Want to stay in the loop with the latest in AI? The XDA AI Insider newsletter drops weekly with deep dives, tool recommendations, and hands-on coverage you won’t find anywhere else on the site. Subscribe by modifying your newsletter preferences<\/a>!<\/p>\n Here’s what the average coding session turns you into<\/p>\n I was the loop<\/p>\n If you’ve been keeping up with the memes and TikTok surfacing lately, you’ve likely seen at least one post about someone walking around with their laptop open just so they can keep tabs on what their coding agent is doing. A lot of the time, it’s to prevent the screen from closing and Claude Code (or whatever agent they’re running) from pausing mid-task, waiting for a permission prompt that no one’s there to approve.<\/p>\n However, at other times, it’s to catch the agent the moment it starts heading in the wrong direction, before it burns twenty minutes and a pile of tokens confidently building the wrong thing. And while I haven’t reached the point where I’m walking around with my laptop open in my hand, I’ve found myself doing the course-correcting far too many times.<\/p>\n For every task I handed off, I’d hover over the terminal, watch it work, catch it the second it veered off course, and type some version of “no, not like that” until it got there. At other times, I would sit there waiting for it to finish a pass just so I could run tests myself, read what failed, paste the errors back, and send it in for another round. Given how terrible usage limits are nowadays, this back-and-forth would end up eating into my session limits significantly (despite being on the Max plan).<\/p>\n Trust me, I tried everything. I made sure each prompt was as detailed as I could make it, front-loading every constraint and edge case I could think of so it wouldn’t wander off in the first place. One of Anthropic’s engineers’ biggest tips has consistently been giving Claude a way to verify its own work, whether that’s through the Claude in Chrome extension or a test suite it can run against itself, so it can catch its own mistakes instead of waiting on me to catch them. Despite trying that, Claude Code verifying its own work and me verifying the output would always reap different results, and I would ultimately find myself in the same exact situation as before.<\/p>\n The \/goal command was the only thing that finally closed that gap<\/p>\n It runs the loop so you don’t have to<\/p>\n Anthropic rolled out the \/goal command back in May, and if you weren’t paying close attention, you might have just ignored it by labeling it as “yet another slash command” just like I did. However, it turned out to be the one thing that actually took me out of the loop I’d been stuck in. <\/p>\n Here’s what it does. Normally, Claude Code does one pass at your task, stops, and hands control back to you. With \/goal, you set a condition instead, and Claude keeps working until that condition is met or until you stop it manually. The condition isn’t checked by the model doing the work, though. A separate, smaller model reads the session transcript after each pass and decides one thing: whether the condition has been met. If it hasn’t, the loop continues. (I went into the full mechanics of how that evaluator works in a separate piece if you want more details<\/a>). <\/p>\n That checking step is exactly what I’d been doing manually all along. I was the one reading each result, deciding whether it was good enough, and typing “keep going” or “no, not like that” until it got there. The \/goal command took that off my plate entirely. I set the condition once, and Claude handled the back-and-forth on its own instead of stopping to wait on me every round. <\/p>\n So, what does this actually look like in practice? Anthropic lists a handful of use cases in its support documentation<\/a>, and they all share one trait \u2014 a finish line a second model can actually check. Migrating a module to a new API until every call site compiles and the tests pass. Splitting an oversized file into smaller modules until each one’s under a size budget. Working through a backlog of labeled issues until the queue is empty. <\/p>\n![\"developer](\"https:\/\/www.newsbeep.com\/il\/wp-content\/uploads\/2026\/06\/developer-coding-on-laptop-with-terminal-window-displaying-json-and-npm-commands.jpg\")
<\/p>\n
![\"laptop](\"https:\/\/www.newsbeep.com\/il\/wp-content\/uploads\/2026\/06\/laptop-screen-displaying-code-editor-with-terminal-window-showing-html-and-css-programming.jpg\"\/)
<\/p>\n