**The Tools that Great Engineers Use**

When it comes to great engineers, there are certain tools that stand out as essential to their workflow. For many, these tools include text editors and Bash - the go-to programming language for automation and scripting tasks. These two tools provide a solid foundation for engineers to work efficiently and effectively, with Bash offering a powerful set of features for automating repetitive tasks and text editors providing a clean and intuitive interface for writing code.

**The Real World of Engineering**

While flashy models may catch our attention, it's the real-world applications of these technologies that truly matter. For great engineers, their computer tool is often not just about being cool or visually appealing, but about being reliable and efficient in its performance. In this sense, tools like RPA (Robotic Process Automation) play a crucial role in automating tasks and freeing up human engineers to focus on higher-level creative work.

**Vendor Lock-in**

As these technologies continue to evolve, one trend that's becoming increasingly evident is vendor lock-in. This means that companies are investing heavily in proprietary models and tools, making it difficult for users to switch or adapt to new solutions. While this can be a barrier to entry for smaller teams or individuals, it also highlights the importance of being aware of these trends and preparing accordingly.

**Claude 3.5 Upgraded to CLA 3.5 New**

The latest upgrade from Claude 3.5 to CLA 3.5 new (or CLA 3.5 Sonic) has generated significant excitement among engineers and users alike. This upgraded model promises improved performance, speed, and accuracy, making it a powerful tool for generating code and text-based content.

**Opus Removal from Documentation**

One interesting development is the sudden removal of Opus 3.5 from Claude's documentation pages. While this may seem like a minor issue, it raises questions about what's happening with this model and why it was removed. Is it being phased out in favor of newer models? Was it simply deleted due to lack of interest or funding issues?

**Predictions on CLA 3.5 Opus**

Several predictions have emerged regarding the disappearance of CLA 3.5 Opus from Claude's documentation pages. One possibility is that it was rebranded as CLA 3.5 Sonic, with some speculating that this model has undergone significant improvements but may not perform exactly as expected. Another theory suggests that CLA 3.5 Opus was simply deleted due to strategic decisions made by the company.

**Anthropic's Decision-making Process**

Given Anthropic's track record of innovation and their position as leaders in AI model development, it's unlikely that they would remove a powerful model like CLA 3.5 Opus without a good reason. One possible explanation is that funding or compute issues led to its deletion, while another theory suggests that the company may be focusing on one powerful model (CLA 3.5 Sonic) and discarding others in favor of it.

**The Future of AI Agents**

As we move forward into 2025, one trend that's clear is the growing importance of AI agents and their role in our development workflows. These tools offer unprecedented levels of automation and personalization, allowing developers to focus on higher-level tasks and deliver more effective results.

**Staying Focused and Ahead of the Curve**

For engineers looking to capitalize on this emerging landscape, it's essential to stay focused and up-to-date with the latest developments in AI model technology. By doing so, we can harness the full potential of these tools and unlock new levels of efficiency, productivity, and innovation in our work.

**Conclusion**

In conclusion, the tools that great engineers use are critical components of their workflow. From text editors and Bash to RPA and AI agents, each tool plays a vital role in enabling efficient and effective development. As we move forward into 2025, it's clear that AI agents will continue to play an increasingly important role in our development workflows.

**Final Thoughts**

The disappearance of CLA 3.5 Opus from Claude's documentation pages raises questions about what's happening with this model and why it was removed. While speculating on the reasons behind this decision may be entertaining, it's essential to focus on the bigger picture - the future of AI agents and their role in our development workflows.

Let us know in the comments where you think CLA 3.5 Opus has gone, and what you're most excited about when it comes to the latest developments in AI model technology.

Stay tuned for more insights into the world of generative AI, as we explore its potential applications, challenges, and implications for developers and engineers alike.

