AI Agents Move From Asking to Doing

Enterprises are accelerating their adoption of artificial intelligence, moving beyond simple question-answering to delegating complex tasks. This shift, detailed in two new OpenAI reports, highlights a growing divergence between companies that are deeply integrating AI and those that are not. The core of this evolution lies in the adoption of agentic AI, which allows AI systems to interact with tools, company data, and execute multi-step workflows.

According to OpenAI's findings, the gap between leading AI adopters, termed 'frontier firms,' and typical organizations is widening. Frontier firms, representing the top 10% of monthly AI users, now generate 8.3 times more output tokens per active user compared to average firms. This metric serves as a proxy for the depth and complexity of AI engagement, indicating that leading companies are not just using AI more frequently but are entrusting it with more substantial work.

The Rise of Agentic AI

The transition from AI as an assistant to AI as an executor is powered by advancements like OpenAI's Codex and ChatGPT Enterprise. These tools can now access company-specific context and integrate with existing business tools, such as CRMs or internal playbooks. Instead of simply asking an AI to draft a presentation, a user can now prompt an agent to gather information from disparate sources, draft the content, and prepare it for review. As of June, Codex accounted for 64% of combined Codex and ChatGPT output tokens among enterprise customers, signaling a significant move towards delegated, multi-step tasks.

The Widening Frontier Gap

The disparity in AI adoption is stark. Frontier firms are significantly more likely to utilize advanced AI capabilities. For instance, 21% of active users at these leading firms use Plugins weekly, compared to just 9% at typical firms. This suggests that companies at the forefront are actively integrating AI with external tools and internal systems to automate complex processes. StartupHub.ai data shows that while OpenAI itself scores a strong 84/100 in our proprietary index, the widespread adoption of specialized tools like Zapier (score 77/100) and UiPath (score 85/100) indicates a broader market trend towards AI-powered automation solutions.

This gap isn't confined to tech giants. The reports indicate that intensive AI use is prevalent across various industries and company sizes, not just within technology sectors. Companies that are more aggressive AI adopters also tend to exhibit stronger financial performance, holding more assets, employing more people, and investing more heavily in research and development, according to an analysis of U.S. public companies in the accompanying working paper.

Agentic Work Spreads Beyond Engineering

While software engineering was an early adopter of agentic AI, its adoption is rapidly expanding across knowledge-work functions. Since February, weekly active enterprise Codex users have seen explosive growth in fields like legal (108x increase), sales (41x increase), and recruiting (41x increase). Marketing departments also saw a 26x surge in usage. This broadens the impact of AI beyond technical roles, enabling professionals in diverse fields to automate tasks and accelerate project timelines.

Early-career employees are leading this charge. Contrary to some surveys that suggest higher AI usage among executives, OpenAI's data indicates that junior employees send more AI-generated messages per week. This suggests that newer entrants to the workforce may have a natural aptitude for integrating AI into their workflows, presenting an opportunity for leaders to identify and disseminate these effective practices across all organizational levels.

Bridging the Gap

Leaders aiming to narrow this frontier gap must focus on more than just providing access to AI models. The reports suggest a practical agenda: connecting AI agents to relevant company context and tools, establishing clear governance and permission structures, and facilitating the transformation of individual effective workflows into shared organizational practices. By enabling agents to securely access company data and execute tasks with appropriate oversight, enterprises can fully transition from AI as an information source to AI as a core operational engine.