AI Software

Enterprise Adoption by Use Case

Generative AI deployment percentages by business function from the McKinsey State of AI surveys 2022 through 2025. 2022 columns are null because the survey predates ChatGPT; the 2025 cut reports a broader AI-use bucket. BCG, Deloitte, and Bain cross-checks expandable below.

Organizations regularly using gen AI in at least one business function

2022

n/d

2023

33%

2024

65%

2025

n/d

2023 to 2024 step-up roughly doubled. McKinsey's 2025 survey reframed the question to 'organizations using AI in at least one business function' at 78% (broader category, includes traditional ML), so we leave the strict gen-AI-only 2025 cell null. The directional message: per-function gen AI adoption broadened across more functions in 2025.

Organizations using AI (including traditional ML) in at least one business function

2022

50%

2023

55%

2024

72%

2025

78%

McKinsey broader AI adoption headline (not gen-AI-only). Provided for trend context; reconciles the 2025 headline to the gen-AI cut.

McKinsey “The State of AI” annual survey, percent of respondents reporting regular gen AI use in the named business function. 2022 cells are null because McKinsey did not survey gen-AI-specific function adoption until April 2023 (after ChatGPT). 2025 cells are null where McKinsey reframed the question to broader “AI use” (including traditional ML); the summary aggregate row shows the broader trend. Click year pills to toggle series. Hover the bars for per-row sourcing notes.

Adoption by Industry

Percentage of organizations actively deploying generative AI by industry, anchored on McKinsey 2025 with Stanford AI Index and IDC cross-references in tooltips. Sorted descending. Technology and Financial Services lead; Manufacturing and Public Sector lag.

Primary source

McKinsey 'The state of AI in 2025: agents, innovation, and transformation'. Fielded May to Jul 2025, n = 1,491 respondents. Percent reporting regular gen AI use in at least one business function.

Adoption rates are not strictly comparable across sources: McKinsey reports “regular use in a business function”, Stanford AI Index aggregates job-posting and corporate-survey data, IDC derives share from IT spending. Hover any bar for cross-source context and per-row notes.

ROI and Productivity Evidence

Ranked table of studies including the GitHub Copilot RCT, Microsoft Copilot randomized trial, Klarna agent disclosure, Stanford / MIT call-center study (NBER), Mollick / BCG consulting study, McKinsey value report, and Goldman 7T paper. Methodology pills color-code rigor (RCT, quasi-experimental, observational, modeled, survey). Disconfirming studies are flagged.
RCTQuasi-exp.ObservationalModeledSurveyDisclosure
Sort:
MethodologyStudyDateHeadline finding
RCT

GitHub Copilot RCT (Peng, Kalliamvakou, Cihon, Demirer)

GitHub Research Working Paper, published Sep 2023

2023-09

Developers using Copilot completed an assigned coding task 55.8% faster than the control group (1 hour 11 min vs 2 hour 41 min), with no significant difference in self-reported task completion confidence.

RCT

Navigating the Jagged Technological Frontier (Wharton/Harvard/BCG, Mollick et al.)

Harvard Business School Working Paper 24-013; Boston Consulting Group experimental

2023-09

758 BCG consultants randomly assigned to GPT-4 or control completed 12.2% more tasks, 25.1% faster, and produced 40% higher-quality results on tasks 'inside the frontier' of AI capability. On tasks 'outside the frontier', AI use led to 19 percentage point lower correct-answer rates than control.

Quasi-exp.

Microsoft Research Copilot Field Experiment

Microsoft Research and external collaborators (Cambon, Hecht, Edelman, Ngwe, Jaffe, Heger, Vorvoreanu, Peng, Hofman, Farach, Bermejo-Cano, Knudsen, Bauman, Lee, Mishra)

2024

Random assignment of Microsoft 365 Copilot to information workers; pilot participants completed common tasks (drafting documents, summarizing emails, building presentations) approximately 27% faster on average; quality differences mixed.

Quasi-exp.

Generative AI at Work (Stanford and MIT call-center study, Brynjolfsson, Li, Raymond)

NBER Working Paper 31161

2023-04 (revised 2023-09; published in QJE 2025)

Access to a generative-AI conversational assistant increased contact-center agent productivity (issues resolved per hour) by an average of 14%. Novice and low-skilled workers gained 34%; experienced workers were largely unaffected. Customer sentiment improved and worker attrition fell.

