The artificial intelligence boom is following a predictable pattern. Last week, the New York Times drew a parallel to the .com era: then, companies like Sapient multiplied revenue from €950,000 to €503 million in eight years simply by helping enterprises go online. Today, the beneficiaries aren't the AI model makers—they're the consultancies packaging and implementing these tools for corporate clients.
IBM has already secured over $1 billion in generative AI sales through its Watson X platform, which helps enterprises build and maintain AI models. Accenture moved $300 million in AI project revenue last year alone. McKinsey, traditionally a strategy firm selling PDFs with ideas, now expects 40% of its annual revenue to come from generative AI work. KPMG projects $650 million in US AI project sales. The broader US management consulting industry is forecast to hit $400 billion this year—a 2% increase that breaks the post-pandemic contraction. These firms are organizing conferences, showcasing third-party models from OpenAI and Anthropic, and bundling AI with robotics projects, multiplying their leverage.
McDonald's tested this model in 100 US drive-thru locations, replacing human order-takers with generative AI chatbots. The logic was sound: taking orders is pure interface work, and English-language models are mature. But the execution failed. The bots confused orders between stations, entered hallucination loops that blocked correction, and generated viral TikTok videos of baskets containing nine ice creams and 200 nuggets. McDonald's paused the project, though it and IBM said they'd return with refined solutions. The real lesson: consulting firms will iterate endlessly, charging for each attempt, while the underlying technology matures at the vendor's pace.
Anthropic's new Claude 3.5 Sonnet model, released this week, outperforms OpenAI's GPT-4o on most benchmarks except mathematical reasoning and undergrad knowledge. The advantage isn't raw capability—the incremental gains from model to model are now marginal. It's the interface. Claude's Artifacts feature lets developers write and test code in real-time without leaving the chat window. The model maintains file context, editing existing code rather than regenerating from scratch, reducing compounding errors. For builders, this matters more than another percentage-point improvement in benchmark scores. Claude remains free up to conversation size limits, while Pro costs €20/month—parity with ChatGPT—making it a genuine competitive threat to OpenAI's default position.