Dark | Light
[GUEST ACCESS MODE: Data is scrambled or limited to provide examples. Make requests using your API key to unlock full data. Check https://lunarcrush.ai/auth for authentication information.]

[@levie](/creator/twitter/levie)
"Another great win for AI agent interoperability. Box is partnering with Google to make Box AI Agents accessible via Google Gemini Enterprise so you can easily work with your enterprise content from anywhere. This is why the future of AI looks like"  
[X Link](https://x.com/levie/status/1976360528256852130) [@levie](/creator/x/levie) 2025-10-09T18:54Z 2.5M followers, 72K engagements


"This is actually extremely pragmatic and realistic from @karpathy based on what is likely to happen especially in an enterprise context. We have rapidly improving AI model capabilities but the diffusion of these capabilities into real life workflows will take time and require lots of integration change management and new solutions that must be built. Basically my AI timelines are about 5-10X pessimistic w.r.t. what you'll find in your neighborhood SF AI house party or on your twitter timeline but still quite optimistic w.r.t. a rising tide of AI deniers and skeptics. The apparent conflict is"  
[X Link](https://x.com/levie/status/1979648582224658538) [@levie](/creator/x/levie) 2025-10-18T20:39Z 2.5M followers, 161.8K engagements


"Having talked to hundreds of IT leaders over the last year alone its clear we have a capability overhang where the current AI models are already very good at solving many problems that havent been adopted yet. The biggest hurdles generally are the imagination for whats now possible the sheer speed at which the tech is changing and the change management to make it happen. In tech its natural to experiment with lots of solutions and do a lot of work to make something work. But outside of tech other than for a small percentage of the largest companies these capabilities need to be packaged as"  
[X Link](https://x.com/levie/status/1979312435694834048) [@levie](/creator/x/levie) 2025-10-17T22:23Z 2.5M followers, 78.3K engagements


"OpenAI just launched Atlas a browser that has AI agents as a native capability. Now you can let browser agents take over your software and automate anything. Here's the agent using Box to automate folder creation for an M&A due diligence process (sped up)"  
[X Link](https://x.com/levie/status/1980710354016669751) [@levie](/creator/x/levie) 2025-10-21T18:58Z 2.5M followers, 346.9K engagements


"Software engineers are the translation layer between the market and the computer. In a world where AI agents will generate 10X or 100X more code than humans this translation layer will be more important than ever. With AI agents dramatically lowering the raw cost of code production the market will decide that it wants even more software built. We will build far more digital experiences in far more industries we will upgrade systems that were once forgotten about we will add more features to every piece of software (like it or not) and generally say yes to building way more. In this world"  
[X Link](https://x.com/levie/status/1977875995484184777) [@levie](/creator/x/levie) 2025-10-13T23:15Z 2.5M followers, 101.3K engagements


"Were going to enter a new phase of AI where the average consumer may not perceive a huge difference in any one model upgrade but specific industries will feel significant improvements occur. The next phase of AI be about making models better and better for AI agents in real world knowledge work domains. Were going to see new evals emerge that test for relevant tasks in these spaces the generation of high quality data and environments that align to real work training against specific tools and software in these spaces and more. The test case of how to improve models on axis of real world"  
[X Link](https://x.com/levie/status/1980856705027305835) [@levie](/creator/x/levie) 2025-10-22T04:40Z 2.5M followers, 84.6K engagements


"Anyone who has been working on AI agents especially in an enterprise context would have been nodding along to Karpathys podcast the entire time. So much has yet to get built out to make it all work at scale. The analogy of self driving cars is especially apt each additional X of reliability takes a massive amount of work. XX% XX% XXXX% are all new levels of complexity that tend to require new breakthroughs more data and so on. The one benefit of AI agents for knowledge work in contrast to self-driving cars though is that you can decide to get value at any point along the way by just tweaking"  
[X Link](https://x.com/levie/status/1980435175160050092) [@levie](/creator/x/levie) 2025-10-21T00:45Z 2.5M followers, 102.7K engagements


"Designing AI agent interaction patterns is a completely wide open space right now. If you use X different tools that do agentic tasks youll see X totally different interaction patterns. Some show all the steps that are being taken by the agent. Some show the thinking tokens. Some show the core thats being executed or computer thats being used. Some show the planning happening. Some let you edit the planning. Some obfuscate this all. Then theres all the UI and interaction patterns for how to work with agent once its produced some output. How do you incorporate what its done into something"  
[X Link](https://x.com/levie/status/1981190193718972802) [@levie](/creator/x/levie) 2025-10-23T02:45Z 2.5M followers, 44.1K engagements


