Critics call OpenAI's partnership with McKinsey and Accenture a bet on AI hype

In a syndicated column about OpenAI, the author makes a harsh thesis: the company's new corporate strategy looks less like a technological breakthrough and more like a model where value is sustained by a constant influx of new clients, partners, and implementation budgets. The catalyst was the Frontier Alliances program, which OpenAI announced on February 23, 2026 alongside McKinsey, BCG, Accenture, and Capgemini. According to OpenAI itself, Frontier is a platform for creating and launching AI coworkers—corporate agentic systems that should receive context from internal data, use tools, and perform real tasks.

The company argues that the market's problem is no longer model quality, but implementation complexity: businesses need integration, access management, process change, and support at the organizational level. This is precisely why the largest consulting and systems integration firms were brought in—to help customers move AI experiments from research into production. Meanwhile, OpenAI itself promises to expand platform access in phases, starting with a limited circle of corporate clients.

The article's author believes that in this construct, what matters is not the product outcome, but the ability to sell the transformation process itself. By his logic, consulting gets an ideal model: spending years accompanying implementation, repackaging problems as change management, optimization phases, and roadmap evolution. If AI coworkers fail to deliver on their promises, responsibility easily diffuses among the vendor, integrator, and the client itself.

In such a scheme, OpenAI gets legitimacy and a market signal that its solutions have been adopted by industry giants, while partners gain access to large budgets for process and IT infrastructure overhaul. A separate line of criticism concerns the actual capabilities of agentic AI. The author argues that marketing rhetoric around "digital employees" does not match the true state of the technology.

He cites the Remote Labor Index study published October 30, 2025 as evidence: the best AI agent tested could automate only 2.5% of complex remote work tasks. From this, the conclusion is drawn that large language models still perform poorly for reliable execution of long, multistep business processes where predictability, control, and repeatability matter.

In other words, demos and pilots look impressive, but when moving to critical operations, errors, hallucinations, and dependence on extensive manual guardrails surface. Another risk the author emphasizes is legal. In corporate contracts, responsibility for validating results, security, and consequences of AI use often falls on the customer.

If an agent makes mistakes in code, document management, client interactions, or internal approvals, disputes over liability can drag on for years. Against this backdrop, executives at companies that have invested tens or hundreds of millions in widespread AI coworker deployment risk not only budget overruns but also claims from boards of directors, shareholders, and regulators. Therefore, the alliance between OpenAI and McKinsey, Accenture, BCG, and Capgemini is described in the article not as a guarantee of product maturity, but as a mechanism to defer payment for overly bold promises.

The practical sense of this critique is that the corporate AI market is entering a phase where glossy demos alone no longer suffice. The louder the promises about autonomous AI coworkers, the more important it is for customers to ask about deployment metrics, error rates, boundaries of responsibility, and support costs. OpenAI's partnership with leading consultants may indeed accelerate the spread of agentic systems, but simultaneously raises the price of a possible error: if expectations do not align with actual technology capabilities, the blow will fall not only on the vendor but on the entire ecosystem of corporate consulting built around it.