Companies Using AI Best Produce 8.3× More. Here Is Exactly What They Do Differently.

On 12 August 2026, OpenAI published two reports based on real data from over 10 million enterprise conversations. Not opinion surveys — administrative data from actual AI usage inside companies.

The central finding: companies in the top 10% by AI usage intensity — which the report calls “frontier firms” — generate 8.3 times as much output per active user as typical firms.

And the difference, the report says, does not come from having access to better models. The models are the same for them as for everyone else.

What Frontier Firms Do Differently

The report identifies a clear pattern: the widening gap appears alongside greater adoption of capabilities that connect agents to company context, tools, and repeatable workflows.

• Context. The AI agent knows how the company operates — what a customer means in their CRM, what a standard contract looks like, what the internal approval rules are. It is not explained from scratch every time.
• Tools. The agent can act directly in existing systems — read from ERP, write to CRM, send emails, generate documents. It does not produce text that a human then copies manually elsewhere.
• Repeatable workflows. The same processes are executed consistently, every time, without being reinvented on each run.

Companies that stay at the level of “an AI chat open in a browser tab” do not get this effect. Not because they are using AI incorrectly, but because they are using it without the three elements above.

Where Adoption Is Growing Fastest — and Why It Matters

Software engineering was the early centre of adoption. But since February 2026, the fastest growth in weekly active enterprise users has been in other departments: 108× in legal, 41× in sales, 41× in recruiting and 26× in marketing — compared with 5× in engineering.

Legal, sales and HR departments do not have in-house engineers. They do not build AI systems themselves. Which means this growth comes from adopting solutions built for them — not by them.

A concrete example from the report: at Virgin Atlantic, engineering teams refactor legacy code in 30 minutes instead of two weeks, and product teams complete in hours competitive research that used to take weeks — research that went on to shape the airline’s five-year digital strategy.

The Wider Context From This Week

Infrastructure is separating from models. Amazon Web Services entered a multiyear partnership enabling enterprises to deploy AI-powered application development entirely within their own private AWS environments, keeping corporate data inside customers’ infrastructure — a clear split between foundation models and the orchestration, governance and security layers around them.

Orchestration is becoming the product. Oracle updated AI Agent Studio with a unified builder for agentic applications — teams of specialised agents that reason, coordinate, decide and execute through business objects, workflows, policies, approvals and logged actions.

Large integrators are positioning around multi-agent. Cognizant announced a dedicated EMEA AI unit with services focused on strategy, rapid prototyping and multi-agent production teams. The market signal: enterprise RFIs increasingly assume multi-agent designs rather than single-model pilots.

Agents are becoming a security layer of their own. Obsidian Security raised $85 million in August at a $1.1 billion valuation, driven by demand for products that monitor AI agents interacting with business data. Nearly 70% of the company’s customers already allow AI agents to interact with business data.

The value is no longer in the model. It is in what you build around it.

What This Means for a Mid-Sized Company

The good news in this data is that the advantage does not depend on budget. Frontier firms do not have better models. They have the same models, correctly connected to their own context.

And “correctly connected” means exactly the things a company of 20 or 200 people can do: document how its processes work, define clear workflows and integrate AI into the systems it already has.

The 108× growth in legal departments or 41× in sales does not come from lawyers learning to code. It comes from someone building a system for them that understands the context of their work.

What We Build at Visual AI Labs

• Agents with your company’s own context — trained on your documents, processes and internal rules, not a generic prompt re-entered every time.
• Direct integrations into existing systems — CRM, ERP, email, cloud storage, internal platforms. The agent acts where your team already works.
• Orchestrated, repeatable workflows — processes defined once, executed consistently, with complete logging and human escalation where needed.
• Governance built in from the start — audit trails, architectural access limits and compliance with GDPR and AI Act requirements.

We deliver fast and without replacing the systems that already work for you.

Want to See What These Three Elements Would Look Like in Your Company?

Write to us on the contact page with a short description of a repetitive process in your company. We will come back with a concrete analysis of what could be automated and what impact it would have. No obligations — an honest assessment of your situation.

Write to us on the contact page →

Sources: OpenAI — Enterprise Signals and From assistance to execution (12 Aug. 2026); MarketingProfs AI Update (7 Aug. 2026); AI Agent Store — AI Agents News, week of 17 Aug. 2026; Orevia News — Enterprise AI News (17 Aug. 2026).