Divya van Mahajan

The State of Enterprise AI

ai-daily-brief-podcast

The State of Enterprise AI

Overview

This episode of the AI Daily Brief (recorded around December 10, 2025) presents a dual-report analysis of enterprise AI adoption, drawing on:

  1. OpenAI’s State of Enterprise AI — based on usage data from enterprise customers and a survey of 9,000 workers across nearly 100 companies.
  2. Menlo Ventures’ Third Annual State of Generative AI in the Enterprise — based on a survey of 495 U.S. enterprise AI decision-makers conducted November 7–25, 2025.

The host argues that enterprise adoption data is now a critical signal in the broader “boom vs. bubble” debate surrounding AI investment, and that the data from both reports points strongly toward continued, accelerating growth.

Source video URL: not provided.

Prerequisites

  • Familiarity with ChatGPT and enterprise AI deployment concepts (custom GPTs, agents, copilots)
  • Basic understanding of the SaaS market and enterprise software procurement
  • Awareness of major AI model providers: OpenAI, Anthropic, Google, Meta
  • Understanding of reasoning models vs. standard LLMs
  • Familiarity with the concept of AI agents vs. copilots

Main Points

1. Enterprise AI Adoption Is Growing Rapidly

  • ChatGPT Enterprise seats increased 900% year over year
  • Weekly enterprise messages grew approximately 800% since November of the prior year
  • The average worker sends 30% more messages than a year ago
  • The median industry sector expanded 6x year over year; even the slowest-growing sector (education services) was up 2x
  • Fastest-growing sectors: Technology (11x), Healthcare (8x), Manufacturing (7x)

2. Depth of Usage Is Increasing, Not Just Breadth

  • While enterprise messages grew 8x overall, weekly users of custom GPTs and projects grew 19x
  • Custom GPTs and projects represent more contextual, repeatable, multi-step task workflows
  • Approximately 20% of all enterprise messages were processed via custom GPTs or projects in recent months
  • Reasoning token consumption per organization is up 320x in 12 months (from a near-zero baseline as reasoning models only recently came online)

3. Workers Are Reporting Tangible ROI

  • 75% of surveyed workers report improved speed or quality of output
  • Enterprise ChatGPT users save 40–60 minutes per day (data science, engineering, communication roles save 60–80 minutes)
  • 87% of IT workers report faster IT resolution
  • 85% of marketing and product users report faster campaign execution
  • 75% of HR professionals report improved employee engagement
  • 73% of engineers report faster code delivery
  • 75% of workers report being able to complete tasks they previously could not perform (e.g., spreadsheet automation, agent design, code review)

4. Coding Is the Clear First “Killer App”

  • Non-technical coding-related messages among enterprise users grew 36% over just six months — and the host suggests this is likely an undercount given that much vibe-coding happens outside ChatGPT
  • Menlo reports enterprises spent $4 billion on AI coding in 2025, representing 55% of total departmental AI spend
  • Individual coding categories: Code completion up 5.1x, AI app builders up 10x, code agents up 36.7x
  • The host’s framing: “2025 was the year of AI agents — it was just coding agents, not general agents”

5. The Gap Between Leaders and Laggards Is Compounding

  • The top 5% of individual users (“frontier workers”) generate 6x as many messages as the median worker
  • Frontier users send dramatically more messages across all task types: 8x creative media, 9x information-gathering, 10x analysis, 11x writing/communication, 17x coding
  • Frontier firms (95th percentile) generate 2x as many messages per seat overall and 7x as many to custom GPTs
  • High-usage individuals also report higher time savings: those saving 10+ hours/week use 8x more credits than those saving zero hours
  • OpenAI’s conclusion: frontier firms treat AI as “a core organizational capability rather than a peripheral productivity tool”

6. Anthropic Has Emerged as the Enterprise LLM Leader

  • Anthropic’s share of enterprise LLM spend: 12% (2023) → 24% (2024) → 40% (2025)
  • Google’s share rose from 7% to 21% over the same period
  • OpenAI’s share fell from 50% (2023) to 27% (2025)
  • Meta’s share fell from 16% to 8%
  • In AI coding specifically, Anthropic holds 54% market share vs. OpenAI’s 21% and Google’s 11%
  • This dynamic is cited as a key reason for OpenAI’s aggressive focus on coding capabilities (“Code Red”)