"Did something change in robotics like exactly one month ago to make all of this possible. This is crazy"  
[X Link](https://x.com/levie/status/1902817611588501964) [@levie](/creator/x/levie) 2025-03-20T20:20Z 2.5M followers, 917.3K engagements


"Product management for AI agents is easily the wildest form of product management in history. Typical product management is trying to figure out how to design interfaces and software for people to interact with deterministic systems. The user generally knows all the context to do their work successfully so its generally a matter of nailing the underlying business logic and surrounding UX. But with AI agents the user you care about most is the agent and they dont know anything by default. Theyll happily run in any direction to perform the task often without success. So as a PM (or engineer)"  
[X Link](https://x.com/levie/status/1961884479427080255) [@levie](/creator/x/levie) 2025-08-30T20:11Z 2.5M followers, 248K engagements


"@sama Just sent this XX minutes ago after a cursor session with codex"  
[X Link](https://x.com/levie/status/1977180326356296069) [@levie](/creator/x/levie) 2025-10-12T01:11Z 2.5M followers, 36.8K engagements


"Designing for AI agent interaction patterns is such a fascinating process. In theory it should be as simple as designing the interaction that you would want to have with another person that was doing work for you. A big difference however is it costs this worker nothing to show all the work its doing along the way or how it came to its final result which is a huge asset. And at the same time you generally want the confidence that its doing the right thing or you may need to tilt it in a different direction if it goes off the rails. In the era of automation and robots seeing more context about"  
[X Link](https://x.com/levie/status/1981223827440116037) [@levie](/creator/x/levie) 2025-10-23T04:59Z 2.5M followers, 4568 engagements


"The next big breakthroughin AI is AI Agents.This is when AI goes from being used as an assistant to chat with to using AI to accomplish complete tasks that a human might otherwise have to perform. This moves AI from being a "read-only" operation to fundamentally a "read/write" operation.Ultimately this brings us much closer to the full promise of AI in particularin the enterprise where AI can begin to complete any part of a workflow and we're already seeing examples today of Agents that write entire software applications or respond tocustomersupport tickets. Today in many ways Agents are"  
[X Link](https://x.com/levie/status/1782164161553961069) [@levie](/creator/x/levie) 2024-04-21T21:47Z 2.5M followers, 369.8K engagements


"There is so much opportunity right now in making AI Agents real for enterprises. The models pack insane capabilities but to get them into the middle of an enterprise workflow requires significant scaffolding domain understanding and context engineering. This is the play"  
[X Link](https://x.com/levie/status/1947003139686846797) [@levie](/creator/x/levie) 2025-07-20T18:38Z 2.5M followers, 261.6K engagements


"Grok X seems very strong. Great proof that the scaling laws are not in fact over. Very bullish for the future of AI"  
[X Link](https://x.com/levie/status/1891709981998596565) [@levie](/creator/x/levie) 2025-02-18T04:43Z 2.5M followers, 1.6M engagements


"It can be tricky for people outside of AI to understand how the massive build out of AI infrastructure makes sense right now. Most of the use of AI at this point is basically asking a question and getting an answer back. Its hard to perceive how this could turn into XX or 100X more infra needed to scale. But everyone inside AI is seeing a very different reality of where AI is going with agents. Just in the past year in coding for example weve gone from a world where AI was doing basically type ahead tasks or being able to generate a few hundred lines of code to now agents being able to do"  
[X Link](https://x.com/levie/status/1977079675722801607) [@levie](/creator/x/levie) 2025-10-11T18:31Z 2.5M followers, 123.3K engagements


"Were in a window right now where theres a huge advantage if youre a startup or a team that takes an AI agent-centric approach to workflows. Just in coding we see an incredible spread between in productivity gains between two seemingly only slightly different types of practices. Youll talk to some teams that say theyre getting 20-30% lift from AI and others that are getting 2-3X or more. The biggest difference is really just the latter group is making an explicit choice to design their engineering workflows to actually make agents effective instead of just assuming it will happen organically."  
[X Link](https://x.com/levie/status/1977487728762884249) [@levie](/creator/x/levie) 2025-10-12T21:33Z 2.5M followers, 202.8K engagements


