OpenAI Says Businesses Are Moving From AI Assistance to AI Execution

Artificial intelligence is changing the way companies work, but the biggest shift may not be the growing number of employees using AI. It is what companies are asking AI to do.

OpenAI says enterprise AI adoption is increasingly moving from simple assistance — answering questions, drafting text and summarising information — to execution, where AI agents use company data and software tools to carry out multi-step tasks. In a report published on August 12, 2026, OpenAI said organisations are increasingly connecting AI to their workflows, tools and internal context rather than using it only as a standalone assistant.

From asking AI to delegating work

The traditional use of generative AI is straightforward: an employee asks a question, AI provides an answer, and the employee performs the work. Agentic AI changes that sequence. An employee gives the system a goal and allows it to perform several steps using tools and business information.

Ask it to “prepare a sales presentation for this client” and an agent could gather information, retrieve previous proposals, analyse relevant data, create the presentation and prepare it for human review. OpenAI describes this transition as moving from assistance to execution.

Enterprise use of AI is becoming more agentic

OpenAI reports that, as of June 2026, Codex generated 64% of the combined output tokens produced by Codex and ChatGPT among enterprise customers. The company says this reflects a growing shift toward agentic workflows, which involve longer, multi-step tasks and therefore generate more output than simple question-and-answer interactions.

The number needs careful interpretation. Output-token volume measures AI usage and depth of activity, not financial productivity. A company producing more AI output is not automatically creating more business value; the more important question is whether those outputs are being turned into useful work.

The gap between AI leaders and other companies is growing

OpenAI defines its “frontier firms” as businesses in the top 10% of monthly AI usage. It says those companies now generate 8.3 times as many output tokens per active user as typical firms, up from 2.6 times in January. The competitive difference is no longer whether a company uses AI — almost every major business is experimenting — but how deeply AI is embedded into its workflows.

The technology is moving beyond engineering

Software development was one of the earliest areas where AI agents became widely useful, but OpenAI says agentic use is spreading quickly into other knowledge-work functions. Since February, weekly active enterprise Codex users reportedly grew:

• 108× in legal
• 41× in sales
• 41× in recruiting
• 26× in marketing
• compared with 5× in engineering

These figures come from OpenAI's own enterprise usage data, so they should not be treated as a complete measure of global AI adoption. They do point to a direction: AI agents are increasingly used for business processes that depend on information, documents and software systems.

AI agents need access to business systems

A chatbot operating in isolation has limited capabilities. An AI agent connected to the right systems can do much more — a sales agent reading a CRM, customer history and previous proposals before drafting a response, or a recruitment agent moving from CVs to an applicant database to interview scheduling.

OpenAI says advanced capabilities such as plugins and skills are significantly more common among frontier firms because these tools let agents connect to company context, data and actions. Among weekly active users, 21% at frontier firms use plugins compared with 9% at typical firms, and 19% use skills compared with 3%.

The lesson is that the value of AI increasingly depends on what the AI can access and what it is allowed to do.

The new AI stack inside businesses

A mature enterprise AI system looks less like a chatbot and more like a technology stack. At the centre is the model; around it sit company data, business applications, APIs, workflow tools, agent instructions, permissions, monitoring and human review. That creates an environment where AI operates within defined boundaries rather than simply generating text.

Access alone isn't enough

OpenAI says complementary investments in continuous employee learning, shared workflows, data infrastructure and governance help organisations move from experimentation to deeper use. Buying ChatGPT or another AI product is not an AI strategy; the technology needs to be connected to specific business problems and workflows.

Early-career employees are using AI more

OpenAI's data challenges assumptions about who uses AI most. Early-career employees send more AI messages than executives — 13 more messages per week six months after adoption. The company suggests organisations identify employees with strong AI habits and turn their workflows into repeatable processes for everyone else.

AI is starting to change who does what

In an analysis of more than 800,000 work-related ChatGPT messages, OpenAI found that 16.8% of work-related messages and 43.5% of occupation-specific messages involved tasks associated with another occupation. A salesperson can analyse data, a marketer can troubleshoot a website, and a small-business owner can draft a contract or run basic financial analysis.

What AI could do for a small business

Consider a small business with ten employees. Today the workflow might be: customer enquiry, employee responds, employee checks stock, employee prepares a quotation, employee updates records, employee follows up.

With AI and automation, much of that becomes: customer enquiry, AI understands the request, AI checks approved business data and inventory, AI prepares the quotation, an employee approves, AI follows up. The human stays where judgement matters, while repetitive coordination happens automatically.

OpenAI is building products around this shift

In February 2026, OpenAI introduced Frontier, a platform for building and managing enterprise AI agents across business functions. In July 2026, it introduced OpenAI Presence, focused on putting voice and chat agents into customer and internal workflows reliably enough for production use.

Governance becomes more important as agents gain access

There is a major difference between an AI that drafts an email and an AI that can send one — and a larger one between an AI that recommends a refund and one that can issue it. As systems gain the ability to act, companies need to define what the agent can access, what actions it can take, what requires approval, how actions are logged, how errors are detected and when a human must intervene.

What this means for Kenyan businesses

The shift is particularly relevant to Kenyan SMEs, many of which still run on a mix of WhatsApp, phone calls, email, spreadsheets, accounting software and manual administration. AI agents can increasingly act as the layer connecting these systems.

• WhatsApp enquiry → AI qualification → CRM → quotation → human approval → payment link → follow-up
• Email invoice → AI extraction → accounting system → verification → approval
• Website visitor → AI sales assistant → needs assessment → product recommendation → CRM lead → salesperson

Our take

The first phase of enterprise AI was dominated by copilots. The next phase is about delegation — employees giving AI a task and allowing an agent to perform multiple steps using company information and software. The advantage will go to organisations that identify valuable workflows, connect AI to the systems those workflows depend on, and establish enough governance to let agents operate safely.

For businesses in Kenya and across Africa, that is an opportunity. Companies don't need to build their own frontier model; they can use existing models and focus on applying AI to local business problems and measurable workflows. The question is shifting from “how can our employees use AI?” to “which work can we safely delegate to AI?”

Sources: OpenAI's August 12, 2026 enterprise research, Enterprise Signals: How Organizations Use AI: Evidence from ChatGPT, OpenAI's research on AI and work, and its 2026 enterprise-agent product announcements.