Latest / AI Prioritized / Claude 5.5 Updates You Need to Know: How to Use The New AI Models
Transcript
- 0:00SPEAKER_00: Anthropic just dropped four major updates in the last two weeks.
- 0:04SPEAKER_00: Chat and co-work are becoming one interface, and the full 5.5 model family is here.
- 0:11SPEAKER_00: Opus 5.5, Sonnet 5.5, and Haiku 5.5.
- 0:16SPEAKER_00: If you're still choosing between separate tabs and using the same model for everyday tasks, you're working from the old AI playbook.
- 0:24SPEAKER_00: I'm gonna leave you with two things.
- 0:26SPEAKER_00: First, you're gonna see how the combined chat and co-work interface can take one request and work across your files, your applications, and complete a multi-step workflow and scheduling all in one shot.
- 0:39SPEAKER_00: Second, you'll learn which model to use for which job.
- 0:42SPEAKER_00: Opus for complex judgment, sonnet for clearly defined work, and haiku for speed and volume.
- 0:48SPEAKER_00: By the end of this video, you'll have one simple model selection rule that you can add to Claude with your instructions or your team SOP.
- 0:58SPEAKER_00: Welcome to AI Prioritized.
- 0:59SPEAKER_00: I'm Nicole McCain.
- 1:01SPEAKER_00: And after 20 years leading operations, product, and technology initiatives for the top Fortune 500 companies, I now help business owners build practical AI workflows, upskill their teams, and turn AI investments into measurable ROI.
- 1:18SPEAKER_00: Now, working with founders and B2B clients day in and day out, I see most of them are still using AI the old way.
- 1:24SPEAKER_00: And what I tell them is you have to change the way you work with AI because using it the old way can cost you more time, more usage, and put the most expensive model on work that a smaller model can easily handle, faster and cheaper.
- 1:40SPEAKER_00: And you don't want to make those expensive mistakes in your AI workflow.
- 1:44SPEAKER_00: But the bigger risk comes when you're given Claude access to your files, your applications, and everything else before deciding where human intervention is really needed.
- 1:54SPEAKER_00: So stay with me and I'm gonna give you one decision rule that you can paste into your Claude instructions that will tell Claude which model should handle the work, at what point, and when a task must stop, and when a human is needed for decision making.
- 2:09SPEAKER_00: Alright, so let's start with the model that Anthropic built for our most complex work, which is Opus 5.5.
- 2:18SPEAKER_00: Now, Opus 5.5 is Anthropic's newest top model.
- 2:22SPEAKER_00: It launched on September 22nd, and Anthropic calls it the first model in the Claude 5.5 family.
- 2:30SPEAKER_00: Now, in our three-level system that we're gonna go through, this is the judgment level model.
- 2:36SPEAKER_00: You can see here the recent updates that now it only costs$4 per million input tokens, which is 40% lower cost for you while you're working with this model.
- 2:48SPEAKER_00: With Opus 5.5, you can see that they have rated it as a top-performing model.
- 2:55SPEAKER_00: And they've also rated it for one of their safest models.
- 2:59SPEAKER_00: So Opus 5.5 achieved the best scores of any model to date in Anthropics Automated Behavioral Audit.
- 3:07SPEAKER_00: They're aligned suite that tests Claude across thousands of simulated scenarios.
- 3:12SPEAKER_00: It's much less, it is much less likely than recent models to take hard-to-reverse actions or act outside the boundaries it has been given.
- 3:22SPEAKER_00: And it's more resistant than Opus 5 to prompt injection.
- 3:27SPEAKER_00: So those are critical updates that will keep your business protected and make you feel safe when you're using the model for your daily work.
- 3:36SPEAKER_00: So let's jump into the claw interface on the web and let's scroll down to the bottom here where on the chat screen.
- 3:44SPEAKER_00: And now you can select the model that you would like.
- 3:47SPEAKER_00: So just toggle up to Opus 5.5, and then you can also select the effort.
- 3:53SPEAKER_00: So the effort could be low, medium, high, extra high, or max.
