with Ed Anuff and Anant Jhingran
Building the Infrastructure for Agentic AI
May 29th, 2025 | 27:10 Runtime
Episode Transcript
TIMESTAMPS
00:32:8 - Discussion about AI-generated versions of themselves and making it a regular practice
01:13:1 - Key observations from IBM Think keynote, focusing on agents being front and center
02:45:1 - Discussion of IBM's bold quantum computing predictions for the next 4-5 years
03:51:0 - Windsurf becoming part of OpenAI confirmed and concluded
05:30:8 - IBM's focus on business process impact rather than just models
08:39:3 - Observation about applications using ensembles of models rather than single models
11:51:2 - Discussion of Anthropic's computer use model and agent interfaces
16:40:9 - Debate within IBM about API design quality for AI agents
22:30:5 - Prediction about the evolution toward APIs designed specifically for agents
25:21:1 - Transition to discussing the IBM-DataStax partnership
QUOTES
"The most pervasive thing was just seeing agents front and center, and I'm really blown away at the level that these have actually been deployed within IBM." —Ed Anuff
"It was really interesting to see Arvind make some bold predictions that in the next four or five years Quantum will actually have a material impact on the things that they're doing. And that's the kind of thing that you'll discover that at IBM ... we are able make some good bold bets and stick with them because some of these require really, really long term investments." —Anant Jhingran
"Agents are, in a very sort of expansive use of the term, really the new form of applications." —Ed Anuff
"The more accuracy comes from cleaner APIs with richer semantics, better documentation, more logical structure ... the more that you can express, the better and more accurate the models will be." —Ed Anuff
"If you have a well designed API, then the model is able to get it right the first time. It's able to interpret the results in a structured way... The end result for the user is that it gets completed faster, more accurately." —Ed Anuff
"So in the world of APIs, we always felt that APIs needed to be fit-for-purpose for the applications... So is there a world in which new interfaces will be designed for agents? Or, the opposite point of view would be that, look, the interfaces that have been designed for humans, the agents will be able to operate against and therefore you don't really need to invent something new." —Anant Jhingran
TIMESTAMPS
00:32:8 - Discussion about AI-generated versions of themselves and making it a regular practice
01:13:1 - Key observations from IBM Think keynote, focusing on agents being front and center
02:45:1 - Discussion of IBM's bold quantum computing predictions for the next 4-5 years
03:51:0 - Windsurf becoming part of OpenAI confirmed and concluded
05:30:8 - IBM's focus on business process impact rather than just models
08:39:3 - Observation about applications using ensembles of models rather than single models
11:51:2 - Discussion of Anthropic's computer use model and agent interfaces
16:40:9 - Debate within IBM about API design quality for AI agents
22:30:5 - Prediction about the evolution toward APIs designed specifically for agents
25:21:1 - Transition to discussing the IBM-DataStax partnership
QUOTES
"The most pervasive thing was just seeing agents front and center, and I'm really blown away at the level that these have actually been deployed within IBM." —Ed Anuff
"It was really interesting to see Arvind make some bold predictions that in the next four or five years Quantum will actually have a material impact on the things that they're doing. And that's the kind of thing that you'll discover that at IBM ... we are able make some good bold bets and stick with them because some of these require really, really long term investments." —Anant Jhingran
"Agents are, in a very sort of expansive use of the term, really the new form of applications." —Ed Anuff
"The more accuracy comes from cleaner APIs with richer semantics, better documentation, more logical structure ... the more that you can express, the better and more accurate the models will be." —Ed Anuff
"If you have a well designed API, then the model is able to get it right the first time. It's able to interpret the results in a structured way... The end result for the user is that it gets completed faster, more accurately." —Ed Anuff
"So in the world of APIs, we always felt that APIs needed to be fit-for-purpose for the applications... So is there a world in which new interfaces will be designed for agents? Or, the opposite point of view would be that, look, the interfaces that have been designed for humans, the agents will be able to operate against and therefore you don't really need to invent something new." —Anant Jhingran