"The issue that were going to have in predicting the future of jobs is that those predicting things dont have any way of measuring (in advance) the fact that job requirements will just change over time because of AI agents. Initially we imagine AI will just make what we do today more efficient; but in practice the job requirements will more likely change as a result. AI agents both free up people to do more of what they couldnt do before which will lead to all new work being done. But it also accelerates work in ways that that essentially redefines the output expectations. Here are a couple"  
[X Link](https://x.com/levie/status/1979597667031511064) [@levie](/creator/x/levie) 2025-10-18T17:17Z 2.5M followers, 30.3K engagements


"The latest Karpathy podcast was actually laying out the bull case for AI agents + SaaS. We know that were years away from a magical superintelligence that will do anything and everything for us. What we have right now increasingly are AI models that exhibit strong general reasoning capabilities with improving skills in specific domains. This will all get better and better over time unquestionably. But they still need a ton to be useful in most domains. What they need to be most effective in real world situations is context and usually an applied set of features to make the AI useful. They"  
[X Link](https://x.com/levie/status/1979941267753824738) [@levie](/creator/x/levie) 2025-10-19T16:02Z 2.5M followers, 175K engagements


"One critical point made by Karpathy on the Dwarkesh podcast was about the difference between AI coding agents and the rest of knowledge work. Its clear that AI agents in coding have taken off extremely quickly. There are many reasons for this including the fact that a substantial amount of training data is already online compared to other domains developers are quick to try new tools and the labs have a lot of incentive to get this right. But theres a subtle and fundamental reason as well that Karpathy covers: the domain is comprised almost entirely of text (great for LLMs) and we have clear"  
[X Link](https://x.com/levie/status/1980085588549362123) [@levie](/creator/x/levie) 2025-10-20T01:36Z 2.5M followers, 139.2K engagements


"Agent evals are a completely new ball game. Right now most AI evals operate within a self contained world of the model. This will remain critical for ensuring we get continued improvements in domain-specific skills and general reasoning capabilities. But agents necessarily will have access to a world that is thousands or millions of times larger than the model parameters. Knowledge sources software tools the web and so on. Thus knowing how to interact with that world will be the new thing we need to train and eval against. You can already feel this today by giving an agent many MCP tools and"  
[X Link](https://x.com/levie/status/1981403279780434058) [@levie](/creator/x/levie) 2025-10-23T16:52Z 2.5M followers, 58.2K engagements


"AI Agents will completely transform how we interact with corporate knowledge by applying reasoning across multiple data sources. Here's an example of ChatGPT's new knowledge feature to pull information from Box and Gmail to combine it an answer a question"  
[X Link](https://x.com/levie/status/1981479811500724297) [@levie](/creator/x/levie) 2025-10-23T21:56Z 2.5M followers, 48.1K engagements


"One big effect of AI agents is that they take previously scarce resources and skills and make them abundant and available to everyone. As a consequence you can self-serve many more of the hard problems that you used to have to go to others to solve. The product marketer can do more parts of marketing. The designer can do more parts of product management. And so on. As a result of that it seemingly should eventually blur the lines between adjacent job functions. However theres a counter force which is the experts in any given field are now far more potent than they ever were before which"  
[X Link](https://x.com/levie/status/1981552772542869574) [@levie](/creator/x/levie) 2025-10-24T02:46Z 2.5M followers, 40.8K engagements

[GUEST ACCESS MODE: Data is scrambled or limited to provide examples. Make requests using your API key to unlock full data. Check https://lunarcrush.ai/auth for authentication information.]

@levie "Another great win for AI agent interoperability. Box is partnering with Google to make Box AI Agents accessible via Google Gemini Enterprise so you can easily work with your enterprise content from anywhere. This is why the future of AI looks like"
X Link @levie 2025-10-09T18:54Z 2.5M followers, 72K engagements

"This is actually extremely pragmatic and realistic from @karpathy based on what is likely to happen especially in an enterprise context. We have rapidly improving AI model capabilities but the diffusion of these capabilities into real life workflows will take time and require lots of integration change management and new solutions that must be built. Basically my AI timelines are about 5-10X pessimistic w.r.t. what you'll find in your neighborhood SF AI house party or on your twitter timeline but still quite optimistic w.r.t. a rising tide of AI deniers and skeptics. The apparent conflict is"
X Link @levie 2025-10-18T20:39Z 2.5M followers, 161.8K engagements

