, from its own research^{1, 10}
Privacy	Score: 4 out of 5.Named or anonymous, set per survey; results from the first answer, with no minimum group size; AI Coding Usage never sends prompts, code or file names	Score: 3 out of 5.Results by team, person or type of work; its recorder captures prompts, the agent's tool calls and file edits, redacts secrets on the machine, and shows a session to its developer alone unless you widen that^{1, 5, 6}
Security	Score: 4 out of 5.No access to your systems or code; answers never go to an AI model, and anonymous surveys store no names, emails or IP addresses; no SOC 2 or ISO certification yet (opens in a new tab)	Score: 5 out of 5.SOC 2 Type II, per its trust center; the recorder isn't a proxy and never sits in the path of requests^{1, 13}
Price	Score: 5 out of 5.Free plan; paid plans on request	Score: 2 out of 5.No public price: per code contributor, from a one-year contract, with a free proof of concept for eligible teams; its own comparison pages list $45 per contributor a month^{11, 12}
Summary	How well people use AI and what holds them back, in engineering and every other team: levels, skills, barriers and worries, from surveys with nothing to connect and a free plan to start.	Deep, measured data on AI coding tools: use, cost, AI-written code and agent sessions, with benchmarks and SOC 2. But engineering only, with your code and AI tools connected.

Where each one sits

AI Adoption Score Surveys and quizzes on how people adopt AI, in engineering and business roles: levels, skills, barriers, attitudes.
Span AI coding tools' use, cost and AI-written code, from engineering data, plus surveys for engineers.

Positions are our reading of each product's public pages in October 2026, not a measurement.

Where Span is stronger

Measured, not reported. It reads your code and your AI tools' admin data, and tells AI-written code from human code in every pull request. Our surveys only know what people tell us, with checks, and AI Coding Usage counts sessions without reading code.^{3, 15}
AI's effect on delivery. It sets the throughput, review cycles and rework of AI-assisted work against everyone else's, by how much AI each developer uses.⁶
What AI costs. Spend by team, tool and model, actual and projected, in one view.⁷
Agent sessions. Its recorder keeps each coding agent's session, with its prompts, tool calls and file edits, tied to tickets and pull requests.^{1, 5}
Benchmarks and certification. Benchmarks from its research across 103 engineering teams, and SOC 2 Type II. We have no cross-company benchmark or certification yet.^{10, 13}

Using both

Span shows what AI changes in your engineering: the code, the cost and the delivery. AI Adoption Score shows the people behind it, in every team: who is at L2 and who at L4, how well they know Claude, what stops them and what worries them. Outside engineering, where Span doesn't go, the 360°'s business questionnaire puts sales, marketing or finance on the same scale.

When Span shows a team where AI isn't paying off, run AI Adoption Barriers (opens in a new tab) there: about 5 minutes per person, and it says who can fix each barrier. The Claude Skills Assessment (opens in a new tab), with extra questions for Claude Code, shows whether people know the tool or just use it a lot.

Questions

Does Span read our code or prompts?

It reads your repositories to measure delivery and to find AI-written code. Its recorder, on developers' machines, captures prompts, the agent's tool calls and its file edits, but not keystrokes, browsing or what developers type in the terminal themselves. It removes secrets and personal details on the machine, and by default only the developer can read their own sessions.^{3, 5}

Span says surveys fall short. Why use one?

Span's own post says self-reports are poor at measuring productivity, and that surveys still capture how people feel, where they hit friction and what leaders should watch. It recommends using both. Our surveys are built for that second part: why people don't use AI, what worries them, and how well they know the tools.¹⁴

Does AI Adoption Score measure AI coding tools at all?

Yes. AI Coding Usage (opens in a new tab) counts sessions, repositories, tools and connectors from Claude Code and Codex, for each engineer who connects their Mac with one line. Prompts, code and file names never leave the Mac. It doesn't read your repositories, so it can't tell AI-written code apart or measure delivery: for that you need a tool like Span.

Sources

Code shows what AI wrote. Ask what holds people back.

See how well every team uses AI and what stops the rest, engineering and beyond,
with nothing to connect and a free plan to start.

Start Free Book a Demo