Latest / AI Fire Daily / #534 Neil: ChatGPT Work Brings OpenAI Into The Claude Cowork Battle
Transcript
- 0:00Imagine waking up tomorrow. Beat. The house is
- 0:05just completely quiet. Best part of the day.
- 0:08Right. You grab your coffee. You sit down at
- 0:10your laptop. And before you've even touched the
- 0:12keyboard, your day is already entirely structured.
- 0:15Sounds like a dream, honestly. It does, right?
- 0:18Your AI has already read your emails. It parsed
- 0:21your schedule. It mapped out your exact priorities
- 0:23for the morning. The cognitive load of wondering
- 0:26where to start is just, it's gone. Yeah, totally
- 0:30eliminated. So welcome to the deep dive. Today,
- 0:32we are looking at what I think is a massive paradigm
- 0:35shift in how we actually work. Absolutely massive.
- 0:38We are dissecting chat GPT work. So this is OpenAI's
- 0:42direct answer to Claude Cowork. And it's a fundamental
- 0:45architectural shift. I mean, we're moving completely
- 0:47past simple chat windows here. Right. OpenAI
- 0:51is building an automated multi -step workspace
- 0:53now. They clearly do not want to stay on the
- 0:55sidelines anymore. Yeah, they really don't. But
- 0:58I have to admit something right up front here.
- 1:00Oh, boy. What's that? I still wrestle with prompt
- 1:02drift myself. Oh, yeah. Everybody does. I mean,
- 1:05I find myself explaining the exact same context
- 1:08of the AI every single morning, and it just gets
- 1:11exhausting. So repetitive. You end up wasting
- 1:13so much time just setting the stage for it to
- 1:16do its job. Yeah, and that friction is exactly
- 1:18why people abandon these tools. The setup cost,
- 1:22you know, it often kills the actual... utility.
- 1:24Exactly. But that is exactly what this new release
- 1:27solves. So today we're going to explore how to
- 1:30move from those single prompt chat bots to a
- 1:33fully automated workspace. Right. We're going
- 1:36to test a real morning briefing. We'll build
- 1:38permanent reusable structures called skills.
- 1:41Right. And then we put it all on autopilot with
- 1:43scheduled tasks. Exactly. But to really automate
- 1:45our day, we first have to understand the foundation.
- 1:48Like, we have to understand how this new workspace
- 1:50actually functions. Well, let's look at the old
- 1:52model first. Standard chat is purely reactive.
- 1:56Right. Waiting for you. Yeah. You ask a simple
- 1:59question. It gives a simple answer. And when
- 2:01you close the window, it just forgets everything.
- 2:03Blank slate. Exactly. But chat GPT work is totally
- 2:06different. It's an actual persistent workspace.
- 2:10It holds context across complex multi -step goals.
- 2:14And context just means the background memory
- 2:16the AI relies on. Precisely. And it gathers that
- 2:20context from a few very specific places. OK,
- 2:22like where? Well, first, you choose an underlying
- 2:25model. Then you open or create a dedicated project.
- 2:29you upload your core reference files to that
- 2:31project, you define your custom instructions,
- 2:34and finally, you directly connect external apps.
- 2:37Right, so a project is basically a walled garden.
- 2:40Yeah, that's a good way to put it. It keeps your
- 2:42related chats and reference files bundled together,
- 2:45so it stops the AI from wandering off topic.
- 2:47Yeah, it creates strict boundaries. It really
- 2:49anchors the AI's attention on the task at hand.
- 2:52It's kind of like giving a new assistant a permanent,
- 2:54dedicated desk. I like that analogy. You know,
- 2:57you have all your files already open for them
- 2:59on the desk. The sticky notes stay exactly where
- 3:01you left them yesterday. You aren't forcing them
- 3:04to relearn your entire business from scratch
- 3:06every morning. That is a great way to visualize
- 3:08it. because you literally set up this workspace
- 3:11before you send a single prompt. Right. And once
- 3:14the desk is set up, the outputs change dramatically.
- 3:17It doesn't just casually chat with you anymore.
- 3:19No, it generates actual complex deliverables.