"WEBVTTKind: captionsLanguage: enif you want to win any generative AI age you must have a vision with every release it's 100% clear anthropic has a vision so much Vision that they gave it to our computers with the brand new computer use tools by now you've already seen how the computer tool Works running on the new claw 3.5 Sonet let's go deeper and explore the less disgusted tools Engineers are sleeping on because it's not flashy and it's not cool at first glance but make no mistake the text editor and The Bash tool are critical infrastructure for asymmetric software engineering in the age of generative AI why because Engineers that fully control their terminal and their editor can ingest and synthesize information like no other I've wired up a proof of concept of these tools and have started putting them to work already let's walk through productivity gains you can get using these tools today and then we'll speculate on the question where is Claude 3.5 Opus I don't know if you've noticed but any mention of claw 3.5 Opus has been removed from anthropics Models page Why has anthropic completely removed any mention of it from their documentation so in this proof of concept you can see I have three classes here we have a logger we have the editor class and then we have the bash session let's go ahead and start start with anthropics new bash tools in the terminal I'm using astrals UV python package manager so I'll type UV run Main and then I'll specify my prompt so let's just start with something simple and work our way up I'll say list all files in current D tree form depth 2 and I'll use the mode bash so before we run this we'll open up the file explorer so you can just see a couple of files we have in this directory and now let's go ahead and run our tool so we have a bunch of information here let's go ahead and review the log if you open up sessions you can see a couple previous sessions here and now you can see our logging for this run so this is really interesting right we have our bash command logging everything that's happening and you can see our output for the tool run right so output for tree- L2 and remember I asked for list all files in the current dur tree form depth 2 and that's exactly what we got we open up AI docs we have those files exactly and of course we have read me we have backup we have an app database that we'll look at here in a second we have our editor dur and then we have our images and of course we have these session dispatch tool converted our natural language command into the correct bash command let's go ahead and push this further let's clear and run another command so let's just add on to it right so we're on UV run main list all files Tre form depth 2 so just running that same prompt and then we'll say first five lines of the newest log file and let's make sure we get our D- mode bash so let's kick this off open up our file explorer you can see okay so let's check out our log let's see what we have now so you can see same as before or tree command Rand just as before right so we can see here the new claw 3.5 model is consistent that's good to see and now if we move down to our second command we can see here that plot is picking up on this command here right so it's saying here let's go ahead and collapse this now let's find the newest log file and read the first five lines you can see here there's some type of Chain of Thought some type of iterative process happening within claud's thought process it's thinking through what it needs to do to accomplish the next task we've given it now it's referring to the files in the sessions directory it's looking at our newest file and interestingly right the newest file is the file that's getting logged actively so it's getting a little meta here but what we end up with here is Claude properly printing out you can see here here's the actual run CLA is properly printing out its previous brand new messages and if we go ahead and search right first let's list all the files in tree form and scroll up you can see that line exactly I'm running commands in natural language and not only is Claude getting the order of the commands properly it's getting all the commands right you can see here we have our pricing calculations set up here so far no hiccups really impressed things are looking really good we are running arbitrary bash commands with a brand new bash tool let's just keep pushing this tool further so inside of this codebase I have this data app DB and this is a simple SQL light database with a single table inside of it so we can go ahead and see exactly what's in this SQL light database with 3 we'll pass the path to this and then we'll say schema so you can see here we have a simple logging table so let's have the cloud bash computers tool modify this database let's go ahead and reuse the same command and I'll say using SQ light 3 update existing data/ app. DB read schemas then insert these five lines in our SQL like database logging table okay so I'm giving it all the information it needs to complete this request without going through any of the Motions of actually writing these commands out myself so let's see how well this does here we'll fire this off and we'll load up that log file you can see it's running that tree command again let's go ah and get a refresh here you can see it is looking at the definitions right so this it's automatically looking at that right it ran that schema method and fantastic and it selected back out the rows here so you can see a summary here at the end of the log I have completed all the requested tasks listed files in tree form with depth two found and display the first five lines read the schema inserted five lines into the logging table and then verified the insertions okay you can see the pricing here that cost us a total of 4 cents mostly input costs right you