ChatGPT released agent mode 4 days ago. And you still don't when to use it (and not) Here's what it's good for - and what it sucks at 🧵:
GOOD USE CASE 1: Sending connection requests to people with personalized messages.
MEH USE CASE 1: Creating your final presentation. Tools like Genspark perform much better in head to head.
GOOD USE CASE 2: Updating your spreadsheet with some manual data pulled from a report. (You'll love this if you hate excel.)
BAD USE CASE 2: Anything related to calendars. These foil it completely. Images: @jonathanroomer
GOOD USE CASE 3: Browse a website and order the best things that you actually want from it.
USE CASE 3 CONTINUED: It can even find and order shoes in your size!
MEH USE CASE 3: Doing a presentation and analysis all in one.
GOOD USE CASE 4: Create a FIRE (Financial Independence, Retire Early) presentation for you based on your own financial situation.
GOOD USE CASE 5: Combing through your data (like market and strategy data stored in a drive) to produce the outline of your presentation.
BAD USE CASE 5: Doing things that take 5 minutes manually. Often, the agent will take longer than 5 minutes. Sometimes... even an hour.
AI Edge
AI Edge22.7. klo 12.30
My biggest disappointment with the new ChatGPT agent is simply the amount of time it takes to complete a task. If a 5-minute manual task takes an agent 1 hour, I'd rather just do the task myself. I think V2+ is where we will really start to get efficient agents being deployed.
GOOD USE CASE 6: Beat the first level of the ARG AGI 3 Challenge. (Designed by the ARC prize to measure AI system generalization and intelligence through skill-acquisition efficiency in novel, unseen environments)
Here's the launch video with a full summary of all its capabilities:
Benchmarks wise, it's a big step up from o3. On Humanity's last exam (considered one of the best evals out there), it scored a whopping 41.6% vs 20.3% for plan o3.
My take: 6/10 release ChatGPT Agent mode is a lot like Manus, but worse. You get to see the AI browsing websites. Which is a plus. But, sometimes, it can be 58 minutes for cupcakes. It's an upgraded operator, but it's not there yet. I'm excited for v2.
Finally - if you want to use ChatGPT Agent mode, you'll need access to a paid plan (pro, plus, or team accounts). Then, you find it under tools: (If you don't have access yet, sit tight. They're still rolling it out to all accounts.)
20,16K