How good is ChatGPT Agent Mode?
|
ChatGPT rolled out agent mode earlier this month. Basically, it uses your computer for you. Think of it as an AI-powered virtual assistant that browses the web on your behalf. I tried agent mode for the past few days to see if it could save me time. First, you’ll need a ChatGPT Plus subscription to access Agent mode. You’ll find it under tools in your chat interface. I started by asking agent mode to find job vacancies in my niche on LinkedIn. ChatGPT set up a virtual desktop and started interacting with LinkedIn directly. It was oddly entertaining to watch. Since ChatGPT knows I’m a content creator and writer, it tailored the job search using its memories of me. After about 5 minutes, it delivered a list of relevant positions: web content specialist, SEO content manager, sales content strategist. Each came with links, role summaries, and salary information. Was it faster than manually searching LinkedIn? Debatable. But it freed me up to work on other things while it ran in the background. Next, I asked it to help me buy creatine powder. I kept the prompt deliberately vague to test its capabilities. ChatGPT asked about preferred brands and price ranges, but when I said “just continue,” it spent 3 minutes researching suppliers in Ireland and the UK. It sourced 250 grams of creatine powder from a popular fitness site. The virtual browser felt clunky, though. I couldn’t imagine entering credit card details into something that slow and cumbersome. Then I asked agent mode to review my YouTube channel for LinkedIn post ideas. ChatGPT analyzed my content and generated 14 different ideas. But here’s the thing - it pulled ideas from my Medium publication and newsletter, not just my YouTube videos. I had to redirect it to focus specifically on my channel. The refined list was much better. Ideas like “going from zero to paid newsletters” and “what to do when AI fails” directly matched my video content. These weren’t groundbreaking topics, but they beat staring at a blank page, wondering what to post. I also asked it to research competing YouTube channels in my space. It identified channels focused on copywriting and AI, explained their positioning, and highlighted their growth metrics. Most were channels I already knew, but for competitor research, this could save significant time. The AI spreadsheet export was disappointing, though. Agent mode struggled with Google Sheets and created a sparse Excel file instead. The original table format was more useful. For my final test, I asked it to research triathlon races in Ireland. This worked well. ChatGPT spent 5 minutes compiling popular races, dates, locations, and highlights. It even mapped them to my Chicago Marathon training schedule. But when I asked it to check if specific races were sold out, things fell apart. It spent 10 minutes struggling with this niche task and basically just snapped a screenshot of the website rather than providing actionable information. My takeaway? Agent mode feels like a 1996 IBM computer - impressive but overwhelming. It’s fascinating to watch AI interact with the web in real-time, but everything takes ages. For basic research tasks, it’s genuinely helpful. For specific, nuanced requests, I’d be faster doing it myself. That said, these AI agents are only going to get more powerful. They’re going to dramatically change how we interact with the web. I’d wager typing will feel antiquated in a few years. Need help? Check out Prompt Writing Studio​ |