ChatGPT Work Was Just Launched. Try These 11 Use Cases

0:00 Everyone keeps telling you that Chad GPT got better with Chad GPT work, but nobody tells you what to do with it. So, here are 11 jobs that you can hand it this week. The real ones, I do them

0:10 myself on my own files. They get harder to believe as they go. And number 11 is probably the one that I would show someone if I only had 30 seconds. Starting with a file that I've been

0:19 avoiding for 6 months. And no, that's not because it's boring. It's because I don't believe it. Now, who is this for? I think it's actually for anyone who pulls a report out of something,

0:29 anything. a booking system, your accounts, a survey tool. It arrives as a spreadsheet and you have never fully trusted a number in it. So you can go into chat GPT work and ask it open the

0:39 sales export in this folder. Rebuild it properly. Give me a clean tab, a separate tab listing everything that was wrong and what you did about it and add charts for the three things that

0:50 actually make a difference. Format it like a finished report. Don't touch the original. Of course, you could always upload a spreadsheet and ask questions about it. What you couldn't do is this.

1:00 It opens the file where the file already lives. It rebuilds it and saves it back with new tabs in it. No upload, no download. That's the spreadsheet connector. And it's the reason this is a

1:10 job and not a conversation. And as you can see, you've got clean tabs, charts, allin-one. And if we take a look, loads and loads of duplicates, loads of rows where the date format changes halfway

1:22 down, two columns that disagree with each other, everyone listed, and everyone already fixed. You go in with a file that you avoid and you don't trust, and you come out with a file that you

1:33 could forward to anyone, even a client. That's a file where the problem was hiding in the numbers. The next one, the problem is something that is not there at all. Now, before you point this at

1:43 anything of yours, there are three settings. I call them the three locks, and they'll take a minute. Okay. Number one, permission mode. It will ask before it does anything outside the folder that

1:54 you gave it. That's the default, and in the beginning, I suggest you leave it as is. Number two, folder scope. I suggest you limit to one folder, not your drive, not your entire documents folder, just

2:04 one playground folder to begin with. And number three, work on a copy. I call that the backup insurance. Okay, this is the one that a lot of people don't think about, but files it deletes cannot be

2:16 brought back as you do from the recycle bin. So, be mindful of that. You want to have a backup insurance. I have made an entire video on setting up guardrails properly. It's built around claude and

2:28 every principle in it actually also applies here without changing anything. So, I'm going to make sure to link it up there and in the description down below. And if you do want one built

2:37 specifically for Chad GPT work, just let me know in the comments and I'm going to make sure to make one. Let's come back to the second use case. The problem with this agreement is a clause that isn't in

2:46 it. So, who could use this? Honestly, anyone who signs things, they just skim through a lease, a supplier terms, a renewal that turned up by email. If you have any type of circumstance where you

2:59 come across contracts in your work, this one's for you. And what you do is you go into chat GPT work and say review this agreement. Mark it up with tracked changes and put a comment in the margin

3:10 explaining each one. Flag anything a document like this normally contains that this one doesn't. And then draft my reply to them. Firm but not aggressive. Here is who I am in this contract. And

3:20 then you need to explain what your position is. Now why is this new? Well, pasting a contract into old chat GPT and telling it what you want to change, that was possible. But before it wouldn't

3:34 make the changes, it would just tell you the changes and then you would go and change it yourself. This one opens the document, makes the changes tracked, so you can see every single one and throw

3:43 out the ones that you disagree with. And it moves everything in the document that you provided on the drive or in the folder that you selected to have the file or the document or the agreement

3:54 saved in. So basically, it moved from being your adviser to your in-house lawyer. So let's take a look. That's track changes in the actual file. Those are real comments explaining why. And

4:05 this is the section that for me was worth the entire exercise. You can take a look at the comments regarding the cap on liability notice period. There is no way that you can miss important aspects

4:17 in the clauses of a contract if you go about it this way. And if you would just skim, you probably would have missed it because there are absences from the contract. So you might have not thought

4:27 about them. Now there's one very important thing that I want to be clear about. It does read your document. It does make changes in it but it is not your legal advisor and you should not

4:38 trust it with important decisions. The day you do that it might be a day it becomes very expensive. So take it with a grain of salt. It does not give you legal advice but it can be a good way to

4:49 anticipate things that you might then need to bring to your legal adviser. Now both of those were sitting on my hard drive. Now let's take a look at something that is not. I have a call in

4:59 20 minutes with someone that I have met once. Take a look at what I know by the time the meeting starts. Now, who could use this? Literally anyone who has joined a call and thought, "Remind me

5:11 who that person was again." You can have chat GPT help you with that. Now, you can say, "I am meeting Sarah at Northridge in 20 minutes." Or if you have your calendar connected, you can

5:22 say, "I have a meeting with this person later today." and they can look at your calendar and find the information about the person you're meeting with. Please look up the person and the company.