Observational

Disconfirming

Apollo Global 'AI Capex vs Revenue Gap'

Apollo Chief Economist Torsten Slok

2025-08

Aggregate AI infrastructure capex by hyperscalers ($350B+ FY2025 plan) far exceeds aggregate AI revenue captured by the same firms (estimated $50B to $60B). Frames the gap as needing either step-function revenue acceleration or capex pullback.

Observational

Anthropic Economic Index (Claude usage and economic impact)

Anthropic Economic Index reports

2025-02 and 2025-09

Analysis of millions of anonymized Claude.ai conversations classified by O*NET occupation: AI use concentrates in software engineering, technical writing, business analysis, and data work. Approximately 36% of jobs use AI in at least 25% of their tasks (Sept 2025 update).

Observational

Goldman Sachs internal developer Copilot rollout (2024)

Goldman Sachs Engineering disclosures (Marco Argenti, CIO public statements; Bloomberg interview Jul 2024)

2024-07

Approximately 12,000 developers using GitHub Copilot at Goldman Sachs; CIO Marco Argenti cited 20% to 40% productivity uplift in internal pilot metrics.

Observational

Disconfirming

Goldman Sachs 'Gen AI: Too Much Spend, Too Little Benefit?' (featuring Jim Covello)

Goldman Sachs Top of Mind Issue 129

2024-06-25

Jim Covello (Goldman head of equity research) argues that AI capex (modeled at $1T+ over coming years) is unlikely to be recouped because gen AI use cases do not yet justify the cost; comparable transformative technologies (internet, smartphone) replaced expensive existing solutions with cheaper ones, whereas gen AI replaces cheap labor with expensive compute.

Modeled

McKinsey 'The economic potential of generative AI: The next productivity frontier'

McKinsey Global Institute

2023-06

Generative AI could add $2.6 trillion to $4.4 trillion in annual value across 63 use cases globally, or 15% to 40% on top of all non-generative AI and analytics impact.

Modeled

Goldman Sachs 'The Potentially Large Effects of Artificial Intelligence on Economic Growth' (Briggs and Kodnani)

Goldman Sachs Global Economics Analyst

2023-03-26

Widespread AI adoption could raise global GDP by 7% (approximately $7 trillion) over a 10-year period and lift productivity growth by 1.5 percentage points annually.

Survey

Disconfirming

MIT Sloan Management Review 'GenAI Divide: State of AI in Business 2025' (Project NANDA)

MIT Sloan and Project NANDA

2025-08

95% of organizations are getting zero return on their generative AI pilots. Only 5% of enterprise gen AI pilots have moved into production with measurable P&L impact.

Survey

Federal Reserve Bank of St. Louis 'The Impact of Generative AI on Work Productivity' (Bick, Blandin, Deming)

FRBSL Economic Synopses 2024-32

2024-10

Survey-based estimate that gen AI raised aggregate US labor productivity by approximately 1.1% from August 2024 levels, equivalent to a roughly $100B annual GDP uplift at then-current labor compensation.

Survey

Microsoft and LinkedIn 'Work Trend Index' Copilot productivity cut

Microsoft and LinkedIn 'Work Trend Index Annual Report 2024'

2024-05

75% of knowledge workers globally report using AI at work (May 2024 survey). 90% of Copilot users say it helps them save time; 85% say it helps them focus on the most important work. Microsoft's own randomized pilot found knowledge workers completed tasks 29% faster on average.

Disclosure

GitHub Octoverse 2024 (developer productivity disclosure)

GitHub Octoverse 2024 annual report

2024-10

GitHub Copilot had over 1.8M paid subscribers in October 2024; companies using Copilot reported up to 55% faster code completion in the GitHub-internal Productivity Survey wave fielded mid-2024.

Disclosure

Klarna AI Assistant (powered by OpenAI)

Klarna press release and Q1 2024 update; Sebastian Siemiatkowski podcast disclosures

2024-02

In its first month, Klarna's AI assistant handled 2.3M customer-service conversations (two thirds of total), performing the equivalent of 700 full-time agents, with comparable customer-satisfaction scores. Klarna estimated $40M USD profit improvement in 2024 from the deployment.