"Having talked to hundreds of IT leaders over the last year alone its clear we have a capability overhang where the current AI models are already very good at solving many problems that havent been adopted yet. The biggest hurdles generally are the imagination for whats now possible the sheer speed at which the tech is changing and the change management to make it happen. In tech its natural to experiment with lots of solutions and do a lot of work to make something work. But outside of tech other than for a small percentage of the largest companies these capabilities need to be packaged as"
X Link @levie 2025-10-17T22:23Z 2.5M followers, 78.3K engagements

"OpenAI just launched Atlas a browser that has AI agents as a native capability. Now you can let browser agents take over your software and automate anything. Here's the agent using Box to automate folder creation for an M&A due diligence process (sped up)"
X Link @levie 2025-10-21T18:58Z 2.5M followers, 346.9K engagements

"Software engineers are the translation layer between the market and the computer. In a world where AI agents will generate 10X or 100X more code than humans this translation layer will be more important than ever. With AI agents dramatically lowering the raw cost of code production the market will decide that it wants even more software built. We will build far more digital experiences in far more industries we will upgrade systems that were once forgotten about we will add more features to every piece of software (like it or not) and generally say yes to building way more. In this world"
X Link @levie 2025-10-13T23:15Z 2.5M followers, 101.3K engagements

"Were going to enter a new phase of AI where the average consumer may not perceive a huge difference in any one model upgrade but specific industries will feel significant improvements occur. The next phase of AI be about making models better and better for AI agents in real world knowledge work domains. Were going to see new evals emerge that test for relevant tasks in these spaces the generation of high quality data and environments that align to real work training against specific tools and software in these spaces and more. The test case of how to improve models on axis of real world"
X Link @levie 2025-10-22T04:40Z 2.5M followers, 84.6K engagements

"Anyone who has been working on AI agents especially in an enterprise context would have been nodding along to Karpathys podcast the entire time. So much has yet to get built out to make it all work at scale. The analogy of self driving cars is especially apt each additional X of reliability takes a massive amount of work. XX% XX% XXXX% are all new levels of complexity that tend to require new breakthroughs more data and so on. The one benefit of AI agents for knowledge work in contrast to self-driving cars though is that you can decide to get value at any point along the way by just tweaking"
X Link @levie 2025-10-21T00:45Z 2.5M followers, 102.7K engagements

"Designing AI agent interaction patterns is a completely wide open space right now. If you use X different tools that do agentic tasks youll see X totally different interaction patterns. Some show all the steps that are being taken by the agent. Some show the thinking tokens. Some show the core thats being executed or computer thats being used. Some show the planning happening. Some let you edit the planning. Some obfuscate this all. Then theres all the UI and interaction patterns for how to work with agent once its produced some output. How do you incorporate what its done into something"
X Link @levie 2025-10-23T02:45Z 2.5M followers, 44.1K engagements

"Did something change in robotics like exactly one month ago to make all of this possible. This is crazy"
X Link @levie 2025-03-20T20:20Z 2.5M followers, 917.3K engagements

"Product management for AI agents is easily the wildest form of product management in history. Typical product management is trying to figure out how to design interfaces and software for people to interact with deterministic systems. The user generally knows all the context to do their work successfully so its generally a matter of nailing the underlying business logic and surrounding UX. But with AI agents the user you care about most is the agent and they dont know anything by default. Theyll happily run in any direction to perform the task often without success. So as a PM (or engineer)"
X Link @levie 2025-08-30T20:11Z 2.5M followers, 248K engagements

"@sama Just sent this XX minutes ago after a cursor session with codex"
X Link @levie 2025-10-12T01:11Z 2.5M followers, 36.8K engagements

"Designing for AI agent interaction patterns is such a fascinating process. In theory it should be as simple as designing the interaction that you would want to have with another person that was doing work for you. A big difference however is it costs this worker nothing to show all the work its doing along the way or how it came to its final result which is a huge asset. And at the same time you generally want the confidence that its doing the right thing or you may need to tilt it in a different direction if it goes off the rails. In the era of automation and robots seeing more context about"
X Link @levie 2025-10-23T04:59Z 2.5M followers, 4568 engagements

"The next big breakthroughin AI is AI Agents.This is when AI goes from being used as an assistant to chat with to using AI to accomplish complete tasks that a human might otherwise have to perform. This moves AI from being a "read-only" operation to fundamentally a "read/write" operation.Ultimately this brings us much closer to the full promise of AI in particularin the enterprise where AI can begin to complete any part of a workflow and we're already seeing examples today of Agents that write entire software applications or respond tocustomersupport tickets. Today in many ways Agents are"
X Link @levie 2024-04-21T21:47Z 2.5M followers, 369.8K engagements