5:32 Well, everything I have with them out of my files and calendar, one page, who they are, what's changed at that company recently, what I last promised them, and five questions worth asking. Now, you

5:43 might be wondering why this one is new. Search could always look up somebody or company. What it could not do is look them up on the web and read your calendar and go through your own folders

5:54 all inside one job and hand back one single summary page. Those are two different abilities that used to be two different conversations. And then you get the page and you can see where they

6:04 worked before, what their company announced 3 weeks ago. One line that might be very very significant for your meeting with them later today. For example, in my case, it says in March

6:14 you told her you'd send pricing for the second site. There is no record of you sending it. I could have completely forgotten that. 20 minutes of nothing and I walk in looking like I did three

6:24 hours of homework. Now, that's before the meeting. This next one is what happens the second the meeting ends. Let's say I finished a call at 11:00. By 4 minutes past, four people had been

6:36 told what they owe me. I was making coffee. If you're wondering who can use this, literally anyone who works with even one other person, especially if they've watched things get agreed in a

6:46 meeting and then never happen. Do you know anyone like that? You can go into JAGP work and say, "Take the transcript of this morning's call from my recorder. Build an action register, every

6:55 commitment, who made it, when it's due, flag anything agreed with no answer, and then post it to the team channel in Slack, for example, and tag every person that has an action item to their name."

7:04 Now, why is this a new one? Well, this one job connects to two different tools. It reads out of my recorder and it writes into Slack and I'm not standing in the middle passing things between

7:16 them. Obviously, that's chat GPT work now. It is no longer just reporting on my tools, but it is able to work in between them. And if you're thinking what we get back, well, it's not a

7:29 summary. You literally already get your summaries and probably nobody reads them. This time you get a very useful column with the action items and the person who owns them because otherwise

7:42 no one owns anything. And if you look in Slack, there it is in the channel with everybody's name against their own line and no one has to be chased and no one has to be reminded. You can go and make

7:54 your own coffee. So far, it's been things that already existed. Now, what should make something out of a total mess? Somewhere in these 41 emails is what we agreed. And I've been putting

8:06 this off since Tuesday. And if you're wondering who could use this, honestly, anyone who has said, "I'll write that up and send it over." And then not done at all for a week or two. So what would you

8:18 do? You would go to chat GPT work and say, "Read the whole thread. Work out what was actually agreed and what was only floated. Write it up as a proposal in my template. scope, timeline, what's

8:29 excluded, pricing, placeholders. Separately, list everything we discussed and never pinned down. And this is new because in the past you used to do this by copying emails into a chat box. But

8:39 if you have 41 of them, you would do that 41 times and then copying the answer back out into a document. I mean, that's a lot of manual work that is really unnecessary. Now, neither of that

8:51 happens. It reads the thread where the thread lives and then writes the document directly into my folder. That's a proposal, notes towards one. And this is the one that saves you. Six things we

9:03 talked about and never settled. Who supplies the data? What happens if it slips past October? Every one of those is an argument that you're going to have in a few months. And now you get to have

9:13 it this week, I guess, or never. Now, that was one document out of a mess. Let me show you four. A lot of the time, I hear people running away from hiring. And it's not the choosing part that they

9:25 don't like. it's the paperwork around the choosing that they don't like. This is for anyone who's about to bring in their first person and maybe don't want to be writing the briefs for the type of

9:36 role that they're looking to hire into. So, you can go to Chad GBT work and say, "Here's the role I'm hiring for." One paragraph. Build me a scorecard of what good looks like, a structured interview

9:46 plan, a scoring rubric so I'm not going on a gut, and a 30-day onboarding plan for whoever gets it. In the past, this used to be four separate prompts. and then four separate copy paste and by the