- 3:22Exactly. It builds comprehensive briefings. It
- 3:25writes full multi -page documents. It structures
- 3:27messy data into clean spreadsheets. It can even
- 3:30build presentations and interactive sites. Yeah,
- 3:33and you can track its progress while it works.
- 3:35It operates much more like an agent now. You
- 3:38can answer its clarifying questions. You can
- 3:40pivot its direction midway through a task. You
- 3:43can even approve critical actions before it takes
- 3:45them. But here's the technical friction point
- 3:47for me. If it is managing these massive deliverables,
- 3:50how does it not lose the thread? What do you
- 3:52mean? Well, standard chat usually forgets the
- 3:55first page by the time you reach page 10. So
- 3:57how does it manage complex deliverables compared
- 4:00to standard chat? Ah, right. It comes back to
- 4:03that project structure. Because the project bundles
- 4:06your related chats and specific files together,
- 4:09it constantly grounds the AI. Oh, I see. Yeah,
- 4:12it's forced to reference those specific guardrails
- 4:15over and over. It prevents the model from pulling
- 4:18in generalized, irrelevant internet data. So
- 4:22it builds complete documents, not just isolated
- 4:25chat bubbles. Yeah. Beat. Precisely. It changes
- 4:28the entire dynamic of the interaction. You know,
- 4:30you're managing a process, not just firing off
- 4:33random queries into the void. OK, so the desk
- 4:36is set up. We understand the workspace. Now,
- 4:39let's look at a practical, high -value test case.
- 4:42The fun stuff. Yeah. We need to wrangle the daily
- 4:44morning chaos. And the guide we're looking at
- 4:47suggests building a morning briefing. Which is,
- 4:50honestly, the ultimate test for this architecture.
- 4:52Right. Because you're asking the AI to juggle
- 4:55multiple live data streams. Right. So you start
- 4:57by connecting Gmail and Google Calendar directly
- 4:59to the workspace. OK, but this is where I want
- 5:01to push back a little on these tools. OK. What's
- 5:04the sticking point for you? I just see a massive
- 5:06flaw in how most people use AI for emails. Which
- 5:10is? AI naturally wants to summarize absolutely
- 5:12everything it touches. Oh, yeah. Always. If you
- 5:15give it 50 emails, it proudly gives you a 50
- 5:18bullet point list. That isn't helpful. Not at
- 5:21all. A summarized messy inbox is still just a
- 5:24messy inbox. It is completely overwhelming. That
- 5:27is a very common trap. And it's exactly why the
- 5:30initial prompt structure is so critical here.
- 5:32Right. You do not ask it to summarize. You explicitly
- 5:36instruct it to filter. You have to be absolutely
- 5:39ruthless with your instructions. Yeah, the guide
- 5:41says you have to define strict priorities first.
- 5:44You don't just walk up and say, read my email.
- 5:46You tell it, my priorities today are content
- 5:48planning, team feedback, and urgent client deadlines.
- 5:51Right. You are giving it a specific lens to look
- 5:54through. You're kind of stacking Lego blocks
- 5:56of data. Yeah. First, you set the calendar context.
- 5:59Then you apply that priority lens to the inbox.
- 6:02You literally command it to ignore the noise.
- 6:04Exactly. You tell it to explicitly ignore newsletters,
- 6:07ignore promotional offers, ignore routine plans.
- 6:10If an item doesn't require my physical action
- 6:13today, do not include it. And it has to justify
- 6:16why an email made the final cut. For every single
- 6:21item it shows you, it must provide one clear
- 6:23next step. Ah, I love that. That's the secret
- 6:26sauce. Yeah. It forces the AI to act as a gatekeeper,
- 6:31not just a passive reader. OK, so after you lock
- 6:33in those instructions, you run the briefing prompt.