can see all the input tokens from ingesting the relevant information to make the commands right we can search our output for observability is ultra important when you're working working with these tools I always like to start with a simple loging file and then progress from there we can just search these right you can see we have a total of four outputs tree- L2 let's go ahead and look at that next command ls-t sessions. log tar agent has read this content and now it's checking our sqlite database schemas right so you can see here executing bash command it's first reading in the database structure just like we asked and then if we scroll down you can see here now it's going to actually execute this statement this is not a trivial bash statement writing this command and then it's doing some wonderful said string parsing automatically for us grabbing those lines placing it inside our database then Claude is going back in and verifying that these items are in there right so we're saying select star from logging order by ID descending limit five and it's getting those items right back out and you know we can manually just go in and run this exact same command to verify because because you know with great logging again observability is really important you can always just backtrack and see what your agents and see what your prompts are doing so we're just going to literally copy paste to send and you can see there we're getting our new previous logging messages so this is really cool right if you want to you can also run this entire chain again with this-- no AGI flag and all this is going to do is it'll run and instead of actually executing the subprocess commands it will just return a mock response and that'll be fed back into the AI agent obviously it won't completely work because it's not running the real commands and it's not getting the real results but you know just a test you can pass in this flag if you want so that you know Claude doesn't fully take over your computer this is pretty incredible stuff we have a simple Loop giving Claude the space to call tools get feedback from the tool and then rerun the next command right so it's a really simple process of letting Claude in just a prompt and then it runs the llm decides what tool to run it passes us the tool we execute the tool and then we give the response back to Claude and it just repeats in that Loop and if you were to build a real application or a real agent around this what you would actually do is after you finish an execution you would simply create some type of human in the loop feedback the simplest version of this is to run a while loop or set up a personal AI assistant to control the history and to interact with you with voice commands right so that this Loop would continue you can also set up user interfaces to interact with this or any other agent so as always let's start simple you can see we have our editor directory here with this ping pong. text I'll say UV run Main and I'll pass in the prompt update pong to Ping in our ping pong. txt file okay and we'll run this the default mode is text so we can just kick this off you can see we're in file editor mode here and we should be able to see that value Kate update to Ping so that just happened in real time um you know we didn't do or say anything let's go ahead and just you know play with this a little bit add 30 more pings 30 pongs ping pong. update okay so this is like the dumbest simplest thing we can do with this file but it's a good place to start right so we're just literally saying update this file in some silly arbitrary way okay great so now we have you know 30 ping 30 pong fantastic so let's do something more legitimate with this tool right we'll run UV run main write a detailed three use case document for llms to a LM use cases markdown then into three going into detail about each use case so here we have two highle commands that we want done right we want an original use case Doc and then we want Claud to break it up into three additional files so let's kick that off eyes on the editor directory here as Cloud's going to be generating these files for us okay you can see LM use case here we can hop into this and get a nice breakdown view here right content generation code development programming data analysis you can see content generation now has its own file very very nice right that is one of the goto use cases for llms and of course code development this is a big one and all all these documents are happening they're getting generated on the fly live and here's our data analysis that was really incredible let's go ahead and backtrack you can see the cost here that cost around 8 cents to fire off and a lot of this is the input tokens right after every tool call we're always passing back the response from the tool back into Claud right so that's going to see our our cost stack up over time let's go ahead and just break down the exact calls so if we search command you can see all the commands that Claud is running we should just have for create command and that's exactly what we have right so for create commands you can see all the content for every single creation statement you can see the exact path right we're always getting back this SL repo SL file and the SL repo is what you'll update or replace to be the true path that you're writing to so we update and replace that to be our editor dur that's configurable via an environment variable and you can see just like the bash tool the file editor tool is just doing all of this work for us right so you can imagine you're passing in a lot more important information and you're treating this tool like an agent I hope it's becoming really really clear what's happening with every video release we have on the channel every Monday we're getting closer to having these pieces of software that are operating