- 3:58SPEAKER_00: So you pick the level of effort that is best suited for the complexity of the work you're trying to achieve, the deep research that you're trying to achieve, and that specific task and outcome that you're looking for.
- 4:10SPEAKER_00: Now, for me, what I usually use this for every day, I use it for when I'm doing deep research.
- 4:15SPEAKER_00: I want more critical analysis, and I keep it primarily on the medium level because this model is one of the best and most capable models.
- 4:23SPEAKER_00: So medium is sufficient for the type of work that I'm doing.
- 4:27SPEAKER_00: Now, let's say I'm building out a client database in Cloud Code, then I would use Opus 5.5 effort level high because I'm asking it to build something more complex and reason with code.
- 4:39SPEAKER_00: But the medium is fine for your more complex everyday tasks, such as your financial dashboards, projections, deep research, and things like that.
- 4:48SPEAKER_00: So I love using this model for multi-step workflows and the deeper work.
- 4:52SPEAKER_00: And what Afro has said is that it takes fewer risky actions, it's less likely to do things that are hard to undo, right?
- 5:00SPEAKER_00: So I love that.
- 5:01SPEAKER_00: So I'm gonna jump into a prompt that I promised that I would give you.
- 5:05SPEAKER_00: Now I'm gonna drop it over here inside of Claude.
- 5:08SPEAKER_00: So let's try Opus 5.5 with a financial dashboard and a prompt.
- 5:13SPEAKER_00: You are my operation operations analyst.
- 5:16SPEAKER_00: Use only the attached file and bit build a one-page report with four parts.
- 5:21SPEAKER_00: List what changed, the three biggest movements with numbers, why it matters, the likely cause of each, label each one as a fact from the file or from your inference.
- 5:32SPEAKER_00: And what I should do this week.
- 5:34SPEAKER_00: Three actions, each which is suggested owner and a due date.
- 5:39SPEAKER_00: And fourth, what you're unsure about, anything missing, unclear, or worth checking.
- 5:44SPEAKER_00: If a file doesn't have the data or section, just tell me so instead of guessing.
- 5:49SPEAKER_00: Alright, so this is a financial projection dashboard.
- 5:52SPEAKER_00: This is a prompt example that you can use in your business when you're running your own financial analysis, deep dives with your numbers.
- 6:00SPEAKER_00: So this is a mock file demo.
- 6:02SPEAKER_00: We don't need to comp I'm not gonna complete it here.
- 6:04SPEAKER_00: I'm gonna move on.
- 6:05SPEAKER_00: Now, this prompt is key because you want Claude to not fill in the gap with something that is sneaking up or something that sounds right.
- 6:13SPEAKER_00: Asking you to separate facts from guesses and to show you what is insure about is what keeps you in charge instead of you outsourcing your judgment to AI.
- 6:24SPEAKER_00: So while it's running, I'll come over to a clean screen.
- 6:27SPEAKER_00: Now in anthropics doc, they say that the effort level is medium.
- 6:31SPEAKER_00: That's typically needed for those deep dive tasks.
- 6:34SPEAKER_00: And you can use the maximum effort for a fraction of the cost.
- 6:39SPEAKER_00: But I always want you to test the lower level of the model effort before you jump into a deeper model so that you don't waste your time or your tokens.
- 6:50SPEAKER_00: So try Opus 5.5 in these three steps that we walk through.
- 6:55SPEAKER_00: Pick the model, pick the level of effort for that model, and run your report and or run deep research and start with a medium effort so that you can see how it analyzes and synthesizes the data and what is bringing you back before you decide to try your higher before you decide to try a higher effort model.
- 7:14SPEAKER_00: So here are some jobs that you can give to Opus first if you want to test it out.
- 7:18SPEAKER_00: You can give it a competitor report, you can give it a proposal to write, your sales or cash review, or your monthly summary.
- 7:26SPEAKER_00: And you can always use the Claud for Small Business plugin, which I will show you really quick.
- 7:31SPEAKER_00: You can go into customize and you can go into plugins and then make sure you're on the Discover tab.