What the evidence says, overall

Per-task RCT productivity gains are real and large (typically 14% to 56%) for focused knowledge-work tasks.

Supporting evidence: GitHub Copilot RCT, NBER Brynjolfsson, Mollick BCG study, Microsoft field experiments.

Per-task gains do not translate cleanly into firm-level P and L gains. The majority of enterprise gen AI pilots have not delivered measurable bottom-line impact yet.

Supporting evidence: MIT Sloan 'GenAI Divide' 2025; Apollo 'capex vs revenue gap'; Bain 'beyond pilot' share approximately 35%.

Modeled long-run macro value is very large ($2.6T to $7T annual range) but is upper-bound, requires high diffusion, and has not been re-validated with realized data.

Supporting evidence: McKinsey 2023, Goldman Sachs Briggs and Kodnani 2023.

Skeptical view from the same firms (GS-Covello, Apollo-Slok) is now mainstream and pushes back on the bull case.

Supporting evidence: GS Top of Mind Issue 129; Apollo 'AI capex vs revenue gap'.

Confirming and disconfirming studies are both included. Methodology pill color reflects rigor: RCT and quasi-experimental at the top, modeled and survey-based at the bottom. Disconfirming studies are flagged with a rose tint. Click a row to expand context and limitations.

Token Consumption Growth

Monthly token volume served by major providers. Google has the most consistent disclosed series (9T per month in May 2024 stepping to 480T per month at I/O 2025). Microsoft Build keynote and OpenAI Dev Day disclosures captured. OpenAI and Anthropic mostly null.

Google: Tokens per Month, Across Products and APIs (log scale)

The only major provider that has disclosed a continuous per-month token series.

Other Providers: Limited or No Token Disclosure

Microsoft Azure AI Foundry

Quarterly only

not disclosed. Microsoft Build 2024 disclosed only Azure OpenAI Service customer count (53,000+ customers); did not name a token volume.

OpenAI API

Not disclosed

not disclosed. Microsoft Build 2024 disclosed only Azure OpenAI Service customer count (53,000+ customers); did not name a token volume.

Anthropic (API and Claude.ai)

Not disclosed

not publicly disclosed; Anthropic Economic Index reports usage by occupation rather than total token volume. Anthropic published the Anthropic Economic Index updates in Feb and Sep 2025, characterizing usage breakdowns by occupation across millions of conversations. Did not disclose tokens per month.

Step-function Launches

  • 2023-11OpenAI DevDay 2023; GPT-4 Turbo and Assistants API launch. Doubled OpenAI API throughput in following weeks per The Information reporting.
  • 2024-05Google I/O 2024; Gemini 1.5 Pro general availability. Triggered the inflection in Google product-AI token volume reported at I/O 2025 (50 times year-over-year growth).
  • 2024-09OpenAI o1 preview launch (Sep 12 2024). Reasoning models drove sharply higher per-query token consumption (10 times to 100 times more tokens per response). Reasoning-token disclosure became a separate billing line.
  • 2025-02Anthropic Claude 3.7 Sonnet (extended thinking) launch. Extended-thinking-token billing line added; per-query token volume rose materially.
  • 2025-05Google I/O 2025; 480T tokens/month disclosed. Canonical disclosure pegging Google's per-month consumption.

Only management-disclosed figures are shown. Google’s “tokens per month across products and APIs” is the cleanest continuous series; Microsoft discloses quarterly Azure AI Foundry totals; OpenAI and Anthropic have not published per-month token volumes. Log scale captures the roughly 50 times year-over-year growth Google disclosed between I/O 2024 and I/O 2025. Hover each point for the underlying disclosure.

AI App Active Users

Weekly and monthly active user disclosures for the top AI apps. ChatGPT is the dominant continuous series with explicit milestones (100M Nov 2023, 200M Q2 2024, 300M Aug 2024, 500M+ through 2025, 800M+ late 2025). Other apps surfaced as disclosure callouts because they do not publish continuous WAU.

ChatGPT Weekly Active Users (disclosed by OpenAI)

ChatGPT is the only AI app where the operator has disclosed a continuous WAU series. Gemini, Claude, Perplexity, and Grok numbers are not disclosed or not strictly comparable, see callouts below.