for us and with this release anthropic is pushing us into the future with these opinionated domain specific set of tools to solve a specific set of problems now I hope that sounds familiar because it is we've been talking about on the channel when you canst prompts around a specific ific use case and you compose those tools that your prompt can call what do you get you get an AI agent this is where everything's going this is what's coming next AI agents are here in 2025 they're going mainstream you're here early we're talking about this early get your hands on these tools and start utilizing these we can now craft arbitrary prompts pass any information we want to and instead of a text to text prompt what we're getting now is text to action prompt these are text to action prompts with bash commands and with file editing and with computer use our llms are starting to control things they're starting to manipulate things create read update delete they're modifying information they're running concrete commands that allow them to do more now my llm can operate on all of my files with and for me this is a huge huge deal this is why I wanted to emphasize these two tools because it's a great way to work up to where everything is going as an engineer the file editor and the terminal are critical tools for us and if you have agents operating in your terminal and your file editor for you in parallel to you you're going to get asymmetric Returns on your time so this is awesome let's go ahead and make a couple tweaks and just show off a couple other things you can do with this be run Main and we'll say read llm use cases right so we'll have it fire off it's read functionality then update to contain a mermaid diagram of the use cases and we'll say LR flow chart okay now we're going to read this file in right so this is an existing file it already exists and it's going to modify you can see there that modification came right in automatically we now have a flowchart right it ingested the file it reread the file and then it made a modification to it right if we go into preview mode here we can see our mermaid chart so now we're rapidly creating this documentation this markdown file you know this plan whatever you're working on you now have these agents to help you operate on it very very quickly it's so important to be wiring these tools up and to be keeping track of what's important this is an important tool the file editing tool and The Bash tool are Ultra important I'm going to be wiring these up in my CLI in Ada the personal AI assistant for engineering we've been building on the channel I'm going to be using and deploying these tools every single day so this code base is is going to be in the description for you to check out and to get yourself up and running with your text editor and your bash AI agents I want to hop back to the high level here and just discuss where things are going so you might have noticed these are just tool calls right there's actually nothing super sophisticated going on here at least not at first site right not at first glance what they're doing here is actually really interesting the value composition of these tools is that claw 3.5 Sonet has been fine-tuned to operate these tools very well what we're seeing with this is an interesting direction for model providers like open aai Google Facebook and anthropic to release sets of tools to perform specific tasks very well with a model fine tune to excel at their tools and I hope this sounds familiar these tool belts built to solve a specific set of problems s are effectively pre-baked AI agents we dove into the composability of prompts into AI agents in our previous video and it looks like we're right on track this is a great direction from a business perspective because now anthropic can build models and sets of tools that perform well on specific benchmarks and sell that tool belt as an AI agent so these companies are starting to turn all their work with llms into concrete value into concrete output they're saying fantastic we've built these next generation tools let's actually solve real problems with them and they're starting at a great place they're starting at the bottom level of where all software problems are solved the engineers tool belt these tools stood out to me right away for that reason text editor and Bash this is where Engineers live this is where great engineers get tons of work done quickly their computer tool obviously very flashy very cool great for RPA but the real world work for engineering happens with these two tools as a side note these companies are trying to create vendor lockin with their specific sets of tools and they're you know fine-tuned models that are built to use those tools well so keep your eyes on releases like this this is a trade-off and it's a risk we just kind of have to accept and the generative AI age because these companies are leading the way for what's possible this is fantastic I'm loving Claude 3.5 upgraded or clae 3.5 new definitely could named that a little bit better but that's old news what is interesting is the question where is Opus where is claw 3.5 Opus I always keep a close eye on the big three model pages and AMA to you know just kind of see where the models are what's going on and also I've referenced these documents a lot one thing I noticed right away during this upgrade and one of my favorite Engineers Simon Willison also o pointed out claw 3.5 Opus has been completely removed from their documentation what is going on why have they done this is Opus 3.5 off the table I have a couple ideas as to why anthropic removed it let's just walk through them and I want to get your thoughts on what you think is going on here so my first kind of prediction is that claw 3.5 Sonet is claw 3.5 Opus and they've just rebranded it although this model is a massive