Concurrency Viral Topics: AI Models, Agents & Risk

What’s Changing and What It Means for Your Business

Concurrency Viral Topics is a fast-paced webinar series exploring the technology trends dominating boardroom discussions, inbox conversations, and LinkedIn feeds.

Join Concurrency Founder and CEO James Savage, Solutions Architect Brian Haydin, and Practice Manager Brad Zepecki as they break down the AI topics everyone is talking about, what they mean for your business, and how to separate the hype from the real opportunities.

This month, the CEOs of Anthropic, OpenAI, and xAI publicly agreed that AI development needs to slow down. Days later, new flagship models were released anyway. Three topics, ten minutes each, straight answers:

Is the AI pause real? What “pacing the frontier” actually means, what it doesn’t, and whether it should change your AI strategy.
A flood of new models. What’s genuinely better, what’s just model churn, and how to evaluate new releases without chasing every announcement.
Agents off the leash. Recent examples of AI agents reaching systems they were never intended to access, and the governance and security controls organizations need before deploying their own.

Join us for a fast-moving discussion that cuts through the headlines and focuses on what matters most for business leaders evaluating AI investments, risk, and opportunity.

Organizations are navigating an expanding choice of AI models while deciding how much responsibility to delegate to agents. The conversation is moving beyond generating answers toward systems that perform tasks on behalf of people and business functions. The real challenge is delivering useful AI outcomes with controls that match the risks of the work.

Benchmarks provide only part of the picture, lower token prices can conceal higher reasoning costs, and upgrades can introduce API compatibility problems. When agents also gain access to sensitive functions, model selection becomes an operational decision involving permissions, testing, information protection, and oversight.

In [Webinar title], Brian Haydin, Solutions Architect at Concurrency, James Savage, Chief Executive Officer at Concurrency, and Brad Zepecki, Practice Manager at Concurrency, explore the AI pause debate, safety and alignment, risk-based governance, model proliferation, and agent autonomy. The discussion connects engineering decisions with organizational responsibilities, including model diversity, intellectual property concerns, separate agent identities, and visibility into model behavior.

Rather than focusing on which newly released model deserves immediate adoption, the session examines why the surrounding harness matters, why switching models needs a material justification, and why rapid development increases the importance of preparation and testing. The speakers connect those choices to AgentOps and differentiated governance for chat and purpose-built agents.

The key takeaway is clear:

Organizations do not need to wait for frontier models to implement evaluations and proper controls. Model changes should be justified through testing against the work the solution needs to perform.

Organizations can use this approach to make model decisions around demonstrated outcomes rather than release headlines, bringing permissions, guardrails, and operational oversight into the conversation before agents take on sensitive responsibilities.

WHAT YOU’LL LEARN

The AI Pause Debate, Safety, and Alignment

Understand the distinction between safety, which concerns preventing serious harm, and alignment, which concerns following intended goals.
Explore competing views around pausing frontier AI development while safeguards catch up.
Examine the regulatory capture concern: rules intended to manage risk may also create barriers for competitors.

Delivery Architecture and Risk-Based Governance

Learn why the engineering discussion centers on the harness and solution delivery rather than following every model release.
Understand the recommendation to match governance to the use case’s risk profile.
Explore how the speakers connect evaluations, guardrails, and Agent 365 with building and managing AI solutions.

Model Choice, Diversity, and Information Protection

Examine why changing an established solution requires a material reason.
Understand the relationship the speakers draw between model diversity and protecting organizational intellectual property.
Explore model selectors, Microsoft 365 Copilot auto mode, and decisions about local versus cloud models.

Upgrade Risk and the Full Cost of Reasoning

Learn why model upgrades can create additional work when API compatibility changes.
Understand why benchmarks should not replace evaluation of the intended workload.
Examine how lower per-token prices can still produce higher total costs through additional reasoning.

Agent Autonomy, Permissions, and AgentOps

Understand how moving from requesting information to delegating actions changes organizational risk.
Explore why agents performing sensitive functions need management and controls, not simply deployment.
Learn why the discussion highlights separate agent identities rather than treating agents as extensions of individual users.

Testing, Platform Controls, and Operational Visibility

Examine how faster development shifts engineering effort toward preparation, testing, and guardrails.
Explore the discussion of Foundry, Purview, and Agent 365 alongside different governance needs for chat and custom agents.
Understand the closing emphasis on adaptability, visibility into model behavior, internal controls, and auditors.

FREQUENTLY ASKED QUESTIONS

Is this session technical or business-focused?

Both. The speakers connect engineering concerns such as model compatibility, evaluation, reasoning costs, and agent identities with leadership responsibilities for sensitive functions and oversight.

Who should attend?

Technology leaders, solution architects, AI engineers, security and governance teams, and business leaders considering agent-enabled workflows. The discussion is particularly relevant to organizations deciding when to change models or how to control agents acting on behalf of users.

Do I need experience with AI agents?

The discussion introduces safety, alignment, and the move from requesting information to delegating tasks. Familiarity with AI tools provides context, but the session is a conversational exploration rather than a step-by-step implementation tutorial.

What is AgentOps in the context of this session?

The speakers use AgentOps to discuss operational management of agents within an organization. Topics include controlling agents assigned to users or functions, managing sensitive work, and establishing oversight as agents take on responsibilities beyond responding to questions.

Why does AI governance matter right now?

The speakers describe a shift toward agents that perform actions rather than only provide information. That shift introduces questions about autonomy, privileges, and organizational impact, while frequent model changes create a need to assess behavior, compatibility, and cost.

What is the most important takeaway?

Implement evaluations and appropriate controls without waiting for developments at frontier labs. Compare models through testing against the intended workload rather than assuming the newest release is the right choice.

ABOUT THE SPEAKERS

Brian Haydin, Solutions Architect at Concurrency, frames the session’s discussion of risk-based AI delivery. His contributions focus on evaluations, model selection, reasoning costs, agent autonomy, separate identities, and controls.

James Savage, Chief Executive Officer at Concurrency, brings an organizational leadership perspective to AI adoption and oversight. He examines safety and alignment, regulatory capture, model diversity, intellectual property protection, AgentOps, and internal controls.

Brad Zepecki, Practice Manager at Concurrency, brings the delivery and engineering perspective. His contributions emphasize the harness, material reasons for model changes, upgrade compatibility, testing, and adaptability.

