I love this stuff. From my pov the short term use case is using AI agents to create dumb automated workflows quickly. These are much safer to permission.
Yes, today's agentic capabilities are quite powerful. But what Levie envisions is a much different kind of agentic capability than what we are able to do today.
Hm I don’t think these issues arise at the level of a particular model. They’re about ai agents in general not ones which operate on top of a particular model.
Yeah, I meant that if platforms like Microsoft and Google have an incentive to throttle external agents (which I agree makes sense) their own agents should have inherent advantages over those like OpenAI that may not have this inbuilt interoperability with major workplace platforms. Unless OpenAI moves in the direction of further integration with Microsoft of course
Within the Enterprise, agents can work with safely with in-house applications using Interop that respects the entitlement and permissioning by working with the app UI's directly, which also keeps the user in the loop.
I just posted on LinkedIn that Bob Marley has the answer to Satya Nadella's problems with Copilot 'In this great future, you can't forget your past'
The irony is that the best place and way to grow capability and acceptance of agents is at the edge. Since they're so unreliable agents need to be iterated on in low-consequence environments. Home automation, personal automation in your phone or watch, GPS in your car. Make those things accurately predict, become good proxies, and limit their blast radius. They're small potatoes but that's good for the tech, just bad for the money.
I love this stuff. From my pov the short term use case is using AI agents to create dumb automated workflows quickly. These are much safer to permission.
Yes, today's agentic capabilities are quite powerful. But what Levie envisions is a much different kind of agentic capability than what we are able to do today.
1. “The spreadsheet with the “real numbers “ 😂
2. Do “non-aligned” models like OpenAI have a giant hill to climb here?
Content is fire this week, keep it coming
Hm I don’t think these issues arise at the level of a particular model. They’re about ai agents in general not ones which operate on top of a particular model.
Yeah, I meant that if platforms like Microsoft and Google have an incentive to throttle external agents (which I agree makes sense) their own agents should have inherent advantages over those like OpenAI that may not have this inbuilt interoperability with major workplace platforms. Unless OpenAI moves in the direction of further integration with Microsoft of course
Within the Enterprise, agents can work with safely with in-house applications using Interop that respects the entitlement and permissioning by working with the app UI's directly, which also keeps the user in the loop.
I just posted on LinkedIn that Bob Marley has the answer to Satya Nadella's problems with Copilot 'In this great future, you can't forget your past'
The irony is that the best place and way to grow capability and acceptance of agents is at the edge. Since they're so unreliable agents need to be iterated on in low-consequence environments. Home automation, personal automation in your phone or watch, GPS in your car. Make those things accurately predict, become good proxies, and limit their blast radius. They're small potatoes but that's good for the tech, just bad for the money.