7. The Application Layer Is Now Larger Than Infrastructure

  • Total enterprise generative AI spend in 2025: $37 billion (~6% of the $300B global SaaS market, just 3 years post-ChatGPT launch)
  • Application layer spend ($19B) now exceeds infrastructure spend ($18B)
  • Horizontal AI up 5.3x, departmental AI up 4.1x, vertical AI up 2.9x
  • Startups captured ~$2 for every $1 earned by incumbents in the application layer (63% of the market overall), reversing a 2024 preference for incumbents (64%)
  • Examples: Cursor outperforming GitHub Copilot; Clay and Actively outperforming legacy sales tools

8. Buy vs. Build Has Swung Back Toward Buying

  • 2023: 80% buy / 20% build
  • 2024: 53% buy / 47% build (enterprises grew confident and application-layer startups were immature)
  • 2025: 76% buy / 24% build — boomeranging back toward buying as specialized application-layer startups matured
  • However, the build/buy distinction is blurrier than ever: even “bought” solutions require significant internal integration, data wiring, permissioning, and governance work

9. Chinese and Open-Source Models Have Minimal Enterprise Traction

  • Open-source LLM usage fell from 19% in 2024 to 11% in 2025
  • Menlo attributes part of this to stagnation in Meta’s Llama models
  • Chinese open-source models account for only 10% of enterprise open-source usage, or roughly 1% of total enterprise LLM API usage
  • The host does not expect this to change significantly given compliance concerns and the availability of performant closed-source alternatives

10. Agents Are the Future, But Copilots Dominate the Present

  • Copilots represent 10x the enterprise spend of agents
  • Only 16% of enterprise deployments qualify as true agentic systems — and even those are relatively simple
  • Of agentic deployments: 39% are fixed-sequence workflows; only 8% are multi-agent systems
  • A “modern AI stack” for agents is described as still under active development
  • The host notes Anthropic separately released a report on enterprise agent-building in 2026, deferred to a future episode

Key Concepts

  • Custom GPTs / Projects: Configurable ChatGPT interfaces with custom instructions, actions, knowledge bases, and context, used for repeatable enterprise workflows
  • Frontier workers / firms: The 95th percentile of AI adoption intensity; used as a benchmark for the most advanced enterprise AI users
  • Reasoning tokens: Output tokens generated by reasoning models (e.g., o-series) that perform chain-of-thought processing, enabling more complex tasks
  • Copilot: An AI assistant that augments human work interactively, currently the dominant enterprise deployment mode
  • Agentic AI / Agents: AI systems that can autonomously take multi-step actions, access tools, and complete tasks with minimal human intervention
  • Model Context Protocol (MCP): An open standard developed by Anthropic for connecting AI models to external data sources and tools; now donated to the Agentic AI Foundation
  • Agentic AI Foundation (AAIF): A newly created directed fund under the Linux Foundation, co-founded by OpenAI, Anthropic, and Block, to steward open standards for agentic AI including MCP
  • Vibe coding: Informal term for non-technical users generating code via natural language prompts in AI tools
  • Buy vs. Build: Enterprise decision-making framework regarding whether to purchase AI software from vendors or develop it internally
  • Vertical AI: AI applications built for a specific industry or domain
  • Horizontal AI: AI applications designed for broad use across industries and functions
  • Departmental AI: AI tools scoped to a specific business function (e.g., HR, marketing, finance)

Summary

Both the OpenAI and Menlo Ventures enterprise AI reports converge on a consistent narrative: enterprise AI adoption in 2025 has been faster, deeper, and more value-generating than prior years, with no signs of bubble-like contraction. Adoption metrics — seats, messages, spending — are growing at multiples rather than percentages. Workers across functions are reporting measurable time savings and new capabilities. Coding has emerged unambiguously as the category’s first killer application, driving Anthropic’s rise to enterprise LLM leadership at the expense of OpenAI. The application layer has overtaken infrastructure in spend, and AI-native startups have outpaced incumbents despite the latter’s structural advantages. The buy/build balance has stabilized back toward purchasing, Chinese and open-source models remain marginal in enterprise settings, and while agentic AI generates enormous excitement, the overwhelming majority of enterprise deployments remain copilot-style assistants rather than autonomous agents. The most consequential structural finding across both reports is the compounding divergence between enterprise leaders and laggards: the organizations and individuals investing most deeply in AI are generating disproportionate returns, accelerating ahead of peers in a self-reinforcing cycle.