Improvement I don't think it perform forms like we would expect an opus 3.52 so they just rebranded it to claw 3.5 in a panic and you know this is not the most glamorous case you don't want a Top Model provider making strategic decisions like this but it's an interesting prediction I think this is a probably mid to low probability so you know maybe 10 to 30% possible the other prediction is that they just deleted it on purpose right they just got rid of it claw 3.5 is a great model with great speed great performance it follows instructions better than any model it writes the best and minus 01 preview it it's also the best for coding more on AI coding with these models and future videos Cloud 3.5 basically wins in every category this isn't really interesting case if they just simply delete it but then we have to ask the next logical question why did they delete it right is it a funding issue do they not have the compute or is it again is it just a positive strategic decision are are they just FOC focusing on one powerful model and one fast model right which is going to be 3.5 ha cou I don't really see anthropic being short on ideas and Innovation to push on and it seems more likely that it's just a decision that they made more than a funding or compute issue because they're firing on all cylinders right all public sources State anthropic performing really really well so the last case and this is the prediction that's the most grounded in reality is that they simply remove the documentation because it's just not ready yet I put this at like mid-tier probability they just need more time with it they're expanding it they're adding pieces of Chain of Thought or you know next level prompt chaining inside the model who really knows only they really know what's coming next on the Innovation side of things they're one of three leaders in model Innovation so only they really could know what's coming next to be fair though CLA 3.5 Sonic is an incredible model especially after the upgrade so I'm not super worried about where 3.5 Opus is and if it comes out I'll be happy to see it if they move on from it and just go right to Cloud 4 that's fine too let me know what you think in the comments where do you think Claude 3.5 Opus has gone why is it not on their model page anymore so I hope you can see this theme AI agents are here in 2025 they're growing mainstream if you're here if you're watching this video you're here early it's time to capitalize as Engineers the question we must ask ourselves now is how quickly how much and how often are we deploying compute in our development workflows right now this takes the form of prompts AI agents and personal AI assistants these are the big three you need to be focusing your time and effort on if you want those asymmetric results follow the channel to stay locked in to what you can do with this incredible generative AI technology available to you stay focused and keep buildingif you want to win any generative AI age you must have a vision with every release it's 100% clear anthropic has a vision so much Vision that they gave it to our computers with the brand new computer use tools by now you've already seen how the computer tool Works running on the new claw 3.5 Sonet let's go deeper and explore the less disgusted tools Engineers are sleeping on because it's not flashy and it's not cool at first glance but make no mistake the text editor and The Bash tool are critical infrastructure for asymmetric software engineering in the age of generative AI why because Engineers that fully control their terminal and their editor can ingest and synthesize information like no other I've wired up a proof of concept of these tools and have started putting them to work already let's walk through productivity gains you can get using these tools today and then we'll speculate on the question where is Claude 3.5 Opus I don't know if you've noticed but any mention of claw 3.5 Opus has been removed from anthropics Models page Why has anthropic completely removed any mention of it from their documentation so in this proof of concept you can see I have three classes here we have a logger we have the editor class and then we have the bash session let's go ahead and start start with anthropics new bash tools in the terminal I'm using astrals UV python package manager so I'll type UV run Main and then I'll specify my prompt so let's just start with something simple and work our way up I'll say list all files in current D tree form depth 2 and I'll use the mode bash so before we run this we'll open up the file explorer so you can just see a couple of files we have in this directory and now let's go ahead and run our tool so we have a bunch of information here let's go ahead and review the log if you open up sessions you can see a couple previous sessions here and now you can see our logging for this run so this is really interesting right we have our bash command logging everything that's happening and you can see our output for the tool run right so output for tree- L2 and remember I asked for list all files in the current dur tree form depth 2 and that's exactly what we got we open up AI docs we have those files exactly and of course we have read me we have backup we have an app database that we'll look at here in a second we have our editor dur and then we have our images and of course we have these session dispatch tool converted our natural language command into the correct bash command let's go ahead and push this further let's clear and run another command so let's just add on to it right so we're on UV run main list all files Tre form depth 2 so just running that same prompt and then we'll say first five lines of the newest log file and let's make sure we get our D- mode bash so let's kick this off open up our file explorer you can see okay so let's check out our