- 6:36You tell it to check the connected Gmail and
- 6:38calendar. You ask it to find important meetings
- 6:41and tight deadlines. And you also ask it to flag
- 6:45any scheduling conflicts or missing context for
- 6:48upcoming meetings. It essentially highlights
- 6:50operational risks you need to know about right
- 6:52now. That risk flagging is brilliant. It feels
- 6:55much more like a human project manager. Yeah,
- 6:58it really does. But let's be realistic for a
- 7:00second. What if the very first briefing it generates
- 7:02is still way too long? Well, that happens frequently
- 7:05on the first run. The AI is still kind of calibrating
- 7:08to your definition of urgent. Right. It doesn't
- 7:10know you yet. Exactly. So when that happens,
- 7:12you just use a refinement prompt. And you stay
- 7:14right in the same chat window for that. You just
- 7:17reply and say, refine this output. Remove anything
- 7:21that doesn't strictly need a reply today. Use
- 7:24a cleaner layout. Okay. And it recalculates based
- 7:27on the new friction. You just command it to strictly
- 7:29cut the fluff. Exactly. You iterate until the
- 7:32format perfectly matches how your own brain processes
- 7:35information. But this brings us to the next massive
- 7:39bottleneck. Okay, what's that? The morning briefing
- 7:41works perfectly today, but tomorrow morning,
- 7:44when I wake up, I have to type out that massive,
- 7:47carefully worded prompt all over again. Yeah.
- 7:49The friction is still there. Refining a prompt
- 7:52every single morning is tedious. Oh, it's totally
- 7:54unsustainable. Right. Nobody will stick with
- 7:57a tool if they have to write a paragraph of rules
- 7:58every single day. Right. You'd just give up.
- 8:00Exactly. So that is why we move to the next phase.
- 8:03Turn that perfect, refined prompt into a permit.
- 8:06tool. We automate the process with skills. And
- 8:09a skill is just a reusable set of instructions
- 8:11for a single type of task. Yes. You engineer
- 8:14the perfect prompt once, then you trigger it
- 8:17with a single click whenever you need it. It
- 8:19completely eliminates the daily setup phase.
- 8:22So how do we actually build one in the interface?
- 8:24Is it like a complex developer tool? Not at all.
- 8:28It's incredibly accessible. You just open the
- 8:30plugins menu on the left side bar. You switch
- 8:33to the skills tab. You click the plus button
- 8:35and you select create with chat. Honestly, this
- 8:38mechanism is wild to me. It opens a new conversation,
- 8:41but you aren't talking to the standard AI. Right.
- 8:45You are talking to a builder agent. The AI literally
- 8:48helps you build the constraint system. Yeah.
- 8:50You don't write any code yourself at all. Yeah.
- 8:53You just described the workflow in plain English.
- 8:55You say, create a reusable skill called morning
- 8:58briefing. You tell it to always use your connected
- 9:01email and calendar. And then you feed it the
- 9:04exact filtering rules we just perfected. Check
- 9:07for deadlines, find actionable emails, ignore
- 9:10the newsletters, rank the tasks by urgency. Exactly.
- 9:14You even tell it exactly how to format the output.
- 9:16Like, always use three sections. Today's schedule,
- 9:19top priorities, and risks. You're permanently
- 9:21locking in your preferences. And here is a crucial
- 9:24architectural detail. The AI actually takes your
- 9:27conversational description and translates it
- 9:30into code. It generates a file called skill .md
- 9:33in the background. Let's clarify that real quick.
- 9:36A .md file is just a simple text document storing
- 9:39your instructions. Right. But it acts as an immutable
- 9:42set of guardrails. It's not a fluid chat memory
- 9:45that changes. Ah, I see. It's a hard rulebook
- 9:48the AI must check against before executing the
- 9:50skill. Whoa. Imagine just typing one command,
- 9:53and it instantly reads your calendar, sorts your
- 9:56inbox, and plans your day. The amount of cognitive
- 9:59load that removes is just staggering. It really
- 10:01is. It functions like a digital chief of staff.
- 10:04You know, it reminds me of recording a macro
- 10:06on a complex spreadsheet. Oh, that's a good way
- 10:09to look at it. Yeah, you do a tedious series
- 10:10of clicks once, and you save it to a single button.
- 10:13But a spreadsheet macro only records Rigid clicks.
- 10:18Right. This records reasoning parameters. It's
- 10:20basically a macro for cognitive tasks across
- 10:23entirely different apps. That's a perfect distinction.
- 10:25It applies your logic, not just your clicks.
- 10:27Right. And when the builder finishes, it confirms
- 10:30the skill is installed. It gives you a direct
- 10:33trigger command. It usually starts with a dollar
- 10:35sign. Like dollar sign morning dash briefing.