TRANSCRIPT

Transcription Collapsed Transcription Expanded

All right. Thanks a lot for joining us today. 0:0:8.103 –> 0:0:15.743 Brian Haydin We are with the next edition of the viral hot topics where I’ve been sort of just playing it by ear. 0:0:15.743 –> 0:0:22.583 Brian Haydin What’s the latest and greatest stuff that people are talking about today? And there has been a lot of things going on over the last week. 0:0:23.293 –> 0:0:27.93 Brian Haydin We’re seeing my Twitter feed is just blowing up with a lot of different topics. 0:0:27.213 –> 0:0:34.413 Brian Haydin Some of it models, some of it risk, some of IT policy, and so it’s quite a bit to talk about. 0:0:34.483 –> 0:0:39.323 Brian Haydin About I’m joined today with our delivery lead. 0:0:39.323 –> 0:0:43.723 Brian Haydin It’s Becky and our our team owner, Jim Savage. 0:0:43.723 –> 0:0:53.283 Brian Haydin So yeah, I wanted to start off with the the big topic that’s been happening over the last two weeks is this, you know, pausing on AI to let safety kind. 0:0:53.283 –> 0:0:54.43 Of catch up. 0:0:54.973 –> 0:0:56.933 Brian Haydin And we saw both. 0:0:58.853 –> 0:1:0.573 Brian Haydin You know amine and Altman? 0:1:0.573 –> 0:1:4.213 Brian Haydin You know, leaning into this and a little bit of pushback from the White House. 0:1:4.683 –> 0:1:14.123 Brian Haydin All that kind of stuff and and so for those that you know, maybe haven’t heard that story, the idea here is that, you know, these AI tools are starting to get really powerful. 0:1:14.123 –> 0:1:18.923 Brian Haydin We’re seeing them actually get outside of their sandboxes. 0:1:19.813 –> 0:1:24.613 Brian Haydin And the the the leading kind of theory is, let’s let the safety features catch up. 0:1:24.613 –> 0:1:26.53 Brian Haydin So they’re calling for a pause. 0:1:27.893 –> 0:1:34.213 Brian Haydin And I wanted to ask you, Jim, first about that is this AI pause reals gonna happen? What are your thoughts? 0:1:34.483 –> 0:1:36.243 Sure. Yeah. Lot going on, huh? 0:1:36.243 –> 0:1:38.923 And you know what’s so strange is it’s only about a month old. 0:1:38.923 –> 0:1:46.3 James Savage It the whistleblower, I think that was September 7th, 8th or 8th or 9th. And yeah that really crazy. OK. 0:1:46.3 –> 0:1:53.563 James Savage So I’m gonna set the stage here to find some terms and talk about some of the characters in this drama that’s been unfolding since, I said. 0:1:54.773 –> 0:1:56.53 James Savage September 8. It really got moving. 0:1:57.653 –> 0:2:3.973 James Savage Terms you can’t really talk about the pause without defining some terms, but later talking about some of the personalities that have promoted it. 0:2:4.283 –> 0:2:12.203 James Savage And some of its critics, generally speaking, the concerns focus around safety and alignment to hear those terms come up over and over again. 0:2:14.373 –> 0:2:18.453 James Savage Relative to these, continuously improving frontier language models, right? 0:2:18.453 –> 0:2:18.693 The crux? 0:2:18.693 –> 0:2:21.693 James Savage The argument is mostly about government regulation. 0:2:21.693 –> 0:2:28.973 James Savage So we’ve got the frontier model firms, Sam Dario, Elon, the folks from Google. 0:2:29.973 –> 0:2:32.813 James Savage Essentially asking the government in ways for regulation. 0:2:32.813 –> 0:2:34.613 James Savage Then of course, you got the pundits on the other side saying. 0:2:35.3 –> 0:2:37.763 James Savage Saying, wait a minute, we’ve got way too much regulation already. 0:2:37.763 –> 0:2:39.963 James Savage We’ve got plenty of product liability. 0:2:39.963 –> 0:2:42.43 James Savage There’s there’s, you know, you do your products right. 0:2:42.43 –> 0:2:43.243 You don’t have a problem. 0:2:44.733 –> 0:2:54.653 James Savage The two terms Big 2 terms safety and alignment safety keep the systems from doing serious harm, breaking into networks, running off on their own, helping with weapons or causing large real world damage. 0:2:54.653 –> 0:2:55.573 James Savage Right. That’s what we’re talking about. 0:2:55.733 –> 0:2:57.573 James Savage We’re talking about safety alignment. 0:2:57.613 –> 0:3:1.933 James Savage Make sure the systems are doing what people actually meant them to do right. 0:3:2.843 –> 0:3:9.683 James Savage Not the loophole that scores the goal. If you say optimize working capital, it should not go fake the books, right? 0:3:10.123 –> 0:3:12.603 James Savage So safety is essentially don’t wreck things. 0:3:12.603 –> 0:3:16.123 James Savage Alignment is do what we meant, OK? Another term. 0:3:16.123 –> 0:3:18.163 James Savage I’m gonna. I’m gonna define here right up front. 0:3:18.933 –> 0:3:19.813 Regulatory capture. 0:3:19.813 –> 0:3:27.813 James Savage So regulatory capture, essentially the concept of key players inviting regulation to closing around them and potentially create barriers for competitors. 0:3:27.813 –> 0:3:29.813 James Savage You know, I think that’s probably the biggest issue. 0:3:29.813 –> 0:3:30.413 Would you agree? 0:3:30.413 –> 0:3:31.613 Yeah, yeah. 0:3:32.43 –> 0:3:36.723 Brian Haydin And that, you know, one of the things that I think about is that government touching things never ends well. 0:3:36.723 –> 0:3:37.43 Yeah. 0:3:37.43 –> 0:3:40.323 Brian Haydin You know, so the, you know, I like. 0:3:40.323 –> 0:3:46.123 Brian Haydin I like Wang’s like comment on in in September where he said we really don’t need any new laws. 0:3:46.203 –> 0:3:49.843 Brian Haydin You know, I wanna. I wanna make sure that the innovation still has. 0:3:50.573 –> 0:3:52.773 Brian Haydin You know unrestricted, you know, kind of thing. 0:3:52.773 –> 0:4:1.373 Brian Haydin The other thing for me, I I agree with everything you’re saying about the the risks and you know the guardrails, but you know something for me that that really. 