The State of AI Report 2025: AI Moves Faster Than Management
Introduction
Artificial intelligence has crossed a threshold.In three years, it has moved from playground to platform, from experimentation to infrastructure. The State of AI Report 2025, authored by AI investor Nathan Benaich (Founder of Air Street Capital), Zeke Gillman, Nell Norman, and Ryan Tovcimak, confirms what many executives sense but rarely say: the world has adopted AI faster than it has adapted to it.
Enterprise adoption has surged from 5% in early 2023 to 44% by late 2025, while the average AI contract value climbed from $39,000 to $530,000, on pace to reach $1 million in 2026. The world’s largest firms are spending billions to embed AI in daily operations, yet the people, processes, and policies built around it still belong to an earlier era of management.
1. The Infrastructure Race Is Already Won
AI has become the invisible infrastructure of modern enterprise.The standard stack is now clear: API access to models from OpenAI, Anthropic, and Google, fine-tuned open-weights alternatives, and a fast-growing orchestration layer that routes prompts and optimizes costs.
“OpenAI seems to be consistently producing the most intelligent models at the frontier,” Nathan notes, but the frontier is narrowing. Chinese labs DeepSeek, Qwen, and Kimi now sit within a few points on reasoning and coding benchmarks, and Qwen powers more than 40% of all new model derivatives on Hugging Face. Open source is catching up fast.
At the edge of capability, “Physical AI” is emerging. Robots are learning “chain-of-action” reasoning—turning structured thought into movement. Meanwhile, Anthropic’s Model Context Protocol (MCP) has become “the USB-C of AI,” a single standard that connects models to tools and data across ChatGPT, Gemini, Claude, and Microsoft’s systems.
Executives no longer ask whether to use AI, but how to keep it compliant, reliable, and cost-efficient. AI has joined electricity and cloud computing as essential infrastructure.
2. Capabilities Are Fragmented Across Organizations
Adoption is widespread, but capability is uneven.The State of AI Report 2025 survey of 1,183 practitioners shows that 95% use AI at work or home, and 76% pay out of pocket for subscriptions. Productivity gains are near universal—92% report being more productive—but these gains are mostly bottom-up. Employees are building their own automations and workflows, often invisible to management.
Nathan calls this “personal mastery outpacing corporate maturity.” AI has crossed the chasm for individuals, not for institutions.
The top barriers to scaling—time, data privacy, lack of expertise, and integration costs—all point to a gap between capability and governance. Security and compliance teams cannot monitor usage they do not see, and most organizations lack the infrastructure to coordinate the chaos.
The result is a patchwork of brilliance: local innovation thriving inside global confusion. Some teams race ahead; others still resist. The tools are ready, but the structures of leadership and learning lag behind.
3. Fast Adoption Without Adaptation Creates Drag
Speed without structure has a cost.The report reveals a paradox: productivity is up, but performance is flat. 92% of users report higher efficiency, yet few companies can link those gains to measurable business outcomes.
The barrier is structural. Legacy systems move on quarterly cycles while models update weekly. Compliance checks arrive after deployment, and managers are left verifying metrics they barely understand. Nathan calls this “technological adolescence”—organizations trying to manage exponential systems with linear tools.
The economy worsens the gap. “Cheaper intelligence produces more demand,” Nathan writes. As compute costs fall, usage rises, driving a feedback loop known as the Jevons paradox: more efficiency creates more consumption, which drives further complexity. The faster the technology moves, the heavier the organizational drag.
Until leadership structures evolve to match the rhythm of the technology, acceleration will continue to create friction instead of flow.
4. The Winners Are Building New Systems of Work
The leaders of this era are not just using AI, they are rebuilding around it, and AI-first companies “still outrun everyone else.”
According to Stripe’s AI 100, these startups reach $5 million in annual recurring revenue 1.5 times faster than top SaaS peers from 2018, and those founded after 2022 scale 4.5 times faster than pre-2020 companies. The Lean AI Leaderboard highlights 44 firms with over $5 million ARR, fewer than 50 employees, and $2.5 million in revenue per person. Small teams, massive leverage.
But the same transformation is now happening inside established firms.
Duolingo’s AI-First Memo asked: What can we build now that was impossible before? The answer became Video Call with Lily, a one-on-one AI conversation partner. The company launched 100 new courses in a year, instituted AI Fridays, and achieved near-universal tool adoption. CTO Severin Hacker calls this “funding what was impossible before.” Furthermore, CEO Luis von Ahn still hires new graduates, arguing that “early talent and AI fluency are the real long-term advantage.”
At the other end of the spectrum, Walmart is scaling transformation through people. Partnering with OpenAI, it will offer AI certifications to all 2 million associates by 2026, part of a $1 billion training commitment tied to OpenAI’s goal of certifying 10 million Americans by 2030. CTO Suresh Kumar says, “When AI is no longer the domain of a few experts and becomes a language many can speak, everything changes.”
Both companies share one principle: AI transformation only succeeds when it scales through learning.
5. Strategy Is Moving From Experimentation to Orchestration
After two years of experiments, the next phase is orchestration.Paid adoption has reached 43.8% and AI systems are moving from pilots to platforms. Frameworks like LangGraph, LangChain, and CrewAI now form the foundation for coordinated intelligence. Inside enterprises, model registries, governance dashboards, and monitoring pipelines replace ad-hoc scripts. The agent ecosystem is maturing, even as protocols churn and compatibility breaks.
At the geopolitical level, nations are orchestrating too. The US AI Action Plan, part of the $500 billion Stargate initiative, marks a shift from export controls to an export-led “American AI Stack”—a bundle of compute, models, and governance templates for allies. Europe’s AI Act lags in implementation, while China races ahead with new chip capacity and state-aligned labs.
Meanwhile, AI safety funding remains a fraction of capability spending. 11 major US safety organizations will spend $133.4 million in 2025, less than Frontier Labs' burn in a single day. As Senator JD Vance put it, “The AI future is not going to be won by hand-wringing about safety.”
The Bottom Line: Building the Adaptive Organization
The age of AI infrastructure has arrived. The age of AI capability is just beginning. For leaders, this means transforming as deeply and rapidly as the technology itself.
Five steps to begin:
- Diagnose your reality. Map how AI is truly being used—officially and unofficially. If you’re not sure where to start, we can help you uncover your organization’s AI baseline.
- Enable from the top down. Train leadership, align governance, and connect experimentation to strategy through our Corporate AI Programs and Executive Boot Camps.
- Invest in continuous learning. Run executive boot camps and design AI-human workflows.
- Build community and knowledge flow. Create internal networks that share evolving playbooks through our PRO Membership.
- Embed rapid assessment. Use diagnostics to identify quick wins and scale responsibly.
Organizations that act now will turn adoption into an advantage. Those who hesitate will have powerful tools but no coherent way to use them. The technology has already adapted to us. The question is whether we can adapt to it.
Looking Ahead
The State of AI Report 2025 closes with ten bold predictions: from agent-driven retail sales and scientific breakthroughs by open-ended agents to AI neutrality as foreign policy and executive orders shaping state legislation.
How do you think your organization will respond when prediction becomes reality?
