What Claude Cowork Actually Does (And Why It's Different)

What Claude Cowork Actually Does (And Why It's Different) 🔗 **Source:** [What Claude Cowork Actually Does (And Why It's Different)](https://www.youtube.com/watch?v=_42sUaf037k) > [!note]+ Document Notes > tags: [claude, artificial-intelligence, productivity-tools] > author: [MISSING] > source: YouTube > reference: https://www.youtube.com/watch?v=_42sUaf037k > date created: 2025-01-17 21:00 > > # 🏷️ Title: What Claude Cowork Actually Does (And Why It's Different) >  ## 🧰 Action Inbox - [H1] Try Claude Co-work for collaborative AI workflow — replaces traditional chat interface #artificial-intelligence - [H2] Test folder-based context sharing for large projects — better than file-by-file uploads #productivity-tools - [H3] Explore plan mode for complex tasks requiring research — asks follow-up questions unlike standard chat #claude ## 🧭 Table of Contents - [📝 Summary](#-summary) - [📖 Main Content](#-main-content) - [🗒️ Notes](#️-notes) - [📚 References](#-references) - [🗣️ Paraphrase](#️-paraphrase) - [📰 Newsletter](#-newsletter) - [✅ Action Items](#-action-items) - [🧪 Hypotheses](#-hypotheses) - [📊 Methodology](#-methodology) - [❓ Questions](#-questions) - [🔮 Future Research](#-future-research) - [💡 Implications](#-implications) - [🧠 Learning Styles Assessment](#-learning-styles-assessment) - [📊 Accessibility Assessment](#-accessibility-assessment) - [🌍 DEIB Assessment](#-deib-assessment) - [✅ Logical Validity](#-logical-validity) - [🛡️ DARVO Analysis](#-darvo-analysis) - [🎯 Conclusion](#-conclusion) ### 🔗 References - [Claude Co-work](https://claude.ai) - [Anthropic](https://anthropic.com) ## 📝 Summary - Anthropic launched Claude Co-work in January 2025 as an evolution from chat-based AI to collaborative workspace AI - Unlike traditional chat interfaces, Co-work provides persistent file management and can execute tasks rather than just discuss them - Key features include folder-based context sharing, parallel processing with sub-agents, and plan mode that asks clarifying questions - Available on macOS desktop app for $20/month standard plan, with Windows support coming soon - Demonstrated through book manuscript editing and travel planning examples showing comprehensive analysis and deliverable creation ## 📖 Main Content Claude Co-work represents a fundamental shift from conversational AI to collaborative AI. Traditional LLM interfaces like ChatGPT and Claude's original chat mode leave users with conversations but no tangible outputs. Co-work addresses this by creating a shared workspace where AI can actually execute tasks and manage files persistently. The product evolution shows Anthropic's recognition of user behavior: Claude Code launched in February 2024 for developers, but users began applying it beyond coding to project management, research, and analysis. Co-work makes these capabilities accessible to non-developers through a more user-friendly interface. Key technical capabilities include handling large context windows (demonstrated with a 45,000-word book manuscript), spawning sub-agents for parallel processing, and maintaining persistent file workspaces. The interface shifts from "new chat" to "new task" framing, emphasizing productivity over conversation. ## 🗒️ Notes #### ⏱️ [00:00](https://youtu.be/_42sUaf037k?t=0) — [H1] The core problem with chat-based AI - Key idea: Traditional AI chat creates conversations but no persistent value or deliverables - Evidence: "You step back and think, well, where is the result? Where's the value that was created from all that chatting back and forth?" #### ⏱️ [01:15](https://youtu.be/_42sUaf037k?t=75) — [H2] Claude Code origins and unexpected use cases - Key idea: Developers used Claude Code beyond programming for general productivity tasks - Evidence: "Software developers started using Claude to code not just for coding but for everything else... managing their to-do list, move forward their project, do research, create spreadsheets" #### ⏱️ [02:30](https://youtu.be/_42sUaf037k?t=150) — [H1] Claude Co-work launch and positioning - Key idea: Co-work makes Claude Code capabilities accessible to non-developers in January 2025 - Evidence: "Getting the power of Claude code and making it much more accessible and user-friendly so that everyone else can take advantage of it" #### ⏱️ [03:45](https://youtu.be/_42sUaf037k?t=225) — [H2] Interface hierarchy and user progression - Key idea: Three modes represent skill levels - Chat (beginner), Co-work (advanced), Code (expert) - Evidence: "Chat you can think of as the beginner mode, code as the advanced mode, and co-work... like chat plus" #### ⏱️ [04:30](https://youtu.be/_42sUaf037k?t=270) — [H1] Task-oriented framing vs conversation - Key idea: Interface language shifts from "new chat" to "new task" emphasizing productivity - Evidence: "That little change in wording is so interesting... it's really kind of framing this as what is something currently on your to-do list" #### ⏱️ [05:15](https://youtu.be/_42sUaf037k?t=315) — [H1] Folder-based context sharing capability - Key idea: Co-work prefers working in folders with free access rather than individual file uploads - Evidence: "It prefers... to have a shared workspace a folder where it has free permissions and free access to do whatever it needs" #### ⏱️ [07:30](https://youtu.be/_42sUaf037k?t=450) — [H2] Parallel processing with sub-agents - Key idea: Co-work can spawn multiple AI instances to handle large tasks in parallel - Evidence: "It's using sub agents... spinning off different versions of itself, each with their own context window and their own to-do list" #### ⏱️ [09:45](https://youtu.be/_42sUaf037k?t=585) — [H1] Large context window breakthrough - Key idea: Co-work handles entire book manuscripts where standard Claude chat failed - Evidence: "I actually tried to put the entire book manuscript into the standard Claude chat... and it couldn't handle it. Couldn't handle that much context all at once" #### ⏱️ [11:00](https://youtu.be/_42sUaf037k?t=660) — [H3] Pricing and accessibility improvements - Key idea: Co-work available at $20/month standard plan, down from $150-200 previously required - Evidence: "Just a matter of around a month ago, you needed the higher tier... $150 or $200 per month... Now both Claude Code and the new co-work are available at the lowest paid plan" #### ⏱️ [15:30](https://youtu.be/_42sUaf037k?t=930) — [H1] Plan mode asks clarifying questions - Key idea: Co-work asks follow-up questions unlike standard chat which dives in immediately - Evidence: "Standard chat will almost never or rarely ask many follow-up questions. Co-work on the other hand will" #### ⏱️ [18:45](https://youtu.be/_42sUaf037k?t=1125) — [H2] Proactive deliverable creation - Key idea: Co-work creates formatted outputs and suggests practical formats without being asked - Evidence: "It was the one that actually suggested a word doc format... we're going to be on the road in Mexico. There's often not good cell phone service" #### ⏱️ [20:15](https://youtu.be/_42sUaf037k?t=1215) — [H3] Local storage limitation - Key idea: Co-work conversations only stored locally, not synced across devices - Evidence: "Everything you do with co-work... is only stored locally. It's only stored on your machine" #### ⏱️ [21:30](https://youtu.be/_42sUaf037k?t=1290) — [H1] Collaborative workspace paradigm shift - Key idea: Co-work changes from face-to-face conversation to side-by-side collaboration - Evidence: "It's not you face to face with the LLM talking. It's as if they're taking a seat at the table... looking at a shared set of working documents" ## 📚 References - [Claude AI Platform](https://claude.ai) - [Anthropic Company](https://anthropic.com) - [macOS Desktop App](https://claude.ai/download) ## 🗣️ Paraphrase - Claude Co-work transforms AI from a chatbot into a collaborative workspace partner that can actually execute tasks and manage files persistently - The evolution from Claude Code shows how developer tools expanded to general productivity, leading Anthropic to create a user-friendly version for everyone - Key differentiators include folder-based context sharing, parallel sub-agent processing, and plan mode that asks clarifying questions before diving into tasks ## 📰 Newsletter **AI Collaboration Gets Real: Claude Co-work Changes the Game** The days of chatting with AI and getting nothing tangible back are ending. Anthropic's new Claude Co-work transforms AI from a conversation partner into a true collaborator that can handle your files, execute complex tasks, and deliver actual results. What makes this different? Instead of uploading files one by one, you give Claude access to an entire folder. It can spawn multiple AI agents to work in parallel, handle massive context like entire book manuscripts, and even suggests the best output format for your needs. The interface shifts from "new chat" to "new task" - a small change that signals a big philosophical shift toward productivity. The tool shines for complex projects requiring research, analysis, and deliverable creation. Whether you're editing a manuscript, planning travel, or managing projects, Co-work asks the right follow-up questions and creates professional outputs without hand-holding. Currently available on macOS for $20/month (Windows coming soon), Co-work