"There is so much opportunity right now in making AI Agents real for enterprises. The models pack insane capabilities but to get them into the middle of an enterprise workflow requires significant scaffolding domain understanding and context engineering. This is the play"
X Link @levie 2025-07-20T18:38Z 2.5M followers, 261.6K engagements

"Grok X seems very strong. Great proof that the scaling laws are not in fact over. Very bullish for the future of AI"
X Link @levie 2025-02-18T04:43Z 2.5M followers, 1.6M engagements

"It can be tricky for people outside of AI to understand how the massive build out of AI infrastructure makes sense right now. Most of the use of AI at this point is basically asking a question and getting an answer back. Its hard to perceive how this could turn into XX or 100X more infra needed to scale. But everyone inside AI is seeing a very different reality of where AI is going with agents. Just in the past year in coding for example weve gone from a world where AI was doing basically type ahead tasks or being able to generate a few hundred lines of code to now agents being able to do"
X Link @levie 2025-10-11T18:31Z 2.5M followers, 123.3K engagements

"Were in a window right now where theres a huge advantage if youre a startup or a team that takes an AI agent-centric approach to workflows. Just in coding we see an incredible spread between in productivity gains between two seemingly only slightly different types of practices. Youll talk to some teams that say theyre getting 20-30% lift from AI and others that are getting 2-3X or more. The biggest difference is really just the latter group is making an explicit choice to design their engineering workflows to actually make agents effective instead of just assuming it will happen organically."
X Link @levie 2025-10-12T21:33Z 2.5M followers, 202.8K engagements

"The issue that were going to have in predicting the future of jobs is that those predicting things dont have any way of measuring (in advance) the fact that job requirements will just change over time because of AI agents. Initially we imagine AI will just make what we do today more efficient; but in practice the job requirements will more likely change as a result. AI agents both free up people to do more of what they couldnt do before which will lead to all new work being done. But it also accelerates work in ways that that essentially redefines the output expectations. Here are a couple"
X Link @levie 2025-10-18T17:17Z 2.5M followers, 30.3K engagements

"The latest Karpathy podcast was actually laying out the bull case for AI agents + SaaS. We know that were years away from a magical superintelligence that will do anything and everything for us. What we have right now increasingly are AI models that exhibit strong general reasoning capabilities with improving skills in specific domains. This will all get better and better over time unquestionably. But they still need a ton to be useful in most domains. What they need to be most effective in real world situations is context and usually an applied set of features to make the AI useful. They"
X Link @levie 2025-10-19T16:02Z 2.5M followers, 175K engagements

"One critical point made by Karpathy on the Dwarkesh podcast was about the difference between AI coding agents and the rest of knowledge work. Its clear that AI agents in coding have taken off extremely quickly. There are many reasons for this including the fact that a substantial amount of training data is already online compared to other domains developers are quick to try new tools and the labs have a lot of incentive to get this right. But theres a subtle and fundamental reason as well that Karpathy covers: the domain is comprised almost entirely of text (great for LLMs) and we have clear"
X Link @levie 2025-10-20T01:36Z 2.5M followers, 139.2K engagements

"Agent evals are a completely new ball game. Right now most AI evals operate within a self contained world of the model. This will remain critical for ensuring we get continued improvements in domain-specific skills and general reasoning capabilities. But agents necessarily will have access to a world that is thousands or millions of times larger than the model parameters. Knowledge sources software tools the web and so on. Thus knowing how to interact with that world will be the new thing we need to train and eval against. You can already feel this today by giving an agent many MCP tools and"
X Link @levie 2025-10-23T16:52Z 2.5M followers, 58.2K engagements

"AI Agents will completely transform how we interact with corporate knowledge by applying reasoning across multiple data sources. Here's an example of ChatGPT's new knowledge feature to pull information from Box and Gmail to combine it an answer a question"
X Link @levie 2025-10-23T21:56Z 2.5M followers, 48.1K engagements

"One big effect of AI agents is that they take previously scarce resources and skills and make them abundant and available to everyone. As a consequence you can self-serve many more of the hard problems that you used to have to go to others to solve. The product marketer can do more parts of marketing. The designer can do more parts of product management. And so on. As a result of that it seemingly should eventually blur the lines between adjacent job functions. However theres a counter force which is the experts in any given field are now far more potent than they ever were before which"
X Link @levie 2025-10-24T02:46Z 2.5M followers, 40.8K engagements

creator/twitter::914061/posts
/creator/twitter::914061/posts