log let's see what we have now so you can see same as before or tree command Rand just as before right so we can see here the new claw 3.5 model is consistent that's good to see and now if we move down to our second command we can see here that plot is picking up on this command here right so it's saying here let's go ahead and collapse this now let's find the newest log file and read the first five lines you can see here there's some type of Chain of Thought some type of iterative process happening within claud's thought process it's thinking through what it needs to do to accomplish the next task we've given it now it's referring to the files in the sessions directory it's looking at our newest file and interestingly right the newest file is the file that's getting logged actively so it's getting a little meta here but what we end up with here is Claude properly printing out you can see here here's the actual run CLA is properly printing out its previous brand new messages and if we go ahead and search right first let's list all the files in tree form and scroll up you can see that line exactly I'm running commands in natural language and not only is Claude getting the order of the commands properly it's getting all the commands right you can see here we have our pricing calculations set up here so far no hiccups really impressed things are looking really good we are running arbitrary bash commands with a brand new bash tool let's just keep pushing this tool further so inside of this codebase I have this data app DB and this is a simple SQL light database with a single table inside of it so we can go ahead and see exactly what's in this SQL light database with 3 we'll pass the path to this and then we'll say schema so you can see here we have a simple logging table so let's have the cloud bash computers tool modify this database let's go ahead and reuse the same command and I'll say using SQ light 3 update existing data/ app. DB read schemas then insert these five lines in our SQL like database logging table okay so I'm giving it all the information it needs to complete this request without going through any of the Motions of actually writing these commands out myself so let's see how well this does here we'll fire this off and we'll load up that log file you can see it's running that tree command again let's go ah and get a refresh here you can see it is looking at the definitions right so this it's automatically looking at that right it ran that schema method and fantastic and it selected back out the rows here so you can see a summary here at the end of the log I have completed all the requested tasks listed files in tree form with depth two found and display the first five lines read the schema inserted five lines into the logging table and then verified the insertions okay you can see the pricing here that cost us a total of 4 cents mostly input costs right you can see all the input tokens from ingesting the relevant information to make the commands right we can search our output for observability is ultra important when you're working working with these tools I always like to start with a simple loging file and then progress from there we can just search these right you can see we have a total of four outputs tree- L2 let's go ahead and look at that next command ls-t sessions. log tar agent has read this content and now it's checking our sqlite database schemas right so you can see here executing bash command it's first reading in the database structure just like we asked and then if we scroll down you can see here now it's going to actually execute this statement this is not a trivial bash statement writing this command and then it's doing some wonderful said string parsing automatically for us grabbing those lines placing it inside our database then Claude is going back in and verifying that these items are in there right so we're saying select star from logging order by ID descending limit five and it's getting those items right back out and you know we can manually just go in and run this exact same command to verify because because you know with great logging again observability is really important you can always just backtrack and see what your agents and see what your prompts are doing so we're just going to literally copy paste to send and you can see there we're getting our new previous logging messages so this is really cool right if you want to you can also run this entire chain again with this-- no AGI flag and all this is going to do is it'll run and instead of actually executing the subprocess commands it will just return a mock response and that'll be fed back into the AI agent obviously it won't completely work because it's not running the real commands and it's not getting the real results but you know just a test you can pass in this flag if you want so that you know Claude doesn't fully take over your computer this is pretty incredible stuff we have a simple Loop giving Claude the space to call tools get feedback from the tool and then rerun the next command right so it's a really simple process of letting Claude in just a prompt and then it runs the llm decides what tool to run it passes us the tool we execute the tool and then we give the response back to Claude and it just repeats in that Loop and if you were to build a real application or a real agent around this what you would actually do is after you finish an execution you would simply create some type of human in the loop feedback the simplest version of this is to run a while loop or set up a personal AI assistant to control the history and to interact with you with voice commands right so that this Loop would continue you can also set up user interfaces to interact with this or any other agent so as always let's start simple you can see we have our editor directory here with this ping