Post by Gabriel Ceicoschi

If you connect recent AI news, you will see a very interesting pattern. OpenAI accepted a U.S. Department of War deal. Anthropic reportedly refused certain uses of its models. Basically, removing safeguards around mass domestic surveillance and fully autonomous weapons. And the internet noticed. Claude briefly overtook ChatGPT in app downloads during the controversy. Later, Microsoft made Copilot run on both OpenAI and Anthropic models, and Microsoft literally describes it as “model diverse by design.” Then Anthropic launched Claude Marketplace. Companies can buy Claude-powered tools from partners like Snowflake, GitLab, or Replit. Instead of buying SaaS tools and adding AI later, companies start buying AI services directly. 🎒 This aligns almost perfectly with Sequoia’s thesis of “Service as Software.” The next trillion-dollar company might look like a service company, but behind the scenes, it’s just a massive AI system delivering outcomes. -- Anthropic’s research shows AI usage is still concentrated in software development and writing. Meanwhile, industries like logistics, transportation, and physical services show very little AI penetration so far. -- And from above, a great opportunity for service companies is to pick up a tool, learn how to use it, and offer 10x the output. 🛟 And yet many employees get LinkedIn Learning access and are told: “Go learn AI.” I watched a few of those courses. Why does a database analyst need to understand diffusion models? Most of the content focuses on theory rather than teaching professionals how to build useful AI applications. The world is changing fast. But not fast for everyone. -- Curious when companies will start investing in hands-on AI education instead of hoping theory courses will do the job. Or maybe we’re waiting until an AI-first service company eats our markets before teams learn how to build agents. Meanwhile, Nizar Ntarouis and I are organizing a workshop in Tilburg to help young professionals actually build with AI. Read more at https://aibl.to