0:4:1.523 –> 0:4:5.3 Brian Haydin Chancell is that American can pause all they want. 0:4:5.3 –> 0:4:8.203 Brian Haydin Yeah. The only thing I don’t think is either Russia going to stop. 0:4:8.203 –> 0:4:16.563 Brian Haydin No, you know, it’s got to stop doing what they’re doing and they had some high level talks at the White House last week. You know, Brad, I mean, what are your thoughts about this pause? 0:4:16.563 –> 0:4:17.643 Brian Haydin What do you think that? 0:4:18.453 –> 0:4:20.413 Brian Haydin Is, you know, should we kind of move forward? 0:4:20.653 –> 0:4:22.13 Brian Haydin Are you seeing a lot of risk? 0:4:22.13 –> 0:4:24.493 Brian Haydin What are you know, what are are people thinking? 0:4:24.493 –> 0:4:29.733 @3 Yeah. So my perspective just for for everyone revolves around the delivery of the solution. 0:4:29.733 –> 0:4:31.493 @3 So coming at it from an engineering perspective. 0:4:31.763 –> 0:4:37.203 @3 We’re not really thinking about the pause the the number of models that are coming out. 0:4:37.203 –> 0:4:42.813 @3 I mean, they’re coming out a lot faster, obviously for us, it’s all about the harness, what harness? 0:4:42.813 –> 0:4:43.83 @3 Pick a harness. 0:4:43.563 –> 0:4:45.963 @3 There’s so many models coming out that it. 0:4:47.293 –> 0:4:48.453 Really, you know. 0:4:48.453 –> 0:4:51.133 @3 It was hard to keep up five and 10 years ago. 0:4:51.373 –> 0:4:53.493 @3 It’s almost impossible to keep up so. 0:4:53.493 –> 0:4:57.93 Brian Haydin We’ve got a lot of playbooks too kind of in. We’re we’re putting in guardrails. 0:4:53.743 –> 0:4:56.223 We’ve got a lot of, yeah, yeah. 0:4:57.733 –> 0:5:1.493 Brian Haydin Agent OPS is kind of a thing that we’re really leaning in, you know, lately and stuff like that. 0:5:0.213 –> 0:5:0.773 Yeah. 0:5:1.723 –> 0:5:11.763 Brian Haydin And that’s it’s a good perspective like this is you know the esoteric kind of conversations when we get into policy making and we get into you know some of the weeds here about this. 0:5:11.763 –> 0:5:16.43 Brian Haydin But but yeah, like I I honestly think that like. 0:5:16.773 –> 0:5:24.533 Brian Haydin We can be as well intended as we want and and, you know, say that we want to do a pause, but I don’t think that there’s gonna be a pause. 0:5:24.893 –> 0:5:26.333 Brian Haydin Jim, what do you? 0:5:26.333 –> 0:5:27.293 What do you think to do on that? 0:5:27.293 –> 0:5:29.53 Like I think it’s so interesting. 0:5:29.53 –> 0:5:29.653 The paradox. 0:5:29.653 –> 0:5:30.773 James Savage Maybe the dichotomy. 0:5:31.203 –> 0:5:39.763 James Savage Of you know, you consider the safety and security the guardrail requirements of to use Nadella’s favorite term diffusion, right? 0:5:39.763 –> 0:5:44.203 James Savage I mean, I think that’s one of the biggest underlying issues and we saw him on the all in podcast. 0:5:44.243 –> 0:5:44.643 It was kind. 0:5:44.643 –> 0:5:47.123 James Savage Of funny right in the middle of this whole controversy. 0:5:47.853 –> 0:5:48.493 The what is it? 0:5:48.493 –> 0:6:0.293 James Savage The All in podcast had their had their their annual in person stage show giving a forum to, you know, all the players from Satya to Jensen Huang and you know, it’s just fascinating. 0:6:0.683 –> 0:6:2.723 James Savage And hearing from all of them in that week. 0:6:2.723 –> 0:6:4.843 James Savage So yeah, we’ve had a headful this past week. 0:6:4.843 –> 0:6:7.643 James Savage We got a headful of everyone’s opinion. 0:6:9.53 –> 0:6:14.93 Brian Haydin Somebody’s opinion that you know you are probably in the Dallas biggest fan of the organization. 0:6:13.583 –> 0:6:16.503 No fanboy. It’s not. 0:6:15.813 –> 0:6:16.853 So he’s got a very different. 0:6:16.853 –> 0:6:24.173 Brian Haydin Perspective of of you know this AI pause. What do you you know what are you hearing from the Microsoft lens? Yeah. Yeah. 0:6:25.973 –> 0:6:27.613 James Savage He’s saying it’s an engineering issue. 0:6:27.613 –> 0:6:29.413 James Savage It’s a DevOps issue. I mean it’s been a dev op issue. 0:6:30.83 –> 0:6:37.363 James Savage Dev OPS issue. You know what really kicked most of the concern off initially was the hugging face, Hap app hack. 0:6:37.763 –> 0:6:39.283 James Savage Everybody remembers that one right? 0:6:39.283 –> 0:6:43.803 James Savage Suddenly, an Openai model had broken its out of its sandbox. 0:6:44.853 –> 0:6:51.13 James Savage Got up to the to the public Internet and hacked the hugging face site went and found the answers to the test questions and. 0:6:53.733 –> 0:6:55.813 James Savage Yeah, big, big controversy. 0:6:56.573 –> 0:6:58.173 James Savage Nadella took two issues with that. 0:6:58.643 –> 0:6:59.563 You know, why are we? 0:6:59.563 –> 0:7:3.923 James Savage Why are we creating model swarms to, you know, beat these benchmarks? 0:7:3.923 –> 0:7:8.803 James Savage You know, should that really have been one of the underlying objectives of the pre training and post training function? 0:7:10.253 –> 0:7:13.93 James Savage But essentially he refers to it as it’s an engineering issue, right? 0:7:13.173 –> 0:7:19.893 James Savage There’s a wonderful quote of his if you know if you’ve got a show stopper bug stop the show, right? 0:7:19.893 –> 0:7:22.53 James Savage Which makes an infinite amount of sense. 0:7:22.93 –> 0:7:23.973 James Savage Jensen Wang makes a lot of talks. 0:7:23.973 –> 0:7:24.573 A lot about. 0:7:25.533 –> 0:7:27.373 James Savage You know, we have plenty of product liability. 0:7:27.923 –> 0:7:29.523 James Savage Regulation already in place. 0:7:29.523 –> 0:7:34.203 James Savage These guys need to follow that product liability regulation, the essential tenets of it. 0:7:36.133 –> 0:7:48.533 James Savage Wang makes makes the point that there’s there’s the plenty of regulatory, regulated or plenty of product regulation, but these these companies are moving from, you know, their labs into product companies, right? 