- 10:38Exactly. Just type that into the chat and it
- 10:40runs the whole process. But wait, you can still
- 10:42guide it, right? Like you can append specific
- 10:44context for that specific day. Right. And that
- 10:47flexibility is key. You can type, use dollar
- 10:50sign morning briefing, but focus heavily on urgent
- 10:53content work today. Oh, nice. It adapts the daily
- 10:56output without breaking the underlying core skill.
- 10:59But let me ask a highly practical question here.
- 11:02Sure. Systems break. APIs fail all the time.
- 11:05What happens if it simply cannot access Gmail
- 11:08or Google Calendar some morning? Yeah. Does it
- 11:10just panic and start guessing my schedule? No.
- 11:13Thankfully, the system handles connection failures
- 11:15gracefully. The skill is designed to run regardless.
- 11:19Okay. But if a data source is unreachable, it
- 11:21will explicitly mark those details as unverified.
- 11:24It tells you it couldn't connect. It flags missing
- 11:26data instead of hallucinating. Exactly. Hallucinating
- 11:30just means the AI invents false facts when it
- 11:34lacks real data. Right, making things up. Yeah,
- 11:36so by forcing it to flag missing sources you
- 11:39eliminate that risk entirely. You just review
- 11:42the first few automated briefings to verify the
- 11:45connections are actually stable. You check the
- 11:47dates. You check the meeting times. You make
- 11:49sure the emails it marked as urgent actually
- 11:51are urgent. And if it consistently misses something,
- 11:55you just update the skill. You tell the chat,
- 11:57update the morning briefing skill to limit the
- 11:59list to five items. And it rewrites the underlying
- 12:02.md file in the background. You never have to
- 12:05start from scratch. Which is huge. Once that
- 12:07skill proves reliable, you have an incredibly
- 12:10powerful tool. But we still haven't reached full
- 12:13automation yet. Not quite. We still have to manually
- 12:16type the trigger command every morning, so let's
- 12:18put this entirely on autopilot. This is where
- 12:21we cross the final threshold. We make the tool
- 12:23run while you sleep. We use scheduled tasks.
- 12:27OK, so you locate the scheduled feature on the
- 12:29left sidebar. It's basically a dedicated management
- 12:31dashboard for repeated actions. Yeah, you click
- 12:33schedule a task. It asks for the prompt you want
- 12:36to automate, and you simply enter your trigger
- 12:39command. Use dollar sign morning briefing to
- 12:42prepare my day. then you establish the cadence,
- 12:45you set the exact time and frequency, like run
- 12:48this every weekday morning at 8 a .m. local time.
- 12:52And you really have to double check your time
- 12:54zone settings here. Oh yeah. Servers usually
- 12:56default to universal time. You definitely don't
- 13:00want your morning briefing triggering at three
- 13:02in the morning and summarizing the wrong day's
- 13:04emails. That would be a mess. But once you confirm
- 13:07the schedule, you just wait. You open the workspace
- 13:09the next morning, and the result is already sitting
- 13:12there waiting for you. Yep. You review the prioritized
- 13:15emails. You verify the calendar sync worked.
- 13:18And this right here is the fundamental shift
- 13:20in the technology. I was just thinking about
- 13:22that. This totally flips the dynamic. It really
- 13:25does. It shifts the AI from being a reactive
- 13:28tool waiting for your input to a proactive partner.
- 13:31It has the heavy lifting done before you even
- 13:33sit down at your desk. You transition from generating
- 13:36work to reviewing work. It changes your entire
- 13:39morning posture. But we do have to talk about
- 13:43maintenance. Workflows naturally drift over time.
- 13:47Your daily priorities are eventually going to
- 13:50change. So what happens then? If the daily automated
- 13:53output starts getting sloppy or my goals change,
- 13:56do I have to tear down and rebuild the entire
- 13:58skill from scratch? No, you don't touch the foundational
- 14:01skill at all. The underlying logic of how to
- 14:04read an email remains exactly the same. OK. If
- 14:07the output drifts, you just adjust the scheduled
- 14:09prompt window. You add stricter daily rules to
- 14:12the trigger itself. Like adding keep only the
- 14:15top five items and put client deadlines first.