Other AI Apps: Disclosure Status (latest cut, 2025-Q4)

Gemini app

MAU only

Sundar Pichai at Alphabet Q3 2025 earnings (Oct 29 2025) cited Gemini app at approximately 500M MAU.

Claude.ai

Not disclosed

Not disclosed. Similarweb monthly visits to claude.ai approximately 280M Oct 2025.

Perplexity

Not disclosed

Approximately 40M+ MAU exit 2025 (est.).

Grok (xAI)

Not disclosed

Not separately disclosed.

Disclosed ChatGPT Milestones

  • 2023-11-06Sam Altman discloses 100M WAU at OpenAI DevDay 2023. (OpenAI DevDay 2023)
  • 2024-08-29OpenAI confirms 200M WAU to Axios (doubled year over year). (Axios, OpenAI statement)
  • 2024-12-04Sam Altman cites 300M WAU in pre-12-Days-of-OpenAI interview. (The Verge)
  • 2025-02-20Brad Lightcap announces 400M WAU. (CNBC)
  • 2025-04-14Sam Altman cites 500M+ WAU at TED 2025. (TED)
  • 2025-08-07Sam Altman cites 700M WAU at GPT-5 launch. (OpenAI GPT-5 launch)
  • 2025-10-06Sam Altman cites 800M WAU and 4M developers at DevDay 2025. (OpenAI DevDay 2025)

ChatGPT WAU is reproduced as disclosed by Sam Altman or Brad Lightcap at named events. Gemini app, Claude.ai, Perplexity, and Grok numbers are either not disclosed or reported in different units (MAU, queries, or cumulative accounts). Cross-product figures (Meta AI inside Family of Apps, Gemini inside Workspace and Search) are kept in the caveats panel to avoid an apples-to-oranges line on the chart.

Developer Survey Snapshots

Annual snapshots from the Stack Overflow Developer Survey, GitHub Octoverse, and JetBrains State of Developer Ecosystem. Percentage of developers using AI tools daily, top tool, and plan-to-use figures per year 2023 through 2025. Tool-mix shift chart tracks the Copilot to Cursor / Claude Code transition.

2023

First year of mainstream AI-coding-tool adoption. Survey signal was 'most developers have tried it, many are using it regularly, daily use is still a minority'.

Stack Overflow

Any AI use

70%

Daily AI use

n/d

Plan to use next year

77%

n =

89,184

Top tool

GitHub Copilot

55%

ChatGPT (in-IDE or browser)

First Stack Overflow Developer Survey wave to ask gen AI questions. 70% of professional developers reported they are using or planning to use AI tools (44%) plus 26% currently using. Daily use cut not published separately at this granularity.

GitHub Octoverse

Any AI use

n/d

Daily AI use

n/d

Plan to use next year

n/d

n =

platform-derived

Top tool

GitHub Copilot

Octoverse 2023 reported Copilot adoption metrics: 1M+ paying Copilot subscribers; 50,000+ organizations on Copilot Business. Developer-survey-style percentages not published.

JetBrains

Any AI use

77%

Daily AI use

n/d

Plan to use next year

n/d

n =

26,348

Top tool

ChatGPT

68%

GitHub Copilot

35%

JetBrains State of the Developer Ecosystem 2023 found ChatGPT was the most-used AI tool overall (browser-based) while GitHub Copilot led for in-IDE assistance.

2024

Cursor enters the survey universe for the first time at 8% (JetBrains). GitHub Copilot remains dominant by paid subscribers but gen-AI-native IDEs begin to surface.

Stack Overflow

Any AI use

76%

Daily AI use

n/d

Plan to use next year

81%

n =

65,437

Top tool

GitHub Copilot

41%

Tabnine

13%

76% of professional developers are using or plan to use AI tools, up from 70% prior year. 62% are currently using them. ChatGPT remained the most-used AI search/assist tool overall (82% used in past year); GitHub Copilot remained the top in-IDE coding assistant.

GitHub Octoverse

Any AI use

n/d

Daily AI use

n/d

Plan to use next year

n/d

n =

platform-derived

Top tool

GitHub Copilot

Octoverse 2024 disclosed 1.8M+ paid Copilot subscribers (Oct 2024); 77,000+ Copilot Business and Enterprise customers; AI Python overtook JavaScript on GitHub as the top language pushed (driven by AI/ML repos).

