AI Pulse · Under the Radar

Three X accounts are passing around the same line from what they claim was an Anthropic internal workshop: “Nobody is prompting agents anymore, we’re building graphs and loops.” The same accounts are predicting that within three to six months, “graph engineering” will replace prompt engineering as the standard way teams work with AI. No recording or transcript has surfaced, so this is still secondhand gossip, not confirmed fact.

The idea builds on Boris Cherny’s earlier framing, “I write loops, not prompts.” The shift is from asking an AI to do something once to building systems where agents call each other, check their own work, and route tasks based on conditions. Instead of crafting the perfect prompt, you map out a workflow with decision points, fallback paths, and validation steps. The AI becomes one node in a process you control, not the whole process.

Why it matters if it’s real

If this is how Anthropic’s own teams are working, it signals that the industry is moving past single-shot prompting faster than most businesses realize. Writing a good prompt still matters, but it won’t be enough if your competitors are building systems that loop, check, and adapt without human intervention every time. That kind of architecture is what we build into the Omni Command Centre, where agents handle repeatable workflows and only escalate edge cases.

The timeline, three to six months, feels aggressive. Most companies are still figuring out how to get one agent to work reliably. But the direction is clear. If you’re still thinking about AI as a better search bar or a writing assistant, you’re already behind the curve that’s forming. Watch for primary sources. If Anthropic confirms this internally, expect the rest of the field to follow fast.