- 14:17Exactly. Update the schedule prompt. Keep the
- 14:19underlying skill intact. precisely you tweak
- 14:23the daily trigger mechanism but you leave the
- 14:25permanent tool alone it keeps your overall workspace
- 14:28incredibly clean and easy to manage over the
- 14:31long term we're going to take a quick break for
- 14:33our sponsors and when we come back we'll look
- 14:35at the broader implications we'll answer the
- 14:37final question is chat gpt work actually worth
- 14:41overhauling your entire routine for We'll be
- 14:43right back. All right, we are back. All right,
- 14:46let's pull all of these threads together. We've
- 14:48mapped out the shift to the workspace. We tested
- 14:51the morning briefing filtering. We built reusable
- 14:53skills. And we put the whole thing on autopilot
- 14:56with scheduled tasks. So the source material
- 14:59ends by addressing a critical question. Is chat
- 15:02GPT work actually worth using? The guide draws
- 15:06a very clear, practical dividing line here. What
- 15:08is it? It all depends on your current friction
- 15:11points. If you only use AI for isolated questions,
- 15:15like writing a quick email or checking a line
- 15:17of code, you don't need this. just stick to the
- 15:20standard chat. The standard chat is perfectly
- 15:22fine for quick, stateless queries. Exactly. But
- 15:25if your daily work requires managing context
- 15:27across multiple platforms, this changes everything.
- 15:30If you are constantly juggling files, cross -referencing
- 15:33emails, and checking calendars, you really need
- 15:36this architecture. The defining value metric
- 15:39here is repeatability. If you execute a complex
- 15:43cognitive task more than twice a week, This system
- 15:47saves massive amounts of time. Right, like daily
- 15:50stand -up summaries, weekly content planning,
- 15:53client meeting prep. Yep, the ultimate automation
- 15:55stack is clear. Context plus skills plus scheduled
- 15:59tasks equals total repeatability. It drastically
- 16:02reduces the manual steps of your day. You aren't
- 16:05constantly copying and pasting data between tabs
- 16:07anymore. It handles the tedious gathering and
- 16:10organizing phases perfectly. Yeah. But the material
- 16:13is very firm about one specific limitation. What's
- 16:16that? Human review remains absolutely essential.
- 16:19Yeah, you cannot just blindly forward its outputs.
- 16:21You still have to review the logic. Definitely.
- 16:23And especially if the connected apps drop their
- 16:25sync overnight, the AI will flag the missing
- 16:28info like we talked about. But you are ultimately
- 16:30responsible for the final deliverable. Treat
- 16:32it like an incredibly fast, highly capable junior
- 16:35coworker. It is not a replacement for your own
- 16:38judgment. You manage its focus. If you use it
- 16:41to anchor context and eliminate repetitive setup,
- 16:45the return on the time invested is just massive.
- 16:48It truly is. So here's our call to action for
- 16:50you today. Do not try to automate your entire
- 16:54professional life by tomorrow morning. Please
- 16:56don't. That is just a recipe for overwhelming
- 16:59frustration. Pick just one single annoying, highly
- 17:03repeatable daily task. Maybe it's summarizing
- 17:06project tickets. Maybe it's drafting the morning
- 17:08standup agenda. Just build one specific skill
- 17:11for that one specific task. Let the AI handle
- 17:15the repetitive gathering phase. And just see
- 17:18how it changes your workflow. And just experiment
- 17:20with the architecture. Because getting comfortable
- 17:22with these persistent workspaces now, it gives
- 17:24you a massive operational advantage later. It
- 17:26absolutely does. It frees up your cognitive bandwidth
- 17:29for the actual deep creative work that matters.
- 17:31But it also leaves us with a deeper, slightly
- 17:33unsettling question. Oh. Think about that morning
- 17:36briefing running completely on autopilot. If
- 17:38your AI is independently reading your private
- 17:41emails, wait, sorry, independently reading your
- 17:44emails, checking your calendar, evaluating your
- 17:46priorities, and structuring your tasks while
- 17:48you sleep, are you still managing your work?
- 17:51Or are you just managing the AI that does your
- 17:54work? Beat, thanks for joining us on the Deep
- 17:56Dive. We'll see you next time.