pong. text I'll say UV run Main and I'll pass in the prompt update pong to Ping in our ping pong. txt file okay and we'll run this the default mode is text so we can just kick this off you can see we're in file editor mode here and we should be able to see that value Kate update to Ping so that just happened in real time um you know we didn't do or say anything let's go ahead and just you know play with this a little bit add 30 more pings 30 pongs ping pong. update okay so this is like the dumbest simplest thing we can do with this file but it's a good place to start right so we're just literally saying update this file in some silly arbitrary way okay great so now we have you know 30 ping 30 pong fantastic so let's do something more legitimate with this tool right we'll run UV run main write a detailed three use case document for llms to a LM use cases markdown then into three going into detail about each use case so here we have two highle commands that we want done right we want an original use case Doc and then we want Claud to break it up into three additional files so let's kick that off eyes on the editor directory here as Cloud's going to be generating these files for us okay you can see LM use case here we can hop into this and get a nice breakdown view here right content generation code development programming data analysis you can see content generation now has its own file very very nice right that is one of the goto use cases for llms and of course code development this is a big one and all all these documents are happening they're getting generated on the fly live and here's our data analysis that was really incredible let's go ahead and backtrack you can see the cost here that cost around 8 cents to fire off and a lot of this is the input tokens right after every tool call we're always passing back the response from the tool back into Claud right so that's going to see our our cost stack up over time let's go ahead and just break down the exact calls so if we search command you can see all the commands that Claud is running we should just have for create command and that's exactly what we have right so for create commands you can see all the content for every single creation statement you can see the exact path right we're always getting back this SL repo SL file and the SL repo is what you'll update or replace to be the true path that you're writing to so we update and replace that to be our editor dur that's configurable via an environment variable and you can see just like the bash tool the file editor tool is just doing all of this work for us right so you can imagine you're passing in a lot more important information and you're treating this tool like an agent I hope it's becoming really really clear what's happening with every video release we have on the channel every Monday we're getting closer to having these pieces of software that are operating for us and with this release anthropic is pushing us into the future with these opinionated domain specific set of tools to solve a specific set of problems now I hope that sounds familiar because it is we've been talking about on the channel when you canst prompts around a specific ific use case and you compose those tools that your prompt can call what do you get you get an AI agent this is where everything's going this is what's coming next AI agents are here in 2025 they're going mainstream you're here early we're talking about this early get your hands on these tools and start utilizing these we can now craft arbitrary prompts pass any information we want to and instead of a text to text prompt what we're getting now is text to action prompt these are text to action prompts with bash commands and with file editing and with computer use our llms are starting to control things they're starting to manipulate things create read update delete they're modifying information they're running concrete commands that allow them to do more now my llm can operate on all of my files with and for me this is a huge huge deal this is why I wanted to emphasize these two tools because it's a great way to work up to where everything is going as an engineer the file editor and the terminal are critical tools for us and if you have agents operating in your terminal and your file editor for you in parallel to you you're going to get asymmetric Returns on your time so this is awesome let's go ahead and make a couple tweaks and just show off a couple other things you can do with this be run Main and we'll say read llm use cases right so we'll have it fire off it's read functionality then update to contain a mermaid diagram of the use cases and we'll say LR flow chart okay now we're going to read this file in right so this is an existing file it already exists and it's going to modify you can see there that modification came right in automatically we now have a flowchart right it ingested the file it reread the file and then it made a modification to it right if we go into preview mode here we can see our mermaid chart so now we're rapidly creating this documentation this markdown file you know this plan whatever you're working on you now have these agents to help you operate on it very very quickly it's so important to be wiring these tools up and to be keeping track of what's important this is an important tool the file editing tool and The Bash tool are Ultra important I'm going to be wiring these up in my CLI in Ada the personal AI assistant for engineering we've been building on the channel I'm going to be using and deploying these tools every single day so this code base is is going to be in the description for you to check out and to get yourself up and running with your text editor and your bash AI agents I want to hop back to the high level here and just discuss where things are going so you might have noticed these