represents the next evolution in AI assistance - from talking about work to actually doing work together. Resources: [Claude AI Platform](https://claude.ai), [Anthropic Research](https://anthropic.com), [macOS Desktop App](https://claude.ai/download) ## ✅ Action Items #### DevRel - Developer advocacy: Create workshops comparing Claude Code vs Co-work for different audiences - Education: Develop curriculum around collaborative AI workflows and workspace management - Developer experience: Document best practices for folder-based context management - Developer marketing: Target campaigns showing Co-work benefits over traditional chat interfaces - Community management: Facilitate discussions around AI collaboration paradigm shifts #### Community - Contributors: Share Co-work integration patterns and workflow optimizations - Sponsors: Evaluate productivity tool budgets to include collaborative AI capabilities - Ecosystem: Coordinate with other AI tool providers on workspace standards ## 🧪 Hypotheses - Co-work's folder-based approach will become the standard for AI collaboration tools - Parallel sub-agent processing will significantly improve complex task completion rates - Plan mode questioning will reduce iteration cycles compared to traditional chat interfaces - Local-only storage will limit adoption until cloud sync capabilities are added ### 📊 Methodology - Comparative analysis between Co-work and traditional chat interfaces for complex tasks - User productivity metrics measuring deliverable creation time and quality - Context window utilization studies with large document processing ### ❓ Questions - How will Co-work's local storage limitation affect team collaboration workflows? - What security implications arise from giving AI broad folder access permissions? - Can the sub-agent parallel processing model scale to even more complex multi-step tasks? ### 🔮 Future Research - Investigation into optimal folder organization strategies for AI collaboration - Analysis of user behavior changes when shifting from chat to task-oriented AI interfaces - Exploration of collaborative AI integration with existing productivity software ecosystems ### 💡 Implications - Practice: Productivity workflows will shift toward AI collaboration over traditional automation - Policy: Organizations will need new guidelines for AI workspace access and data management - Security: Folder-level AI permissions require updated security protocols and audit trails - Education: Training programs must evolve from AI prompting to AI collaboration skills - Funding: Productivity tool budgets will increasingly prioritize collaborative AI capabilities over traditional software licenses ### 🧠 Learning Styles Assessment Reading level estimate: 10th grade #### 👀 Visual learners - Present: Screenshots of Co-work interface showing three modes and folder selection - Improve: Workflow diagrams comparing chat vs co-work paradigms, sub-agent visualization #### 🔊 Aural learners - Present: Video demonstration with narrated examples - Improve: Podcast-style summary of key Co-work benefits and use cases #### 🗣️ Verbal learners - Present: Clear definitions of co-work vs chat vs code modes, analogies to collaborative work - Improve: Story-driven case studies, terminology comparison charts #### 🧩 Kinesthetic learners - Present: Step-by-step folder setup and task creation examples - Improve: Hands-on checklists for Co-work onboarding, interactive tutorials #### 🧠 Logical learners - Present: Timeline of Claude product evolution, systematic feature comparisons - Improve: Decision trees for choosing between AI modes, ROI calculations #### 🧑‍🤝‍🧑 Social learners - Present: Demonstrations of collaborative workflows - Improve: Team adoption guides, sharing strategies for Co-work outputs #### 🧘‍♀️ Solitary learners - Present: Individual productivity examples like book editing and trip planning - Improve: Personal workflow templates, self-assessment questionnaires #### 🌿 Naturalistic learners - Present: Travel planning example connecting to real-world exploration - Improve: Environmental impact considerations of AI processing, sustainable AI usage patterns ### 📊 Accessibility Assessment #### Screen reader semantics - Present: Clear heading structure and descriptive interface elements - Improve: Alt text for interface screenshots, better description of visual workflow elements #### Keyboard navigation - Present: Desktop app with standard keyboard shortcuts - Improve: Document keyboard shortcuts for Co-work specific features #### Contrast and readability - Present: Standard desktop application contrast ratios - Improve: High contrast mode documentation, text