- 7:39SPEAKER_00: And then you can search for SMB or Small Business, and then you'll see the Small Business plugin, which has different skills that you can use for financial analysis, month end close, competitor analysis, QuickBook analysis, QuickBook or Financial Projections, legal or CRM or tax prep, and so on.
- 8:01SPEAKER_00: And always remember to set up guardrails, set permissions first, make sure that it's working according to the permissions and guardrails that you've set up, and give it one folder to work out of from so that it doesn't have to work across multiple files and only connect to what you need.
- 8:17SPEAKER_00: You may not need to connect it to every app and every tool at the same time.
- 8:22SPEAKER_00: Figure out what you need and start small before you go broader.
- 8:25SPEAKER_00: So let's move into the next update.
- 8:29SPEAKER_00: So let's move into the next big update.
- 8:31SPEAKER_00: So on September 16th, Anthropic announced that co-work and the chat window inside of Claude are no longer gonna be separated, they're gonna become one interface within Claude.
- 8:44SPEAKER_00: And this has been a real pain point for me because I'm always wondering why do I need to have a separate interface to go and automate my work and schedule it?
- 8:54SPEAKER_00: Why can't I just do everything in one window?
- 8:56SPEAKER_00: So now Anthropic has solved for that.
- 8:59SPEAKER_00: So this is what it used to look like.
- 9:01SPEAKER_00: There used to be a toggle, this was the second generation.
- 9:04SPEAKER_00: First, chat and co-work were completely separate windows, and the co work had to be ran on your desktop app, but now you can run co-work inside of the browser app.
- 9:16SPEAKER_00: And it's all in one interface now, which is a big relief for everyone, I'm sure.
- 9:21SPEAKER_00: So now that you don't have to choose which window inside of Cloud your task is gonna go in, that means that this one window can now read your files and your app.
- 9:32SPEAKER_00: And Cloud Announcement has noted that cowork can run, can run, can now run after you've closed your laptop.
- 9:41SPEAKER_00: That was a pain point before because you could you had to keep your computer on open all night for cowork to run your tasks, your schedule tasks, and your automations.
- 9:51SPEAKER_00: And that was always a concern for me because when you're sleeping, you never know what can happen.
- 9:54SPEAKER_00: So now the one window can reach all your files and all your apps.
- 9:59SPEAKER_00: But we always want to make sure that we set up the guardrails first.
- 10:02SPEAKER_00: So the permission setting I would use inside of Claude is so two permission settings I would use when I'm working with Claude.
- 10:12SPEAKER_00: You can check it out here.
- 10:14SPEAKER_00: There's auto, and then you could toggle down to manual.
- 10:19SPEAKER_00: Now I would select manual as the default just when you're starting out.
- 10:23SPEAKER_00: Auto lets Claude keeps going with an automatic safety check.
- 10:27SPEAKER_00: So let's say you're working on a finance report and it's running.
- 10:31SPEAKER_00: If you keep it on auto, it's gonna continue to work on that job without stopping and asking you for every little thing.
- 10:38SPEAKER_00: You may occasionally get a pop-up that says, okay, Claude needs your review or your permissions to do this or to take it the next step.
- 10:47SPEAKER_00: So the model will decide when it wants you to intervene or when it needs your decision.
- 10:53SPEAKER_00: Manual is you run in it, it stops, it asks you a question, it'll ask whether it can proceed, whether it can do this, whether it should update this, and then you say, Yes, go ahead.
- 11:05SPEAKER_00: And then it will run that one, it'll run that one part of the task, and then it'll come back to you and ask permission for the other part of the task, and so on.
- 11:13SPEAKER_00: So once you get comfortable with using Claude, you set up regard rails and permissions, it knows your business and your preferences.
- 11:22SPEAKER_00: Then I would say turn on auto mode because it is a pretty safe model, and I use it primarily on auto when I'm doing daily on almost everything that I'm doing.
- 11:32SPEAKER_00: So the chat and co work options have not rolled out to every account as yet in the US, and it's rolling out gradually.
- 11:40SPEAKER_00: I don't even know if it's rolled out internationally as yet.