9:57 fourth one it had forgotten the first be honest you've been there I'm sure you have the model underneath this holds a long job in its head now so there's one instruction with four finished files and

10:07 they agree with each other thankfully you put one paragraph in and you get four documents out that are aligned the rubric is the one that I have skipped and probably shouldn't have that's the

10:18 difference between hiring on evidence and hiring on whoever you like the most on that afternoon now One thing it wouldn't do, and it's right not to do it. It will not score candidates for

10:29 you, and it won't tell you who to hire. That decision needs to stay with you, especially if you run your own business. Now, we've gone through six, five to go, and this is where they start getting

10:39 harder to believe. Let me show you. I thought I knew the number one complaint about this, and I was wrong, and it was not close. This next one is for anyone deciding what to fix, what to buy, or

10:50 what to build next. If you run a big program or a big product, this one's for you. You could go into Chad GPT work and say, "Go and read the reviews for this product across every site you can find.

11:01 Rank the four things people complain about the most by how often. Real quotes under each one. Link every source." If you've tried to do this before, old research would read the first couple of

11:11 pages and then summarize them. This has an actual browser inside it with tabs. It opens one site, reads it, decides what's worth keeping, and then goes to the next one, and then the next, and

11:21 then the next. That is the difference between summarizing and actually doing proper research. Now, as you can see, it's not asking me for the reviews. It's going and finding them on its own. And

11:30 it found four complaints ranked by frequency, and I had the top one wrong. Now, every quote links back to where it came from, which makes a huge difference because it's wrong sometimes. And that

11:41 link is how you can catch it in a matter of seconds instead of in a meeting in front of people. That one I asked for. This next one I asked for once 3 weeks ago. Now, it has messaged me twice in 3

11:53 weeks and both times I needed to know. So, who is this one for? It's for anyone who keeps checking whether somebody else changed their prices, their offer or their pitch, your competitors. If you

12:04 have those, which if you're in business, you probably do, you can go into chat GPT work and say, "Check these three competitors pricing and homepage once a day. Say nothing unless something

12:14 changes." If it does, tell me what it was, what it is now, and what it means to me or my business. You're probably thinking, "Yeah, but schedule tasks are not new. Chad GPT has had those for a

12:25 while." What's new, however, is what a schedule task is allowed to do when it wakes up. It used to be able to remind you to go and check something. Now it can actually go out and do the check for

12:36 you. And it doesn't stop there. It can build the table itself. It goes to their sites, reads them, and then it goes quiet if nothing changed. So for days you would hear nothing, maybe even

12:47 weeks, and that's a good sign. And then something changed. They moved their mid tier from 90 to 120. And it went out, found that out, and then also worked out what that means for where we are. The

13:00 clever part is not the table itself. It's the silence. Two limits that are worth mentioning here. However, it cannot run more than once an hour. And it cannot fire off an event like you can

13:11 do with routines in cloud code. This is a patient watcher, okay? It's not an alarm. It runs on the clock. It does not run on an event or a trigger. Now, in a minute, I'm going to show you the last

13:22 three, and one of them can completely change how you work. But first, I have to tell you about the one that went wrong, let's say. Because if you skip this, you are going to feel it. So, this

13:33 one did exactly what I asked and not at all what I meant. So, I told it to tidy up a client folder and archive anything old. I meant the drafts. It understood old as anything not touched in 6 months,

13:46 which was most of the finished work as well. It didn't break. It didn't do anything wrong. It did precisely what I asked it to do completely and confidently and I did not notice for

13:56 about an hour. Now, here's what that taught me and it's the most useful thing in this video and why I talk about that insurance so much. The place to catch this is not the output, it's the plan.

14:07 It tells you what it's about to do before it actually does it. And I had stopped reading that part because the first 10 times it was always right. So, please read the plan every time,

14:18 especially once you've started to trust it. and always have a backup, which is exactly what makes number 11 work. But first, the one where I don't touch the keyboard at all. That one's fun. I am

14:28 not going to touch the keyboard for the next 90 seconds, and you can watch the mouse. This one is for anyone with a portal, a supplier form, or an application. That means 40 minutes of

14:38 copying between two windows. You can go into Chad GPT and say, "Open the supplier portal, sign in, and complete the registration form." Every answer is in the company documents in this folder.