scaling options #### Text resizing and reflow - Present: Responsive interface elements in desktop app - Improve: Documentation of accessibility settings, font scaling options #### Media and captions - Present: Video demonstration with spoken narration - Improve: Transcript availability, closed caption accuracy verification #### Forms and interactions - Present: Clear file selection and permission dialogs - Improve: Better error messages for folder access issues, confirmation dialogs ### 🌍 DEIB Assessment #### Cultural breadth - Present: Mexico travel planning example shows non-US context - Improve: More diverse geographic examples, multilingual capability demonstrations #### Inclusive language - Present: Gender-neutral terms for most descriptions - Improve: Avoid assumptions about family structures, work contexts #### Neurodiversity support - Present: Clear step-by-step examples with visual confirmation - Improve: More frequent progress indicators, alternative input methods #### Representation - Present: Family-friendly travel planning inclusive of pregnancy considerations - Improve: Varied user personas beyond families and writers #### Psychological safety - Present: Respectful demonstration of AI capabilities - Improve: Clear boundaries around AI limitations, privacy considerations ### ✅ Logical Validity #### Strong reasoning - Deductive strength: Clear progression from problem identification to solution demonstration - Inductive strength: Uses specific examples to generalize about Co-work capabilities - Causal reasoning: Establishes clear cause-effect between user needs and product development - Analogical strength: Effective comparison between chat interfaces and collaborative workspaces #### Fallacies - Hasty generalization: Limited examples may not represent all use cases - Appeal to novelty: Assumes newer Co-work approach is inherently better than chat #### Tone of voice - Overall tone: Enthusiastic and instructional, appropriate for product demonstration content ### 🛡️ DARVO Analysis - Deny: No significant denials present - Attack: No attacks on competing products or critics - Reverse victim/offender: Not applicable to this content - Emotional cues and rhetoric: Positive enthusiasm appropriate for product demonstration ## 🎯 Conclusion - Claude Co-work represents a paradigm shift from conversational AI to collaborative AI that can execute tasks and manage persistent workspaces - Key advantages include folder-based context sharing, parallel sub-agent processing, and plan mode that asks clarifying questions before task execution - The $20/month accessibility and demonstrated capabilities with large context windows make this a significant advancement in practical AI productivity tools ## 🧱 Refactor Appendix ### 📚 Literature Split Pad Purpose: Stage CEE units that become Literature notes. Split level: H3 ### [H1] Co-work enables persistent AI collaboration beyond chat > [!quote] What Claude Cowork Actually Does (And Why It's Different) by [MISSING] --- ## Full Transcript [00:00] If you've been using something like Chat [00:01] GPT or Claude, tell me if you've had [00:03] this experience. You open up a chat, [00:05] [music] you have a great back and forth [00:07] conversation full of ideas and insights [00:09] and questions and answers, but then at [00:11] the end of it, you step back and you [00:13] sort of think, well, where is the [00:16] result? Where's the value that was [00:18] created from all that chatting back and [00:21] forth? Where did it go? It's felt like [00:23] the AI is held back somehow. It's really [00:25] just a little buddy, a little assistant [00:28] that you can chat with, which is [00:30] valuable, [music] [00:30] but it's really been shackled and held [00:33] back by its inability to do anything [00:35] more except talk. A little bit of [00:37] background. In February of 2025, [music] [00:39] Anthropic, the company that makes [00:41] Claude, tried to address this [music] by [00:44] quietly coming out with a new product [00:47] called Claude Code. What started to [00:50] happen [music] is that software [00:52] developers started using claw to code [00:55] not just for coding but [music] for [00:57] everything else. They were using it to [00:59] manage their to-do list to move forward [01:01] their project to do research to [music] [01:04] create spreadsheets and analyses. [01:06] Basically the use cases exploded to [01:08] practically everything that you might [01:10] want to do on a computer. So in January [01:13] 2026, [music] just a few weeks ago, [01:15] Enthropic recognized this trend and they [01:17] came out with yet another product called [01:19] Claude Co-work, which