OpenAI has tested its latest models, including GPT-5, Claude, and Gemini, on real-world tasks, revealing that despite the proliferation of artificial intelligence tools, their effectiveness remains inconsistent. A recent report highlighted that 95 percent of enterprise artificial intelligence projects have failed, underscoring the pressure on companies to justify their use of these technologies. OpenAI’s new evaluation system, GDPval, measures AI performance on 1,320 tasks across 44 occupations, primarily in knowledge work. This evaluation aims to bridge the gap between theoretical capabilities and actual workplace demands, indicating that while AI can complete tasks significantly faster and cheaper than human experts, it currently lacks the ability to handle the complexities of ongoing projects effectively. OpenAI’s findings suggest that while AI models are becoming increasingly competitive with human performance, they still face challenges in tasks that require contextual understanding and interaction.
Cloudflare has introduced a new feature that allows website owners to block Google’s AI Overviews search product without opting out of Google search entirely. This move addresses concerns from publishers, including Penske Media Corporation, which recently sued Google over traffic losses attributed to AI Overviews that utilize their content without directing users to their sites. Cloudflare’s CEO, Matthew Prince, explained that the feature simplifies updates to the robots.txt file, enabling publishers to specify that Google can index their content but not use it for AI Overviews. However, these instructions are not legally binding, and it’s unclear if Google will comply. If Google or other AI companies ignore these directives, it could lead to legal action from affected publishers.
Why do we care?
Here’s the kicker—95% of enterprise AI projects fail. Not stumble, not slow down—fail. OpenAI’s new tests show these models can crank out tasks fast and cheap, but they fall apart when it comes to real projects. That’s what your clients are up against. And while you’re trying to help them make sense of AI, Google’s out there hoovering up publisher content for its AI Overviews, cutting traffic to websites. Cloudflare’s offering a block button, but whether Google listens? Who knows. For MSPs, this boils down to two jobs: keep clients from throwing money at AI experiments that won’t stick, and help them rethink marketing when SEO isn’t enough.