0:7:48.533 –> 0:7:53.13 James Savage They’re all waiting for the big multi trillion dollar one or $2 trillion IPO. 0:7:54.93 –> 0:7:56.573 James Savage And, you know, racing toward that end. 0:7:56.963 –> 0:8:15.123 James Savage And you know, perhaps they’re doing some things that run a little fast for as opposed to doing taking a more safer cars, introducing the engineering, the guardrails as you mentioned, keeping track of the alignment, the security of these of these models as they’re developing. 0:8:1.993 –> 0:8:2.113 God. 0:8:15.893 –> 0:8:17.133 James Savage I don’t think it’s a bad idea, frankly. 0:8:17.133 –> 0:8:26.853 James Savage I think you know if you look at the each of the players and what they’re saying, I think there’s a general consensus around rather than pushing these models as fast at RSI. 0:8:28.3 –> 0:8:29.723 James Savage Recursive self improvement. 0:8:29.803 –> 0:8:34.523 James Savage Let’s focus a little bit more about on the safety and security of these things. 0:8:34.683 –> 0:8:42.203 Brian Haydin Yeah, governance has really taken off as a conversation we’re having with our customers and definitely seeing that uptick, you know. 0:8:42.973 –> 0:8:57.293 Brian Haydin And your team has been leaning in on things like evals and Agent 365 as well, you know. So I would say, like, my my perspective is that, you know, look at the governance, you know, and. 0:8:57.483 –> 0:9:7.83 Brian Haydin And align that to the risk profile of what you’re trying to build, rather than trying to have this like one-size-fits-all you know kind of kind of thing. 0:9:7.83 –> 0:9:9.723 Brian Haydin And these models are getting really good at safety. 0:9:10.163 –> 0:9:11.803 Brian Haydin I’ve got a talk coming up on. 0:9:12.493 –> 0:9:13.133 Brian Haydin On Friday, where? 0:9:14.933 –> 0:9:27.933 Brian Haydin I talked that I’ve done before, but it’s it’s about the cracking MCPS, you know, and I have to rework the demos that I’ve been doing because these models are they’re preventing the kind of things that I was doing before. 0:9:28.973 –> 0:9:32.173 Brian Haydin So it you know, they’re they’re paying a lot of attention. 0:9:32.613 –> 0:9:34.773 Brian Haydin You know to it which you know kind of leads me. 0:9:34.773 –> 0:9:34.933 I’m gonna. 0:9:34.933 –> 0:9:39.573 Brian Haydin I’m gonna segue this one in the next segment, but I’m gonna use this topic here really quick. 0:9:39.573 –> 0:9:43.373 Brian Haydin So the model flood we get all these models that are coming out. 0:9:44.143 –> 0:9:47.263 Brian Haydin And one of the points that I would say before we transition is that. 0:9:49.463 –> 0:9:51.583 Brian Haydin We like these these policy discussions. 0:9:51.583 –> 0:9:58.113 Brian Haydin They’re centered really more around the big, you know, the big AI frontier labs. But there’s a lot of, like, really open source models that are coming out. 0:9:58.173 –> 0:10:4.453 Brian Haydin That are capable of doing really, really incredible things at really, really small, really compact models. 0:10:4.453 –> 0:10:9.533 Brian Haydin So my point is, like we can pretend that we’re going to pause this stuff all we want. 0:10:9.533 –> 0:10:12.93 Brian Haydin It’s like these open source models are going to continue to evolve. 0:10:13.223 –> 0:10:20.143 Brian Haydin But that leads me into this like you know, you know weird. You know, how do we deal with all these models? Last week we had frontier. 0:10:20.143 –> 0:10:26.23 Brian Haydin We everyone on the Frontier Labs was releasing new models and we’re getting them at such a frequent pace as that. 0:10:26.63 –> 0:10:28.823 Brian Haydin Like, how do you even figure out what which ones to use? 0:10:29.493 –> 0:10:33.813 Brian Haydin So you know Brad, what is your team doing to keep up to date? 0:10:33.813 –> 0:10:35.13 When should we be switching? 0:10:35.13 –> 0:10:36.493 Brian Haydin When should we be thinking about? 0:10:37.903 –> 0:10:45.423 Brian Haydin You know, not getting into the weeds of which models are better and stuff like that, but like, you know just I selected like how are they looking at that problem? 0:10:45.783 –> 0:10:53.503 @3 Well, I think for for my team it’s it’s not necessarily picking one as they come out it’s all about. 0:10:54.503 –> 0:10:58.583 @3 Because they’re so similar, we’re looking at obviously cost as part of the equation. 0:10:59.13 –> 0:11:7.853 @3 And and just selecting things that are easy to swap models in and out depending on the work that you’re doing. 0:11:7.853 –> 0:11:9.613 @3 So it’s really about. 0:11:11.183 –> 0:11:26.743 @3 Creating the best outcome you can in the solution that you’re working working in and my team’s not spending a ton of time looking at the models because once they’ve settled in, it has to be a pretty material change before they they think about doing that because that. 0:11:26.783 –> 0:11:27.103 @3 You know. 0:11:28.183 –> 0:11:29.63 It’s basically. 0:11:29.453 –> 0:11:32.253 @3 Changing how you’re gonna do do your work so. 0:11:33.943 –> 0:11:35.623 That’s that’s typically what depends on. 0:11:36.183 –> 0:11:40.143 James Savage Your model diversity for a lot of organizations is, you know, it’s an imperative. 0:11:40.143 –> 0:11:41.703 You’ve got Alex Karp out there. 0:11:43.343 –> 0:11:58.303 James Savage Blowing up on CNBC on almost a regular basis, talking about losing your alpha right about the specter of of losing an organization’s IP, Nadella refers to it as leaking your IP but the prospect. 0:11:58.653 –> 0:12:11.453 James Savage Act of you know you’re entering the most critical secrets. The organization into these folks models. What are they going to do with that on the other side of things? I mean LLM’s are LLM’s because they’ve been trained on human responses. 