- 11:43SPEAKER_00: So if you don't see just the box here, continue to toggle in between Claude and Co-Work and use it as you normally would.
- 11:50SPEAKER_00: And Frobic has said that they're rolling it out to Pro and Max users first on web, desktop, and mobile, and then the team and free accounts will follow.
- 12:00SPEAKER_00: So I have the Claude Pro Plan, and it took me about two weeks before I got this update.
- 12:06SPEAKER_00: So now that I have it, I can share it with you guys, but everyone doesn't have this update.
- 12:11SPEAKER_00: Now, if you already use co-work, your chats, your projects, your connectors, your skills, and files carry over so you have nothing to worry about.
- 12:20SPEAKER_00: Just continue to chat with interface as you see that it's being updated.
- 12:24SPEAKER_00: With a new Quan and Co-Work interface, you can continue to run your co-work tasks and schedules just the way you ran before.
- 12:32SPEAKER_00: You can just chat with it here inside of the chat box and then ask it to schedule something for you, and it will do it right inside of here.
- 12:40SPEAKER_00: Or you can always go over to the left side and you can go to scheduled and then you can schedule your tasks from there.
- 12:48SPEAKER_00: And remember, I'm on the web version.
- 12:50SPEAKER_00: The old co-work model wouldn't allow you to schedule tasks from the web.
- 12:56SPEAKER_00: You would from the claude.ai web, you would only have you would all you could only schedule those tasks from the from the Claude desktop app that you would have to download.
- 13:07SPEAKER_00: So now everything can be done on the web.
- 13:09SPEAKER_00: Pretty soon it's rolling out to mobile users.
- 13:12SPEAKER_00: So if you don't see it right now, don't worry, it's going to be rolled out to you on mobile and web in the next few days or weeks.
- 13:19SPEAKER_00: So three things that you should know before you start working inside of the new chat and co-work combined interface is that once your killed has that new experience, it cannot switch back to a separate chat and a separate co-work.
- 13:34SPEAKER_00: So before, I would be able to toggle back and forth.
- 13:38SPEAKER_00: So before I would be able to toggle back and forth between the two, but now you cannot switch back.
- 13:43SPEAKER_00: Once the task is started in the chat window, it is what it is.
- 13:47SPEAKER_00: You just continue to chat with it.
- 13:49SPEAKER_00: Now, if you remember, incognito chats, they open up in the older experience.
- 13:54SPEAKER_00: So this is the incognito button here.
- 13:57SPEAKER_00: So when you use incognito, it can't create files for you.
- 14:01SPEAKER_00: So local files, if you're using incognito mode, local files can be used in the built-in browser or on the Claude Desktop app while it's open.
- 14:13SPEAKER_00: So think about the first four things that you want to hand over to Claude and co-work first.
- 14:19SPEAKER_00: Is it your Monday or brief, your Monday report, your end-of-the-week report, a proposal deck, all your receipts that you want organized into your financial dashboard?
- 14:29SPEAKER_00: Or is it just an inbox recap, a daily inbox recap?
- 14:33SPEAKER_00: So, like I said, start on manual mode just so that you feel safe.
- 14:37SPEAKER_00: And then, as you can see, once you see how Claude is working, how it's reasoning, and how it's handling your task, and you feel safer with it, then you can put it on autopilot for some of those for auto mode for some of those tasks that you want to complete.
- 14:52SPEAKER_00: So let's move over to the third update, Sonnet 5.5.
- 14:56SPEAKER_00: Now, Sonnet 5.5 is what I primarily use when I'm doing regular tasks, everyday tasks.
- 15:03SPEAKER_00: It's always on Sonnet 5.5.
- 15:06SPEAKER_00: Let's say I'm running my Monday update, my chief of staff, I'm asking it to go into my emails, I'm asking it to draft a proposal, I'm asking it to help me create content.
- 15:17SPEAKER_00: Those are regular things that Sonnet can handle.
- 15:20SPEAKER_00: And now the 5.5 model, which launched on September 28th, is now 30% faster and cheaper than the Sonnet 5 model.
