A way of building AI coding agents lasted about six weeks this summer, start to finish. Engineers gave the practice a name in early June, then declared the same practice finished online by mid-July.

Boris Cherny, the engineer who built Claude Code at Anthropic, said in early June he no longer writes prompts for the coding assistant himself. For about two years, using a coding assistant meant typing an instruction, reading the result, and typing the next instruction by hand, one exchange at a time. Cherny described stepping back entirely from the turn-by-turn habit. He writes loops instead: small systems sitting between him and Claude, sending the assistant new instructions on their own, checking results, and trying again until a task is finished. His job, he said, is to write the loops, not the individual prompts running inside them. Addy Osmani picked up the line, wrote about the shift, and gave the practice a name: loop engineering.

For the weeks following, the name held. Guides on loop engineering spread across social platforms and company newsletters. Engineers rebuilt their own versions of Cherny’s setup over weekends, comparing notes on what a self-prompting system handled well and where the same setup broke down.

Then, on July 18, Peter Steinberger, the creator of OpenClaw, posted a mostly joking question on X: “Are we still talking loops or did we shift to graphs yet?” A few hours later, Hamel Husain answered with an article carrying a blunt title of its own: “Loop Engineering Is Dead. Enter Graph Engineering.”

One short post was enough to flip the conversation. By the end of the weekend, the same online spaces filled with diagrams, guides, courses, and tool stacks, all explaining why loops were already outdated and graphs were the new way to build agents. Weeks earlier, those same corners of the internet had filled with almost identical material explaining why loops were the future. The pattern repeated at almost the same speed, and few people seemed to notice the same shift happening twice.

Some of this speed follows familiar incentives. Vendors want a new angle to sell, and commentators want a fresh post to write, pressures found in most cycles of online enthusiasm and not unique to AI. Neither habit fully explains why this particular argument moved in a matter of days.

The bigger change: testing an AI engineering claim now costs almost nothing. An engineer hears about a new looping method on a Friday and builds a version by Saturday night. By Sunday, the loop never stops, loses track of its own progress, or drifts from the original task. A failed loop no longer stays private on someone’s laptop. Thousands of engineers run something close to the same experiment within days of a new idea appearing. A shared limitation surfaces fast, in public, so the wider conversation notices together instead of one engineer noticing alone.

Excitement alone used to be enough to keep a new idea circulating for a while. People would call something promising, plan a test, and often never run the test at all, letting the enthusiasm carry on unchallenged. Now, because so many people test the same idea at once, the pace of a hype cycle tracks how fast a claim gets tried and broken. A name spreads across a whole professional community and collapses inside one long weekend, because the public test finishes before the public conversation about the claim even settles down.

Sometimes a fast rename only marks the moment when engineers scattered across the field discover the same limitation at once and move together toward the next attempt. Loop engineering and graph engineering carry those same marks.

Cherny’s original practice, a system prompting itself toward a goal, kept running under the new name the whole time. Only the label around the practice changed by mid-July, settled inside one weekend by two short posts and the wave of guides following them. Six weeks turned out to be enough time this summer for a new way of building AI agents to spread, get tested in public by thousands of engineers at once, and get renamed before most of those engineers had finished reading the first set of guides.

Sources

Addy Osmani, Loop Engineering — addyosmani.com
X, “Are we still talking loops or did we shift to graphs yet?” — x.com
X, “Loop Engineering Is Dead. Enter Graph Engineering” — x.com
AI Advances, Two Engineers Made a Joke About Graph Engineering. Six Days Later It Had Courses. — ai.gopubby.com