is basically [01:22] getting the power of Claude code and [01:24] [music] making it much more accessible [01:26] and userfriendly so that everyone else [01:28] can take advantage of it. Instead of [01:30] chatting, you now work alongside the AI [01:34] on a shared set of files that persist [01:37] [music] over time. Instead of just [01:39] talking about work but leaving you to do [01:41] most of the execution, Claude [music] [01:43] can now actually execute much of the [01:46] work itself. Let me show you what this [01:48] looks like. This is the MacOSS desktop [01:51] app. And what I'll have you notice is [01:54] right up here at the top there are now [01:55] these three separate tabs. Chat was the [01:58] original one which you can see here [02:00] below. This is the standard typical LLM [02:03] interface that we all know and love. And [02:05] then code is over here on the right. and [02:07] that was the second one to come out and [02:09] co-work the new one is right here in the [02:11] middle. It's kind of interesting to [02:12] think about why they're in that order. [02:14] And I think it's basically because chat [02:16] you can think of as the beginner mode, [02:18] code as the advanced mode, and co-work [02:22] mode. I almost think of it like chat [02:24] plus, like an advanced version of the [02:26] chat interface. You can see it looks [02:28] quite similar to chat. There's a [02:30] dialogue box in the middle, a bunch of [02:32] options along the top or bottom, but [02:34] there's also some differences which I [02:36] want to point out. Now, the first thing [02:38] is you'll notice it's an early research [02:40] preview. So, you want to always keep [02:42] that in [music] mind. This is a beta. [02:44] This is an experiment. Could have some [02:45] weird behavior, some things you don't [02:47] expect. The second thing to notice is if [02:49] you open the sidebar here, instead of [02:51] new chat, it says new task. [music] And [02:54] that little change in wording is so [02:57] interesting to me. Before it was a chat. [02:59] It was a little conversation now because [03:02] of what this is called and also because [03:03] this says let's knock something off your [03:05] list. It's really kind of framing this [03:07] as what is something currently on your [03:09] to-do list that we can take care of [03:10] right now. That's a much more powerful [03:13] and promising framing in my opinion. And [03:15] then the third thing to notice is if you [03:17] click right here work in a folder, it [03:19] not only gives you the option to add [03:22] some sort of connector or something [03:24] which you can do right here, but it [03:26] actually wants to work in a folder. it [03:28] it prefers it's its preference is to [03:31] have a shared workspace a folder where [03:33] it has free permissions and free access [03:36] to do whatever it needs. So let me give [03:38] you an example of how powerful that is. [03:40] I'm going to go ahead and click work in [03:42] a folder. I'm going to go to my projects [03:45] and then go to the book I'm writing. And [03:47] then we have a folder here called lip [03:49] which stands for life in perspective my [03:51] upcoming book lip context. So, inside [03:55] that folder is the latest version of my [03:58] book manuscript. So, I'll go ahead and [04:00] select that folder. So, it now knows [04:02] that's its little workspace, its little [04:03] playground. You'll notice that I have [04:05] the option to allow this one time or [04:08] always allow it to access that file that [04:11] I've selected and all of its subfolders. [04:13] So, I'll go ahead and click allow. I [04:16] didn't have to remage through and [04:17] individually choose all these specific [04:20] files. I just gave it a space, a folder, [04:23] and now it can play, it can act, it can [04:26] explore, it can research anything that [04:28] is in that space. So, let me show you [04:30] how powerful this can be. I'm going to [04:32] go ahead and say, what are the suggested [04:35] improvements that you would make to this [04:37] book manuscript to make it more [04:39] interesting, impactful, and accessible [04:42] to a broad readership? And then return. [04:45] So, a couple things as that's working. [04:47] You'll notice up here on the right, it [04:49] will show you its progress. Claude [04:51] Co-work actually has like its own [04:53] private to-do list. It actually is [04:55] mapping out the different tasks it will [04:57] need to do and then it checks them off [04:59] one by one. And you'll often see [05:00] evidence of those right here in the [05:03] stream of thought, which is called [05:04] thought