- 15:30SPEAKER_00: So this is a great update for everyone.
- 15:33SPEAKER_00: 30% faster and costing you 30% less for most of your work.
- 15:37SPEAKER_00: Anthropic Updates also scores it at a 70% 70.6% on the terminal bench 4.0 mark.
- 15:46SPEAKER_00: And that's an agentic coding evaluation compared to its previous model, Sonnet 5, which was only 10.3.
- 15:54SPEAKER_00: So this model 5.5, it's much more capable.
- 15:59SPEAKER_00: And it only scores two points below Opus 5.5 on the GPD valve.
- 16:04SPEAKER_00: So that's a test for real-world work across a variety of different occupations.
- 16:10SPEAKER_00: So you can use Sonnet.
- 16:11SPEAKER_00: You don't really have to go to Opus 5.5 for everyday tasks.
- 16:15SPEAKER_00: You can use Sonnet 5.5 for your financial task and feel comfortable and say that it will give you what you need.
- 16:21SPEAKER_00: Now, always analyze what it's giving you.
- 16:23SPEAKER_00: And if you feel you need more deep analysis, deep research, more capabilities, then you can go up to Opus 5.5.
- 16:30SPEAKER_00: Now, the other thing I like with this update is that they say that the Sonnet 5.5 model now writes more clearly than the previous generation models.
- 16:40SPEAKER_00: So I don't know if you noticed with Claude in the last two, three months, it was being very verbose and it was just talking too much.
- 16:48SPEAKER_00: And a lot of folks are getting annoyed with it because they're like, okay, just give me the answer, give it to me at a sixth grade level.
- 16:55SPEAKER_00: I don't need all of this esoteric terms, and I don't need you to keep going and explaining things.
- 17:00SPEAKER_00: So I love that sonnet 5.5 is now more concise, it's more clear, and its writing is more natural.
- 17:07SPEAKER_00: So if we go back into Claude, into the Claude interface, just toggle on to Sonnet 5.5, and there you can choose the effort level.
- 17:15SPEAKER_00: I always keep it on medium, medium is great.
- 17:18SPEAKER_00: I use medium for my helping me with my content, presentations, building PowerPoint presentations, proposals, even diagrams and small animations and things like that that I want to put inside of my presentations.
- 17:31SPEAKER_00: This model is fully capable of doing that now more than ever.
- 17:35SPEAKER_00: I've even used it to help me with video editing when I'm inside of Claude Code.
- 17:39SPEAKER_00: I'll use the Sonnet 5.5 model to start off my analysis of my content, what I wanted to cut and edit for my videos and design for prompts or pop-ups and things like that.
- 17:50SPEAKER_00: I'll start it off with a Sonnet model, and if I feel like my edits are getting more complex, then I'll put it on Opus.
- 17:56SPEAKER_00: But here's a test that you can give Sonnet right now.
- 17:59SPEAKER_00: Make sure you understand what Dunn actually looks like.
- 18:04SPEAKER_00: If the answer is yes, then you could give that job to Sonnet.
- 18:07SPEAKER_00: So you can try this prompt.
- 18:09SPEAKER_00: Let's just give it a prompt and see how you should be speaking to it.
- 18:13SPEAKER_00: So I'm gonna use Whisper Flow to talk.
- 18:16SPEAKER_00: Or you can use you can either use this little microphone here or the voice mode, which is on all the models.
- 18:22SPEAKER_00: You use my proposal template below and my call notes.
- 18:26SPEAKER_00: Fill in every section in my voice.
- 18:28SPEAKER_00: Don't invent prices or dates, list anything that I've missed, and don't send anything unless you have my permission.
- 18:36SPEAKER_00: So this is just a demo, but I'm showing you how you can prompt the models when you're using them.
- 18:41SPEAKER_00: Make sure that you're telling them that you want it to ask you for permission and not to send anything as it's working and generating for you.
- 18:51SPEAKER_00: So this is just a mock template, just for example.
- 18:55SPEAKER_00: Now, if your job needs deeper thinking, then you can move it up to Opus.
