AI in Corporations: Which Scenarios Are Already Working Right Now

Why Most AI Reports Are Lying

There is a lot of noise around corporate AI, but few real numbers. Most studies are based on surveys—companies simply report how they are “using AI.” The loudest of such studies came from MIT: it claimed that 95% of generative AI pilot projects fail.

Analysts at Andreessen Horowitz say this figure raises doubts. They gathered real data from leading AI startups and conversations with top executives at large corporations.

Real Implementation Numbers

29% of Fortune 500 companies and about 19% of Global 2000 companies are already live, paying customers of leading AI startups.

The statistics include only those who:
— signed an official corporate contract
— successfully completed a pilot project
— actually launched the product within the company.

Historically, startups had to spend years selling products to other startups before securing their first enterprise contract. AI has flipped this norm.

The starting point was November 2022—the launch of ChatGPT. It first showed corporations the real potential of AI and forced them to act faster than with any previous technology. The result: in just over three years, nearly a third of the Fortune 500 has real AI implementations.

Three Areas Where AI Actually Works

1. Code Writing — The Clear Leader

Coding is the dominant AI use case. This is evident from the explosive growth of Cursor, Claude Code, and Codex. Growth rates have surpassed even the most optimistic forecasts.

Why code specifically:

It can be tested: you run it and immediately know if it works or not. It is precise and unambiguous, with clear syntax rules. Top engineers increase productivity by 10–20 times with AI tools. The tool is useful even without 100% accuracy: error detection and template generation already save time.

2. Customer Support — The Ideal Task for AI Agents

Support is routine entry-level work that companies often outsource because they consider it too monotonous. This is exactly why AI has thrived here.

Most inquiries have a clear goal. For example, processing a refund—support has established standard procedures that AI agents can follow. ROI is easy to measure: number of tickets, customer satisfaction, percentage of issues resolved. If the AI makes a mistake, there is always escalation to a human, which reduces implementation risks.

3. Corporate Data Search

One of the biggest pain points for large companies is simply finding the right information among fragmented internal systems. Startup Glean became the main player for this scenario. In the legal sector, Harvey took off; in medicine, OpenEvidence.

Which Industries Are Leading the Way

Technology companies are, as expected, the first. They are almost always early adopters.

The legal industry is an unexpected pioneer. Traditional software barely accelerated lawyers’ work. AI changed this: it excels at parsing dense texts, summarizing documents, and drafting responses. Startup Harvey reached approximately $200 million in annual revenue within three years of its founding.

Healthcare is a market that traditional software never conquered. Companies like Abridge, Ambience Healthcare, and Tennr have grown rapidly on specific use cases: medical transcription, search, and automation of administrative processes—all without needing to replace existing systems.

Why Some Industries Adopt AI Quickly While Others Do Not

Industries actively adopting AI share several characteristics:

— work is text-based and repetitive
— results can be verified
— code that runs, a closed ticket—the human is involved in the process and can correct errors
— no strict regulation.

In areas where work involves the physical world, interpersonal relationships, or strict regulation, AI adoption is slower.

Models Are Improving Rapidly—And This Is Changing the Landscape

The most important takeaway: model capabilities are growing rapidly. Over the past four months, accounting and auditing capabilities have grown by nearly 20%, and investigative work by almost 30%. Labs are actively working on interface management, financial spreadsheets, and long-running tasks—this will unlock an entirely new class of work for automation.