process. Here you'll see the [05:06] folder that it's accessing as part of [05:08] its context. As it creates and edits and [05:11] manages other files, you might see those [05:13] appear right here where it says context [05:15] as well. You can see that it found the [05:17] manuscript for my book which is in a [05:20] word document format [music] and now [05:22] it's reading through it and providing [05:24] suggestions. So something you'll notice [05:26] as it's taking a few minutes to do this [05:28] is I can click into this little kind of [05:30] work in progress area and see a [05:33] playbyplay of what it's doing. So you'll [05:35] notice it's uh realizing the manuscript [05:38] is quite long. It's about 45,000 words. [05:40] It's actually trying different [05:42] approaches. So here you can see that [05:44] it's reading it in chunks. Okay, it's [05:46] still quite large, so it's using sub [05:48] agents. It's actually spinning off [05:49] different versions of itself, each with [05:51] their own context window and their own [05:53] to-do list to read different sections in [05:55] parallel and then compile comprehensive [05:57] feedback. So, it's almost like you're [05:59] spawning a team of different AIs that is [06:02] each tackling in parallel different [06:04] parts of the task that you've assigned [06:06] it. So, you can see here it's reading [06:07] and analyzing part one of the book, part [06:09] two, part three, final chapters. I can [06:11] even click into here if I really want [06:13] even more detail as to what it's doing [06:15] specifically. And there we go. That took [06:18] around five or six minutes. And let's [06:20] see what it came up with. It says, [06:21] "Based on my comprehensive analysis of [06:24] life and perspective, here are my [06:25] suggested improvements organized by [06:27] priority and impact." So notice that [06:30] this whole framework for how to provide [06:31] feedback was not something that I told [06:33] it. It came up with this on its own. Has [06:35] critical issues, high impact. There's an [06:37] accessibility gap. [music] The book [06:38] assumes that this is for business [06:41] owners, knowledge workers, people with [06:42] significant time or resources. That is a [06:44] major issue. Chapter 12 may alienate [06:46] most readers. Structural pacing issues. [06:49] I like that it doesn't just identify the [06:51] problem. It also suggest very [06:53] specifically how to fix it. A weak [06:55] introduction to conclusion arc. Then [06:57] there's high priority improvements, [06:59] missing encounter examples, guidance too [07:01] vague. So this is just remarkable, you [07:04] guys. This is so amazing. I actually [07:07] tried to put the entire book manuscript [07:10] into the standard Claude chat just a [07:12] matter of weeks ago, maybe five or six [07:14] weeks, and it couldn't handle it. [07:15] Couldn't handle that much context all at [07:17] once. So, I had to go chapter by [07:18] chapter. The problem with that is that [07:21] Claude as an editor cannot give as good [07:24] of advice if it only sees one chapter. [07:26] It really needs the context of the [07:28] entire book. This is a gamecher in terms [07:30] of how much context you're able to give [07:33] the LLM all at once so that it can see [07:35] the big picture. Now, before we go any [07:37] further, let's talk about what this [07:39] costs. There's two things you're going [07:41] to need. At least a paid standard plan, [07:44] which is the lowest tier at $20 per [07:47] month as of the filming [music] of this [07:48] video. And the second thing is the [07:51] MacOSS desktop app. It's currently not [07:53] available for Windows, although that's [07:55] coming soon. [music] What's pretty [07:56] amazing is just a matter of around a [07:58] month ago, you needed the higher tier, I [08:01] think a $150 or $200 per month plan to [08:05] get access to Claude Code. Now, both [08:07] Claude Code and the new co-work are [08:10] available at the lowest paid plan, which [08:12] is just 20 bucks. I would really [08:14] recommend that anyone who is already [08:16] paying for any LLM, which is likely $20 [08:19] a month, really consider adopting [08:22] Co-work instead. And the reason for that [08:24] is even if you're not a software [08:26] developer or a writer like me or a [08:28] content creator, in other words, you [08:30] don't have a large code base or a book [08:32] manuscript or some big thing you're [08:34] working on, I would still recommend it [08:36] even for your typical day-to-day [08:38] questions and tasks. And I'll show you [08:40] why in the next example. Let's go ahead [08:43] and open up the sidebar, create a new [08:45] task, and I'd like to show you a [08:47] different real life example that I did [08:49] just in the last few days that doesn't [08:52] make use of existing context. A couple [08:54] weeks ago, I decided that I wanted to [08:57] take a little road trip in central [08:59] Mexico with my family. We've been living [09:01] here a year and a half and have not done [09:03] any road trips. So, I thought that would [09:04] be a good way to take advantage of the [09:06] fact that we live here and have a car. [09:08] So, I'm going to go ahead and use [09:09] Whisper Flow, hold down the function [09:11] key, and just give it the trip that I'm [09:14] envisioning. I'd like your help planning [09:16] a road trip within a few hours of our [09:19] home in Vaia de Bravo, Mexico. This will [09:22] be myself, my wife Lauren, who is 5 [09:26] months pregnant, and our two kids who [09:28] are ages 3 and 5. I'd like it to be [09:31] around four nights within a few hours [09:34] drive of where we live. And it should be [09:36] quite easy and nonstrenuous because of [09:40] the two young kids and the pregnant [09:41] wife. So nothing too extreme or too [09:44] taxing. Prioritize nature, culture, [09:47] interesting and unusual experiences. [09:50] Prioritize places that are a little bit [09:52] off the beaten path and not something [09:54] that is going to be too touristy, [09:55] especially something with international [09:58] tourists. All right. So, I'm actually [10:00] going to turn off this context so that [10:03] it's not drawing on my book manuscript. [10:05] And then I'll hit let's go. So, you'll [10:07] notice that it's doing web research, [10:09] which is a capability that co-work has. [10:11] And there you see all the different [10:13] sites that it is researching. 10 [10:15] different websites and then 10 more. I [10:17] could click into one of these if I [10:19] wanted to see more details about what [10:21] it's looking at. Now it's doing another [10:22] round of research around familyfriendly [10:25] accommodations, driving distances, off [10:27] the-beaten path options, taking into [10:29] account my preferences. So here you can [10:31] see it's starting to research more [10:33] specifically a place called Patuado with [10:35] 10 more results. Another one called [10:37] Malino with 10 more results. Few more. [10:40] It's summarized back to me what the [10:42] criteria is that it's using. It's giving [10:44] me the options that it's researched. And [10:46] this is something I want to point out [10:47] right here. This is a feature that is [10:49] unique to co-work and code which is [10:52] called uh plan mode. Taking a step back [10:54] here, you'll notice that in the standard [10:57] chat, anytime you ask the LLM to do [10:59] anything, it just dives head first. It [11:02] just goes for it, right? It's almost [11:04] like an overeager kid. It's like an [11:06] overeager intern that wants to please [11:09] you and really wants to impress you, [11:11] which is good. The problem is that it [11:13] doesn't know stuff. It doesn't it [11:15] doesn't have the right context. is [11:17] making assumptions and the standard chat [11:19] will almost never or rarely ask many [11:22] follow-up questions. Co-work on the [11:24] other hand will. So you can see here [11:25] it's asking me a follow-up question [11:27] which is really a critical piece of [11:29] information which is when are you [11:30] planning to take this trip? And then [11:31] it's even giving me preformulated [11:33] answers that I can choose from or you [11:35] can write in your own answer. So I'm [11:37] going to say midFebruary which is when [11:39] we're planning on making this trip. And [11:41] then another follow-up question which is [11:43] again is pretty crucial. What type of [11:45] accommodations do you prefer? Boutique [11:46] hotels, chain hotels, vacation rentals. [11:49] I'm going to say a mix. So, it's [11:51] recommending that we take a visit to the [11:53] Monarch Butterfly Sanctuaries. It's [11:55] given me a fournight itinerary. Night [11:58] one there, day two, nights two and four. [12:01] And this is where co-work will tend to [12:04] present more detail than the standard [12:06] chat. The standard chat will not usually [12:08] tell you what it's less certain about. I [12:10] will just confidently give you the [12:11] answer, but it's giving me what it's not [12:13] sure about and what would change my [12:15] recommendations. And what I'm noticing [12:17] is that there's some additional context [12:18] that it should use. So, I'm going to [12:20] basically evolve and iterate on