0:12:12.13 –> 0:12:18.373 James Savage You know, I’m not getting all doomsday on this, but, you know, model diversity is definitely a thing. And that idea of the harness. 0:12:19.423 –> 0:12:29.223 James Savage You know, for me it was a very compelling moment when Elon Musk, the world’s first trillionaire, spent his first shekels after becoming the trillionaire. 0:12:29.573 –> 0:12:39.933 James Savage On Cursor right I I in in a way I think that completely invalidated completely and totally validated Microsoft’s business model the hardest. 0:12:40.13 –> 0:12:43.653 Brian Haydin Which is yeah, which is to me like when I think about this like. 0:12:44.383 –> 0:12:46.423 Brian Haydin Model catalog model selection. 0:12:46.423 –> 0:12:48.583 Brian Haydin You know, how do I figure what the best ones are? 0:12:49.223 –> 0:12:52.303 Brian Haydin There’s a couple of technical ways that we solve this problem. 0:12:52.303 –> 0:12:58.703 Brian Haydin One is model selectors like. They have models that the model is actually built to select the model. That’s right. 0:12:59.133 –> 0:13:1.293 This job right Microsoft’s office. 0:13:1.293 –> 0:13:1.493 Yeah, yeah. 0:13:1.493 –> 0:13:2.253 And we’ve got that. 0:13:2.253 –> 0:13:5.13 Brian Haydin But copilot is essentially that, right? 0:13:5.13 –> 0:13:13.253 Brian Haydin So we’ve got auto mode using copilot 365 like it just you don’t have to say I wanna use cloud. I wanna use this. 0:13:13.253 –> 0:13:15.693 Brian Haydin It’s gonna figure out which one of these models are the best. 0:13:16.463 –> 0:13:28.863 Brian Haydin And and I agree with you from that perspective that that really what we’re seeing in the marketplace is validating Microsoft’s approach towards that. And then the safety, you know, safety parts of it as well. 0:13:30.173 –> 0:13:35.293 Brian Haydin And then I also am seeing a lot of this like local mom stuff, right? 0:13:35.363 –> 0:13:39.843 Brian Haydin We’re seeing hardware that’s being developed. It’s being commercialized. 0:13:40.163 –> 0:13:44.43 Brian Haydin I’m waiting for the new copilot plus PC to come out that you know as. 0:13:43.713 –> 0:13:45.433 Yearly so I guess. 0:13:45.773 –> 0:13:52.293 Brian Haydin Yeah, as like this one’s got a pretty nice set up in hardware, but but then we’re going to get into this whole global model selection. 0:13:52.293 –> 0:13:56.93 Brian Haydin Like now I got to figure out when do I do here and when do I go to the cloud? 0:13:56.93 –> 0:13:56.813 So you. 0:13:56.893 –> 0:14:1.213 James Savage Think about those gains and and both on marginal productivity, but also. 0:14:2.253 –> 0:14:3.133 Finance cost efficiency. 0:14:3.133 –> 0:14:4.413 Yeah. I mean it’s, you know, the whole. 0:14:4.963 –> 0:14:7.123 James Savage Game is is is changing. 0:14:7.163 –> 0:14:11.843 James Savage You know, the frontier firms were so comfortable with their trillion dollar capitalizations. 0:14:9.753 –> 0:14:10.273 Yeah. 0:14:11.843 –> 0:14:15.763 James Savage Until what was the what was the open weights model that came out? 0:14:15.763 –> 0:14:20.483 James Savage And it scored, like ridiculously close to where Fable had scored. 0:14:21.253 –> 0:14:22.213 For $0.15 on the. 0:14:22.213 –> 0:14:30.533 James Savage Dollar was far less was like 15 is like a point penny and 1/2 for for anthropics dollar yeah, there’s a lot of change going on in the marketplace. 0:14:30.533 –> 0:14:33.853 James Savage Open weights and all the source models are absolutely going to. 0:14:34.643 –> 0:14:41.323 James Savage Be competing. Microsoft’s got their Mai models which you know you know they’re going to be knit into the harness and. 0:14:41.843 –> 0:14:43.443 Brian Haydin Oh yeah, it’s gonna be baked. 0:14:43.443 –> 0:14:48.283 James Savage Into it, just like IT as we know, it leapfrogs on a continual basis. 0:14:48.283 –> 0:14:53.643 James Savage I mean, you see, you see anthropic and open AI leapfrogging literally every day. 0:14:54.3 –> 0:14:55.883 James Savage Or literally every other day, perhaps. 0:14:56.583 –> 0:14:57.863 James Savage But you know, we’re all benefiting here. 0:14:57.863 –> 0:14:58.743 James Savage Look at look at it. 0:14:58.743 –> 0:15:1.223 James Savage I mean, we truly are since we live in the age of abundance, right? 0:15:1.223 –> 0:15:11.423 James Savage At least when it comes to information and with intelligence. And that’s that’s really I think what’s most important is this intelligence layer. If you boil into Microsoft IQ. 0:15:11.653 –> 0:15:14.493 James Savage The value it has for business it boggles the mind. 0:15:14.493 –> 0:15:15.973 James Savage Truly it’s it’s. 0:15:16.413 –> 0:15:17.173 So go ahead. 0:15:17.693 –> 0:15:26.573 @3 Well, no. I was gonna have the the selection of models like the upgrade upgrading to the latest model isn’t necessarily free. I think anthropic released. 0:15:27.303 –> 0:15:31.23 @3 Their model that broke a handful of API capabilities. 0:15:31.23 –> 0:15:33.863 @3 So yeah, there’s there’s risk in that. 0:15:33.863 –> 0:15:41.783 @3 So I think I think the smart engineers are going to just stick with it has to again back to being a compelling reason to switch. 0:15:42.693 –> 0:15:45.453 @3 What we’re so I was getting ready for this. 0:15:45.453 –> 0:15:50.893 Brian Haydin I was taking a look at the, you know, all the model release that happened with it the last day of the month, right? 0:15:50.973 –> 0:15:56.453 Brian Haydin September had like 25 mod like major model announcements mean it’s almost one a day. 0:15:56.453 –> 0:15:58.293 Brian Haydin How do you even keep up with that right? 