- 18:59SPEAKER_00: But if it's just a small job like this, you're analyzing a proposal and your meeting notes, your meeting transcript, then you can use Sonnet for that.
- 19:07SPEAKER_00: But if it's a smaller job, you can always move it down to haiku.
- 19:11SPEAKER_00: So that brings us to the fourth major update that Claude has released in the last four weeks.
- 19:17SPEAKER_00: And stay with me because I have more for you at the end.
- 19:19SPEAKER_00: I have resources below that you can dive into.
- 19:22SPEAKER_00: And don't forget to subscribe because I'll have more builds, more analysis, and more updates for you on all that's going on in the AI space.
- 19:31SPEAKER_00: So on October 7th, and so on October 7th, Anthropic launched Haiku 5.5, which Anthropic calls its cheapest, fastest, and most capable small model.
- 19:42SPEAKER_00: So in the three-level system of Opus being the highest, most capable model for deeper thinking, deeper judgment, deeper analysis, and then sonnet for everyday work.
- 19:55SPEAKER_00: Haiku now is almost on the same level as Sonnet 5.5.
- 19:59SPEAKER_00: Now, personally, I've never used Haiku because I didn't feel that it was capable enough for the type of work that I was doing on a day-to-day basis.
- 20:07SPEAKER_00: I would use it once in a blue moon to ask simple questions, brainstorm, get clarity, but I wouldn't use it for analysis or to build and create anything.
- 20:18SPEAKER_00: So Anthropic says Free Pro Max team and enterprise users can select it on the web, on iPhone, and in on Android.
- 20:29SPEAKER_00: And that means you can try it even on a free account.
- 20:32SPEAKER_00: So Anthrobic says to use it for quick, high volume work.
- 20:36SPEAKER_00: And as you can see here, it's only 10 cents per million input tokens, which makes this one of the cheapest models, but it is still very capable.
- 20:45SPEAKER_00: So Anthrobica says that it's Haiku is designed for quick, high volume work, such as sorting, labeling, summarizing, routing requests, live chat and support, or filling out a form and data entry.
- 21:00SPEAKER_00: So those small lower level tasks can be done so much cheaper and quickly with Haiku.
- 21:09SPEAKER_00: So let's say you're creating a chatbot and you want to select a model that can take use that chatbot to answer customer calls, customer inquiries on the phone or on the web.
- 21:22SPEAKER_00: Haiku is the cheapest model and it could do that very well without raising your voice agent bill every month.
- 21:29SPEAKER_00: So I know a lot of folks are now migrating to their voice agents to Haiku because this is exactly what it's for.
- 21:37SPEAKER_00: Those smaller, low-tiered, everyday tasks.
- 21:40SPEAKER_00: And you can see the benchmarks here of how it compares now for knowledge work and computer use.
- 21:46SPEAKER_00: Before the Haiku 4.5 model only scored 735 for knowledge work on the GDP Valve.
- 21:54SPEAKER_00: Now it scores more than double that, it scores 1620.
- 21:59SPEAKER_00: And And for computer use, it's it used to score 15.7%, but now it's scoring 72.4% on Osworld 2.0 benchmark.
- 22:11SPEAKER_00: So this is amazing.
- 22:12SPEAKER_00: You're getting 83% more capacity and use cases out of a smaller, cheaper model that you can use for everyday tasks like customer support and browser use, like lower level tasks, like basic customer support.
- 22:27SPEAKER_00: So let's say that you want to use haiku for a quick task.
- 22:31SPEAKER_00: For example, you have 10 messages in your inbox or your contact form, and you need haiku to label each message from your prospects as you need haiku to label each message from your prospects as hot, warm, or cold lead.
- 22:45SPEAKER_00: And you want it to write a one-line reply for each person.
- 22:49SPEAKER_00: And you also want it to list anything that it's not unsure about in the escalator in an escalation path for you, maybe a draft, and it won't send that message out.
- 22:59SPEAKER_00: So you could run this example through Haiku, see how it works for you, compare it to Sonnet as a customer service model, or even Opus for a more complex, detailed customer service type model.