this [12:22] plan through conversation until I arrive [12:24] at something that I have confidence in. [12:26] We live in VIA de Bravo, which is very [12:29] nearby, one of the biggest monarch [12:31] butterfly sanctuaries. So, it's very [12:34] easy for us to visit those. Revise this [12:36] plan to not include the butterflies. All [12:39] right. So, it has now removed the [12:41] butterflies. It's recommending we go to [12:43] Morelia to Patu to a place that I cannot [12:47] pronounce [12:48] and a few others. This looks great. [12:50] [music] There's some things that I can [12:52] adjust like if a 3-hour drive is too [12:54] much. No, I think that's fine. If I want [12:56] to go more off the beaten path, if you [12:57] want more nature. So, in real life, I [13:00] did go back and forth a few more times [13:01] to revise this. But what I want to show [13:04] you is what we actually ended up with. [13:06] So, I'll say this looks perfect. go [13:08] ahead and summarize this overall [13:10] itinerary with all the details that I'll [13:13] need daybyday in a format that I can [13:16] print out and take with me. You can see [13:18] right here inside a new working folder [13:20] that it's created, it has now created a [13:23] word document to store all these plans. [13:25] And there we go. Let's go ahead and open [13:27] this in Microsoft Word. It's going to [13:29] ask you for permission. And voila. Look [13:32] at this thing. It's got a day-by-day [13:33] itinerary with everything including [13:36] specific driving times, what we do in [13:37] the morning, what we do in the [13:38] afternoon. It's got practical notes like [13:41] accommodations, roads, weather, food. [13:43] None of this were was things that I [13:45] asked for. This was all on its own [13:47] initiative. In fact, the first time I [13:48] did this, it was the one that actually [13:50] suggested a word doc format. At first, I [13:52] was like, a word doc? What is this like [13:55] 1995? [13:56] But then I realized, wait a minute, [13:57] we're going to be on the road in Mexico. [14:00] There's often not good cell phone [14:01] service. So, I can't have it just on my [14:03] phone. It's actually very helpful if [14:05] there's something I can print out and [14:07] keep with me in the car. All right, [14:09] we've now looked at a couple in-depth [14:11] examples. Before you dive into co-work, [14:13] one thing to keep in mind is that [14:15] everything you do with co-work, just as [14:16] with code, is only stored locally. It's [14:19] only stored on your machine. So, if you [14:21] switch to a different computer or a [14:22] tablet or a phone, you will not find [14:25] that conversation stored there. [14:27] Something you should also keep in mind [14:28] is that context works differently. [14:30] Instead of clicking into a project and [14:32] uploading, you know, project files one [14:34] by one, you'll just have a folder on [14:37] your computer where all the context and [14:39] the work in progress is stored. This is [14:42] actually a much better approach, but it [14:43] does take some [music] getting used to. [14:45] It takes a little bit of a mindset shift [14:47] to work with co-work. We're used to just [14:48] opening up a chat without having to [14:51] think of some of these other [14:52] considerations as if we're opening up a [14:54] Google search. Although it does require [14:55] a little more that you have to keep in [14:57] mind, especially when compared to code, [14:59] I think co-work is a huge step forward [15:02] [music] that anyone who's using LLMs [15:04] today should go ahead and at least try [15:06] out, if not adopt. It's not you face to [15:09] face with the LLM talking. It's as if [15:12] they're taking a seat at the table or at [15:14] the desk next to you. You're both [15:16] looking at a shared set of working [15:18] documents [music] and taking turns and [15:20] collaborating and handing off different [15:23] tasks to each [music] other in a very [15:25] dynamic way that is more natural, more [15:28] exciting, and of course, far more [15:30] productive. We've covered a lot in this [15:32] video, [music] so we've actually created [15:34] a completely free getting started guide, [15:36] which you can find in the description or [15:39] the comments below. It includes a [15:40] summary of everything [music] that I've [15:42] discovered, all the dos and the don'ts [15:44] and the tips and the tricks for how to [15:46] take advantage of co-work. And if you [15:48] discover any other tips or tricks that [15:50] you want to share with us, go ahead and [15:52] share those in the comments below. [15:58] >> [music]