JetBrains

Any AI use

84%

Daily AI use

n/d

Plan to use next year

n/d

n =

23,262

Top tool

ChatGPT

70%

GitHub Copilot

41%

JetBrains 2024 found 84% of developers used AI assistants in the past 12 months (vs 77% in 2023). JetBrains AI Assistant (their own product) at 11% use. Cursor at 8% (first year tracked).

2025

The mix flipped. Cursor and Claude Code became co-leaders alongside Copilot in active in-IDE AI use; GitHub Copilot retained the largest paid-subscriber base because of enterprise lock-in. Windsurf entered the top five. Daily AI tool use crossed 50% for the first time per Stack Overflow.

Stack Overflow

Any AI use

84%

Daily AI use

51%

Plan to use next year

84%

n =

49,123

Top tool

Cursor

38%

GitHub Copilot

36%

Claude (Claude Code plus claude.ai)

29%

Stack Overflow 2025: 84% used AI tools in development (up from 76%). For the first time, daily AI tool use crossed half: 51% of pro devs used AI tools daily (the precise figure varies by SO cut and is in the 50% to 55% band). Cursor pulled ahead of GitHub Copilot among professional developers in the in-IDE category, with Cursor at 38% past-12-month use vs Copilot at 36%; Claude tools at 29%. Caveat: SO 2025 had a smaller respondent base than prior years.

GitHub Octoverse

Any AI use

n/d

Daily AI use

n/d

Plan to use next year

n/d

n =

platform-derived

Top tool

GitHub Copilot

Octoverse 2025 (Oct 28 2025): Copilot paid subscribers exceeded 5M (up from 1.8M); Copilot Pro+ tier launched mid-2025; new agent-mode Copilot features pushed average tokens-per-session sharply higher. Python remained the no. 1 language on GitHub; TypeScript moved up.

JetBrains

Any AI use

91%

Daily AI use

n/d

Plan to use next year

n/d

n =

23,000

Top tool

ChatGPT (browser plus in-IDE)

73%

GitHub Copilot

43%

Claude Code

25%

JetBrains 2025: AI tool use approaching saturation at 91% any-use. Tool mix shifted sharply: Cursor jumped from 8% to 24%, Claude Code (released Feb 2025) reached 25%, Windsurf entered at 12%. GitHub Copilot share of pro use rose modestly (41% to 43%) but lost relative position to Cursor and Claude Code among heavy users.

Tool-mix Shift, 2023 to 2025

Percent of professional developers reporting use of each AI coding tool. JetBrains is the consistent year-over-year series (JB). Stack Overflow 2025 (SO) added for the latest snapshot. Hover bars for the trend narrative.

Headline Summary, 2023 to 2025

Professional developers using AI tools in development

2023

70%

2024

76%

2025

84%

Stack Overflow Developer Survey 2023 to 2025

Plan to use AI tools next year (sum of currently using plus plan to use)

2023

77%

2024

81%

2025

84%

Stack Overflow Developer Survey 2023 to 2025

Daily AI tool use (first year reported by Stack Overflow)

2023

n/d

2024

n/d

2025

51%

Stack Overflow Developer Survey 2025

JetBrains: developers using AI assistants in past 12 months

2023

77%

2024

84%

2025

91%

JetBrains State of the Developer Ecosystem 2023 to 2025

GitHub Copilot paid subscribers (Octoverse annual disclosure)

2023

1M

2024

1.8M

2025

5M

GitHub Octoverse 2023 to 2025

Three sources are reproduced separately because their populations and methodologies differ: Stack Overflow surveys ~50,000 to 90,000 self-selecting devs, GitHub Octoverse is platform-derived, JetBrains surveys ~23,000 to 26,000 balanced respondents. Any-use rates cluster in the 70 to 91 percent band; daily-use crossed 51 percent for the first time per Stack Overflow 2025. Cursor and Claude Code rose sharply in 2025 while GitHub Copilot kept the largest paid-seat base. Click a year-pill in the tool-mix chart to toggle the year.

Ask Sterling

Register for a premium account to gain access to Sterling AI.

Get Started

Things you can ask Sterling:

Summarize Tesla's latest earnings reportWhy did NVIDIA's margins expand?Compare Apple vs Microsoft's cash flowWhat's driving the EV sector growth?
AI Software - Adoption | Sterling