- 23:12SPEAKER_00: And compare the speed and compare the quality.
- 23:16SPEAKER_00: For sorting and quick replies, you might find that the small model is just all that you need.
- 23:21SPEAKER_00: So when you're inside, and when you're inside the cloud interface, just go to haiku, as I showed you before, and then choose the effort that you want for that model.
- 23:30SPEAKER_00: Higher effort means more thorough responses.
- 23:33SPEAKER_00: It may take a little bit longer and just use a little bit more tokens, but with haiku, it's definitely worth it if you put it on medium or high.
- 23:41SPEAKER_00: So if you compare the smartest model to haiku, Opus 5.5 is$4 per input tokens, Sonnet 5.5 is$2 per million input tokens, and Haiku 5.5 is only 10 cents per million input tokens.
- 23:58SPEAKER_00: AI is getting so much cheaper and so much more capable.
- 24:02SPEAKER_00: And that's 40 times less for the same text, the same type of messages and queries that are going out.
- 24:09SPEAKER_00: So for your three-level model system inside of Claude, Opus 5.5, which is used for deeper judgment, deeper analytics, deeper research, more capable, complex tasks, Sonnet 5.5 for everyday execution work task, and Haiku 5.5, which is also now capable of everyday work tasks, but it's also faster and cheaper.
- 24:33SPEAKER_00: So now it's up to you to pick the model based on the type of work that you need to complete for your business.
- 24:39SPEAKER_00: And the thing to ask yourself is does this type of work need careful thinking?
- 24:46SPEAKER_00: Does it need deep research?
- 24:48SPEAKER_00: Does it need a human review?
- 24:50SPEAKER_00: And most important, what is the risk level for this type of task?
- 24:54SPEAKER_00: If it's higher risk, then I would say use Opus 5.5 because it has deeper judgment and reasoning to discern whether something may be risky that it's doing or that it's bringing back.
- 25:04SPEAKER_00: And Sonnet 5.5 is also capable of that as well for everyday tasks.
- 25:08SPEAKER_00: Before we wrap up, I want to show you the model decision rules that you should be pacing inside of your instructions for Claude or even in your team SOP or your skills so that Claude can read this before it acts.
- 25:22SPEAKER_00: So here are the decision roles to give your AI.
- 25:25SPEAKER_00: Number one, before you start, name the model that fits this type of work.
- 25:30SPEAKER_00: Number two, stop and ask before you send, pay, or delete anything.
- 25:35SPEAKER_00: Number three, stop when the data is missing.
- 25:38SPEAKER_00: Four, ensure that a human decides on clients, money, and anything permanent.
- 25:44SPEAKER_00: And you can also note in here your risk levels or your policies.
- 25:49SPEAKER_00: And number five, end every task with what you're unsure about.
- 25:53SPEAKER_00: Let me know what you're not sure about, what you're not clear on, and what you need more answers, instructions, or context on before you proceed.
- 26:01SPEAKER_00: So these are the decision rules that you can adopt for yourself, edit them, make them your own, and train your AI on these decision rules.
- 26:12SPEAKER_00: Because any fast answer that AI is giving you can look right, but can still be wrong.
- 26:16SPEAKER_00: And remember that your name and your business is on the final work.
- 26:19SPEAKER_00: So here's your one action for today.
- 26:22SPEAKER_00: Paste the decision rule that I gave you inside of your instructions in Claude or your SOP.
- 26:27SPEAKER_00: Now you can run a real life use case report through this and you'll see the difference.
- 26:33SPEAKER_00: Now, if you want to learn how to use AI in your business, implement agents safely, and upscale your team, join me inside of the AI prioritized community where you'll get step-by-step training, live implementation help, and stay on top of the best AI practices for your business.
- 26:51SPEAKER_00: 2027 is almost here and the AI landscape is changing rapidly day after day.
- 26:58SPEAKER_00: So this is how you'll stay on top of it all and ensure that your business isn't disrupted and continues to run seamlessly.
- 27:05SPEAKER_00: Thanks for watching, and I'll see you on the inside.