0:15:59.63 –> 0:16:9.823 Brian Haydin So you know you have that aspect and then I think one of the things you’re referring to and alluding to is that the, you know, we’re seeing the news and the noise around these new models and capabilities against the benchmarks. 0:16:10.573 –> 0:16:24.423 Brian Haydin And those benchmarks tell a little bit of a story, but they don’t tell the whole story. And you know some of these, like some of the models that come out, have a lower per token cost or, you know, kind of changes that dynamic of it, but it incre. 0:16:24.423 –> 0:16:28.893 Brian Haydin The amount of reasoning that has to go into it, so at the end of the day, you wind up actually spending more money than you. 0:16:28.983 –> 0:16:32.23 Fasten your seat. 0:16:29.703 –> 0:16:39.383 Brian Haydin Have a really weird dynamic you’re seeing, so you can’t just say, oh, this is the latest and greatest model I’m going to go ahead and use it and all of a sudden your costs are going to like, you know, spike through the roof and the. 0:16:39.383 –> 0:16:40.103 Way in which. 0:16:40.103 –> 0:16:40.463 You’re using. 0:16:41.253 –> 0:16:42.933 James Savage Is changing so drastically. 0:16:42.933 –> 0:16:43.813 James Savage You’ve got chat. 0:16:43.813 –> 0:16:45.773 James Savage You’ve got coord. You’ve got. 0:16:47.263 –> 0:16:55.23 James Savage Dev completion, you know, I mean it’s it’s it’s it’s it’s really mind boggling kuwa this idea of computer use agents. 0:16:55.63 –> 0:16:56.543 James Savage I mean, yeah, blew me away. 0:16:56.543 –> 0:16:57.423 At first a few. 0:16:57.423 –> 0:16:59.543 James Savage Months ago, when they first veiled themselves. 0:16:59.543 –> 0:17:1.63 James Savage Anthropics Co work. 0:17:1.63 –> 0:17:1.903 I mean absolutely. 0:17:2.863 –> 0:17:11.743 James Savage Mind blowing the the the technology open claw. Of course, being the the initial variant, but you know the industry pundits last week especially. 0:17:12.133 –> 0:17:18.293 James Savage We’re talking about overhanging the concept of, you know, maybe these frontier firms have. 0:17:18.293 –> 0:17:18.973 Gotten a little? 0:17:18.973 –> 0:17:26.973 James Savage Bit too far out in front of of the business requirements, you know, you know we know how technology often chases a problem. 0:17:27.743 –> 0:17:30.63 James Savage Albeit there’s some just miraculous things. 0:17:30.143 –> 0:17:32.583 James Savage Miraculous things being done in the field of mathematics. 0:17:32.583 –> 0:17:35.23 James Savage Miraculous things being done in the field of healthcare. 0:17:35.103 –> 0:17:37.743 James Savage I mean, it’s it’s mind boggling. A member of my family. 0:17:38.63 –> 0:17:41.743 James Savage I do a monthly update on cancer. 0:17:41.893 –> 0:17:48.533 James Savage Treatments he’s got chickens got can’t. He’s fighting cancer nobly. And this is the prospect of sending him an Astra. 0:17:50.63 –> 0:17:56.303 James Savage Researched all the latest findings in that form of cancer. I mean, it’s incredible. It’s mind boggling. 0:17:56.303 –> 0:18:6.303 Brian Haydin You made me chocolate here because there’s there’s a story that I read a couple of days ago about a consortium of mathematicians. 0:17:57.193 –> 0:17:57.913 You. 0:18:7.223 –> 0:18:12.223 Brian Haydin And they had said, like, basically the book, you know, these models are solving problems that had never been. 0:18:12.813 –> 0:18:13.933 How’s everybody? 0:18:13.933 –> 0:18:28.533 Brian Haydin 100 year old one year old problems and like the map, mathematicians are saying, well it’s not the solution to the problem that is really important. It’s the thought process and the you know how you solve it and the creativity and the you know so. 0:18:20.283 –> 0:18:20.603 Music. 0:18:29.383 –> 0:18:34.263 Brian Haydin You made it laugh, but they’re trying to protect lives, you know, for the residents a little bit. 0:18:30.3 –> 0:18:31.3 The dealership said. 0:18:32.533 –> 0:18:33.693 Oh, God. Risk adjusted a. 0:18:34.933 –> 0:18:38.853 Brian Haydin Yeah, but yeah, you know, the other thing I was gonna say is that like. 0:18:40.333 –> 0:18:46.853 Brian Haydin Taking a step back and just a pause for a second, I I think that Jim, you and I specifically you know. 0:18:48.573 –> 0:19:2.253 Brian Haydin Be were job that’s is supposed to go look at these models and make us and greatest in you running a company that’s really invested in this. We’re in this bubble of people at the cutting edge looking at this stuff and you’re a little bit more measured. 0:19:2.643 –> 0:19:8.563 Brian Haydin About it, you know running the operations to the team and not being an engineer and stuff like that. 0:19:8.683 –> 0:19:12.243 Brian Haydin And so I just want to acknowledge that we’re like. 0:19:12.243 –> 0:19:13.123 We you know. 0:19:14.733 –> 0:19:18.613 Brian Haydin It using these tools is is far different. 0:19:18.613 –> 0:19:23.533 Brian Haydin You know, at the average person doing the work versus the people that are sitting here at the frontier. 0:19:23.533 –> 0:19:24.773 Brian Haydin Let me give you a couple examples. 0:19:26.413 –> 0:19:26.733 Brian Haydin Or what? 0:19:26.733 –> 0:19:32.453 Brian Haydin I’m talking about so the let’s see opening. I just dropped the $200 Max plan. 0:19:32.723 –> 0:19:42.443 Brian Haydin And they’re kinda like, bring it back today, I think. And couple weeks ago I think it was 4th of July, I was running a couple of projects you know trying to get some stuff done. 0:19:42.443 –> 0:19:43.523 Brian Haydin I had two. 0:19:43.963 –> 0:19:48.323 Brian Haydin I had two or three different like projects running at Astra Max, you know. 0:19:49.213 –> 0:19:49.893 And for like token. 0:19:51.353 –> 0:19:52.273 Brian Haydin It was so Manding. 0:19:52.273 –> 0:19:55.913 Brian Haydin I’m like juggling these projects like you know, how do I give you the next set of instructions? 