14:47 Please stop before you submit and let me review or show me. This feature did not exist at all. Not a better version of something. It just was not there. It's called computer use and it means it can

14:59 see your screen and drive your mouse and your keyboard on your own machine. No version of Chad GPT could do this before July. And I want you to just watch this. It's basically reading my documents and

15:10 typing the answers in everything. the company number, certificate number, uploaded everything from the folder and it stopped before submitting because I asked it to. But please always include

15:22 that instruction. Never admit it ever. Now those are 40 minutes I will never need to spend again with probably all forms. Now that's it using my computer. These last two are it using my whole

15:36 month. So number 10, this one used to be my entire Sunday, an actual Sunday every month, sometimes every week. So this is for anyone who rebuilds the same document every month or more often than

15:48 that. A board update, client report, a summary that somebody expects. There are two ways to hand it an example, and almost nobody knows the difference. If you give it a reference file, that means

15:59 do something like this once. A template means this is the format, keep it every time. So, you can go ahead and say, "Use last month's report as a template. Same layout, same fonts, same section order.

16:12 Pull this month's numbers from the sheet in this folder and rebuild it." That word template is what's helping us do the work and it's new. You used to have to describe your format in a paragraph

16:23 and then hope. Now, the format is carried inside the file with your instructions attached to it. So, it holds every single month instead of drifting a little every time. So you can

16:34 see you get the same layout, the same order, just different month data. Can you tell which one I made? Now that is a document that got rebuilt while I was just watching. So here's what happens

16:45 when I stop watching. Number 11. This is the one that I would probably show somebody if I only had 30 seconds. I think it's fun and quite impressive. Who this is for? Literally everybody

16:57 watching. Every job so far, I told it what to do. This time I'm only telling it what I want to end up with. That's steps closer to becoming an agent. You can say, "Have everything ready for

17:08 Friday's meeting. The deck in our format, the numbers from last quarter, a one-page pre-eread for the people who won't read the deck, and the follow-up emails drafted so I can send them the

17:17 moment it ends. This is new because this is what chat GPT work actually is underneath all 11 of this. Old Chad GPT gave you a reply. This one gives you a plan. And it doesn't stop there. It

17:28 waits for you to approve it and then works for hours across different apps while you're not there. Everything else in this video was basically a little piece of that. But this is the whole

17:38 thing. And as you can see, it's asking things that I hadn't decided yet, which is the right order to find that out. And this is the bit that I told you about 10 minutes ago. This is where you catch it.

17:48 Not at the end, but here when you approve. And then I can just go. And when I'm back, you see the deck, the numbers, the pre-eread, the four emails, everything is written, waiting across

18:02 four different apps, and I wasn't even in the room for any of it. Now, one thing that I think is important to be aware of before you go and give this one a try, okay? It's the thing that I'd

18:12 want to know if I were you right now. There is a way that this can go wrong that is not the way that you're expecting. We went through 11 different jobs. 10 I would immediately hand over

18:22 today, this week. one I probably wouldn't. And you've watched all of them, including the one that went sideways. Now, here's the big takeaway from all of this. For 20 years, the

18:34 valuable skill was doing great work. The skill that's becoming valuable is knowing exactly which work to let go of and being right about it. If you want to try something, give number one a try,

18:45 the file you don't trust. You can point it at a copy and go and make yourself a coffee or a tea or whatever you prefer. I would do that as the first step. And before you connect it to anything that

18:56 matters, I think you need to go and watch this because in that video, I talked about all the ways that LLMs can go wrong. The file that doesn't come back, the permission that you gave

19:05 without noticing, the settings that will stop everything and will give you more control. So, I'm going to see you in this next video. But before you go, please like this video if you did. Be

19:16 sure to subscribe if you haven't done so. Share this with anyone in your circle of friends or family or co-workers who you think would benefit from learning a little bit more about

19:23 how to use JH GPT work. And in the meantime, thank you so so much for watching. And if you'd like to take this one step further and apply everything you've learned, you're always more than

19:32 welcome in our communities. We have a free community where you can go through challenges and you can learn so much hands-on together with thousands and thousands of other people. If you want

19:42 to start your own business, we also have the founders community and you get the full claude setup course in that community as well. So, I'm looking forward to seeing you in either or both

19:52 of them. And until next time, like I said, I'm going to see you in this video here. Bye.