are just tool calls right there's actually nothing super sophisticated going on here at least not at first site right not at first glance what they're doing here is actually really interesting the value composition of these tools is that claw 3.5 Sonet has been fine-tuned to operate these tools very well what we're seeing with this is an interesting direction for model providers like open aai Google Facebook and anthropic to release sets of tools to perform specific tasks very well with a model fine tune to excel at their tools and I hope this sounds familiar these tool belts built to solve a specific set of problems s are effectively pre-baked AI agents we dove into the composability of prompts into AI agents in our previous video and it looks like we're right on track this is a great direction from a business perspective because now anthropic can build models and sets of tools that perform well on specific benchmarks and sell that tool belt as an AI agent so these companies are starting to turn all their work with llms into concrete value into concrete output they're saying fantastic we've built these next generation tools let's actually solve real problems with them and they're starting at a great place they're starting at the bottom level of where all software problems are solved the engineers tool belt these tools stood out to me right away for that reason text editor and Bash this is where Engineers live this is where great engineers get tons of work done quickly their computer tool obviously very flashy very cool great for RPA but the real world work for engineering happens with these two tools as a side note these companies are trying to create vendor lockin with their specific sets of tools and they're you know fine-tuned models that are built to use those tools well so keep your eyes on releases like this this is a trade-off and it's a risk we just kind of have to accept and the generative AI age because these companies are leading the way for what's possible this is fantastic I'm loving Claude 3.5 upgraded or clae 3.5 new definitely could named that a little bit better but that's old news what is interesting is the question where is Opus where is claw 3.5 Opus I always keep a close eye on the big three model pages and AMA to you know just kind of see where the models are what's going on and also I've referenced these documents a lot one thing I noticed right away during this upgrade and one of my favorite Engineers Simon Willison also o pointed out claw 3.5 Opus has been completely removed from their documentation what is going on why have they done this is Opus 3.5 off the table I have a couple ideas as to why anthropic removed it let's just walk through them and I want to get your thoughts on what you think is going on here so my first kind of prediction is that claw 3.5 Sonet is claw 3.5 Opus and they've just rebranded it although this model is a massive Improvement I don't think it perform forms like we would expect an opus 3.52 so they just rebranded it to claw 3.5 in a panic and you know this is not the most glamorous case you don't want a Top Model provider making strategic decisions like this but it's an interesting prediction I think this is a probably mid to low probability so you know maybe 10 to 30% possible the other prediction is that they just deleted it on purpose right they just got rid of it claw 3.5 is a great model with great speed great performance it follows instructions better than any model it writes the best and minus 01 preview it it's also the best for coding more on AI coding with these models and future videos Cloud 3.5 basically wins in every category this isn't really interesting case if they just simply delete it but then we have to ask the next logical question why did they delete it right is it a funding issue do they not have the compute or is it again is it just a positive strategic decision are are they just FOC focusing on one powerful model and one fast model right which is going to be 3.5 ha cou I don't really see anthropic being short on ideas and Innovation to push on and it seems more likely that it's just a decision that they made more than a funding or compute issue because they're firing on all cylinders right all public sources State anthropic performing really really well so the last case and this is the prediction that's the most grounded in reality is that they simply remove the documentation because it's just not ready yet I put this at like mid-tier probability they just need more time with it they're expanding it they're adding pieces of Chain of Thought or you know next level prompt chaining inside the model who really knows only they really know what's coming next on the Innovation side of things they're one of three leaders in model Innovation so only they really could know what's coming next to be fair though CLA 3.5 Sonic is an incredible model especially after the upgrade so I'm not super worried about where 3.5 Opus is and if it comes out I'll be happy to see it if they move on from it and just go right to Cloud 4 that's fine too let me know what you think in the comments where do you think Claude 3.5 Opus has gone why is it not on their model page anymore so I hope you can see this theme AI agents are here in 2025 they're growing mainstream if you're here if you're watching this video you're here early it's time to capitalize as Engineers the question we must ask ourselves now is how quickly how much and how often are we deploying compute in our development workflows right now this takes the form of prompts AI agents and personal AI assistants these are the big three you need to be focusing your time and effort on if you want those asymmetric results follow the channel to stay locked in to what you can do with this incredible generative AI technology available to you stay focused and keep building\n"