0:19:55.913 –> 0:19:56.593 Oh **** you’re done. 0:19:58.623 –> 0:20:13.223 Brian Haydin And I still couldn’t Max this thing out, you know? And I know people that can you know, but it takes. I feel like I know what I’m doing. And like the noise that’s going around and stuff is is observed absurd. You know, it’s there’s people aren’t actually using. 0:20:13.463 –> 0:20:15.943 Brian Haydin It these kinds of devils, and they’re really struggling. 0:20:15.943 –> 0:20:26.453 James Savage You and I, we’ve been talking this for a few months, you know, just the the paradigm, the method, you know, you’ve got Pmbok, project management, body of knowledge. 0:20:26.453 –> 0:20:36.533 James Savage Yeah, establishing the requirements within you know a well managed project, SDLC, software development lifecycles, you know we went from waterfall to agile. 0:20:37.263 –> 0:20:37.983 You know what’s next? 0:20:38.103 –> 0:20:47.103 James Savage You know, you stand up a bunch of agents in an organization that you know, roughly assigned to different users, or at least different functions of that organization. 0:20:47.253 –> 0:21:1.373 James Savage How do we control these things? You know, the idea of agent OPS is sooner than later. Honestly, I think the imperative that is that is management and I think you know the the harness makers. 0:21:2.303 –> 0:21:15.343 James Savage Microsoft cursor. Some of the others you know, these are these are top of mind because these are big issues in the organization. You know in the all in podcasts, discussions, the forum last week you know it was a big topic of conversation. 0:21:15.973 –> 0:21:22.13 James Savage Controlling these, you know for the organization and for the managing risk within the organization, this is absolutely an imperative. 0:21:22.13 –> 0:21:33.253 James Savage You’ve got an agent, you know replacing. Let’s say replacing people, but you know performing these super, super important, very sensitive functions within the organization. 0:21:33.943 –> 0:21:35.143 Those things have to be managed. 0:21:35.223 –> 0:21:36.343 Yeah. What’s a great segue? 0:21:36.343 –> 0:21:39.23 Brian Haydin Because last thing I want to talk about was risk, right? 0:21:39.23 –> 0:21:44.103 Brian Haydin Like bringing up like this concept of risk and you know, I’d say this is kind of viral. 0:21:44.333 –> 0:21:47.893 Brian Haydin Only because it’s been hitting the news that’s driving the policy changes. 0:21:47.893 –> 0:21:59.413 Brian Haydin But this is something that I’ve been talking about since the beginning of the year pretty heavily. An agent risk and what happens in autonomy matters and and it’s something that a lot of organizations are are starting to pay attention to now. 0:21:59.413 –> 0:22:2.93 Brian Haydin And the reason why is because we’re moving from. 0:22:2.823 –> 0:22:11.503 Brian Haydin This intern era of AI, where we’re just asking it to give me the information so I could do something with it to hey, I’ve got this task I need to get done. 0:22:11.503 –> 0:22:13.623 Brian Haydin It’s kind of Monday and I want you to go through this for me. 0:22:14.93 –> 0:22:15.413 Brian Haydin So we’re not trusting these. 0:22:15.453 –> 0:22:22.173 Brian Haydin You know these these agents to do things on our behalf and that takes that carries a lot of significant risk to it, right. 0:22:22.733 –> 0:22:33.173 Brian Haydin You know, if you are building an agent for CFO at an organization, it’s acting with those sets of privileges like you can do some pretty serious damage to the organization. 0:22:34.223 –> 0:22:43.783 Brian Haydin So that’s kind of the last topic is you know for me like when models can add you know, how do you know what you know, how do you control what’s going wrong and and then? 0:22:44.173 –> 0:22:46.253 Brian Haydin You know, how do you protect yourselves against that? 0:22:46.253 –> 0:22:50.13 Brian Haydin So, umm, Brad, I’m gonna put you a little bit on the spot here today. 0:22:50.413 –> 0:22:53.853 Brian Haydin Uh, and and I know that. You know, Nick will call him out. 0:22:53.853 –> 0:22:59.453 Brian Haydin Has been spending a lot of time looking at some of these guardrails and stuff like that and. 0:23:0.143 –> 0:23:10.103 Brian Haydin First thing I’ll ask you is has that how’s that changed the delivery in terms of the time that we’re spending or the things that we’re doing, you know what? 0:23:10.103 –> 0:23:12.103 Brian Haydin What’s happening with the teams aspect of it? 0:23:12.173 –> 0:23:13.93 Brian Haydin What are we doing with that? 0:23:13.443 –> 0:23:29.643 @3 Yeah, I would say it’s kind of shifting the workload of you know what it means to be a senior engineer, an architect because the development cycle is so much quicker. But the testing and all the prep work, the guardrails that you’re talking about. 0:23:30.413 –> 0:23:33.373 @3 To make sure something bad doesn’t happen once you put this into production. 0:23:33.693 –> 0:23:37.573 @3 I think that’s at least from what I’m seeing, that’s where kind of the shift in workload is. 0:23:37.573 –> 0:23:43.453 @3 You have to be a lot more careful, but you’re also getting the development done in half the time or 3/4 of the time. 0:23:43.683 –> 0:23:47.123 @3 So yeah, a lot of it’s. 0:23:47.123 –> 0:23:58.923 Brian Haydin Like big thing for me is, you know that these agents that we’re building and these these systems that we’re building aren’t just acting on behalf of Jim Savage or Brad Sapecki. They’re acting as their own agent. 0:23:58.923 –> 0:23:59.923 Brian Haydin They’re getting their own. 0:24:0.613 –> 0:24:3.333 Brian Haydin You know IDs. To me, that’s kind of a big thing. 0:24:3.813 –> 0:24:8.493 Brian Haydin Another thing that kinda is, you know, pretty new this week is copilot. 0:24:8.493 –> 0:24:14.933 Brian Haydin Autopilot is by getting definition. We finally seen what that really is. It was formerly. 0:24:15.83 –> 0:24:17.763 Brian Haydin Scout. I remember coming out of the hill. 0:24:17.763 –> 0:24:19.3 Brian Haydin That was like scout. 0:24:19.3 –> 0:24:20.843 Brian Haydin What’s the difference here? 0:24:21.3 –> 0:24:22.443 Brian Haydin Difference was the name. 0:24:22.443 –> 0:24:24.763 Brian Haydin There was not, you know, it was just the brand, right? 0:24:24.803 –> 0:24:28.883 Brian Haydin Yeah. So, but that’s actually starting to get some legs and people are starting to use that now. 0:24:30.133 –> 0:24:35.853 James Savage There’s a lot of thought around the different metaphors of chat versus cowork versus code. 0:24:36.13 –> 0:24:42.533 James Savage And you know, they’re integrating the statement from from Nadella. As you know, we’re creating apps all the time. 0:24:43.243 –> 0:24:57.963 James Savage Yeah. And then it it be wonderful to have some more clarity in that regard because I I think it’s it almost seems like window dressing, the distinction between trap and cowork in both anthropics and Openai open AI’s interface, but. 0:24:58.933 –> 0:25:13.253 James Savage Yeah, it’s exciting and certainly to have it to have a tight harness around, which is what Microsoft brings to the equation is they’ve got, you know, they’ve got this security safeguards agent OPS. It’s all there in the platform. 0:25:13.523 –> 0:25:17.923 James Savage As opposed to running around and picking them up piece meal. 0:25:18.443 –> 0:25:26.363 Brian Haydin And then we’re getting like we’re getting those models in, in foundry pretty, you know pretty quickly like we’re getting. We’re getting access to those models. 0:25:26.483 –> 0:25:27.963 Brian Haydin So I don’t, I don’t see. 0:25:28.613 –> 0:25:28.653 A. 0:25:28.773 –> 0:25:28.813 A. 0:25:28.813 –> 0:25:34.293 Brian Haydin A real desire need for us to build outside of that the Microsoft platform. 0:25:35.853 –> 0:25:41.613 Brian Haydin You get those you know, purview, safety, safety rails Agent 365, safety rails. All kind of built into it. 0:25:42.53 –> 0:25:43.773 Brian Haydin You you were starting to talk. 0:25:44.123 –> 0:25:54.323 Brian Haydin A little bit about like the differences between, you know, just chat versus low code versus pro code and that really aligns well with this idea of like the risk profile, right. 0:25:54.363 –> 0:25:56.403 Brian Haydin What I’m just chatting with something. 0:25:56.403 –> 0:25:59.83 Brian Haydin The risk profile from a governance perspective. 0:25:59.853 –> 0:26:7.373 Brian Haydin Is is different than if I’m building a custom purpose built AI that’s you know doing something, you know, intentional or acting on user’s behalf. 0:26:7.373 –> 0:26:9.453 Brian Haydin I have to have a lot more security controls. 0:26:9.493 –> 0:26:13.133 Brian Haydin Microsoft’s done a really good job of of laying out that foundation for us. 0:26:13.323 –> 0:26:17.43 Brian Haydin Yeah, you know, so we got a couple more, a couple more minutes. 0:26:17.43 –> 0:26:18.403 Brian Haydin Any last thoughts Brad? 0:26:18.443 –> 0:26:19.843 Brian Haydin Like from your perspective. 0:26:20.833 –> 0:26:22.513 @3 Not not a ton. 0:26:22.513 –> 0:26:24.673 @3 I mean, obviously the industry is moving so fast. 0:26:24.673 –> 0:26:30.593 @3 That’s why we’re talking about the pause from my perspective and my team’s perspective. It’s full steam ahead. 0:26:32.33 –> 0:26:33.593 @3 We’re still developing solutions. 0:26:33.593 –> 0:26:34.913 @3 We’re still putting them into production. 0:26:34.913 –> 0:26:36.793 @3 We’ve all seen challenges that we’ve had. 0:26:36.793 –> 0:26:46.33 @3 It’s just, it’s just a lot faster than engineers I think are used to, so I’ve had some engineers that just can’t keep up. 0:26:46.33 –> 0:26:47.33 @3 They just can’t get it. 0:26:47.33 –> 0:26:47.873 @3 Just doesn’t. 0:26:47.873 –> 0:26:48.713 It’s not what they’re. 0:26:48.723 –> 0:26:50.523 It’s not time to allow people. 0:26:49.833 –> 0:26:50.313 @3 It’s not time. 0:26:50.313 –> 0:26:51.113 @3 And then there’s the ones that adapt. 0:26:51.783 –> 0:27:2.783 @3 Yep, and and have business sense, sense and technology experience and can put it all together and implemented, you know, safely and our clients are happy. 0:27:2.823 –> 0:27:5.503 Brian Haydin So you brought up a really good point. 0:27:6.553 –> 0:27:10.953 Brian Haydin About like some people just are struggling with it and wanting your jobs to coach people. 0:27:11.313 –> 0:27:16.73 Brian Haydin So if you were giving people that are listening today any kind of advice? 0:27:17.833 –> 0:27:21.353 Brian Haydin What’s what’s 1 of the things that has been resonating with people as you get them inspired? 0:27:21.623 –> 0:27:26.743 @3 This technology just tell them to get in the project. Management is really no. I’m kidding. I’m kidding. 0:27:26.743 –> 0:27:27.583 I’m kidding. I’m kidding. 0:27:28.463 –> 0:27:30.943 @3 Yeah, I think it’s embraced the change. 0:27:30.943 –> 0:27:32.583 @3 I mean, I talked to these guys, yeah. 0:27:32.993 –> 0:27:33.193 David. 0:27:35.393 –> 0:27:40.273 @3 You just have to coach them into seeing that like this. Isn’t this isn’t gonna change. 0:27:40.273 –> 0:27:44.913 @3 This is a way of life. Your role, the way you do your job is is changing. 0:27:44.913 –> 0:27:46.193 @3 But it’s not gonna stop. 0:27:46.193 –> 0:27:52.873 @3 And it’s continued ever since. Back when we were, you know, and I want to say when, but a long time ago when we were doing solutions so. 0:27:51.303 –> 0:27:52.23 What time we go? 0:27:52.983 –> 0:27:53.903 We we’ve worked. 0:27:53.903 –> 0:27:55.543 @3 We’ve worked together. Yeah, yeah. 0:27:54.893 –> 0:27:55.493 So. 0:27:55.543 –> 0:27:57.543 @3 So it’s about adaptability. 0:27:57.543 –> 0:28:3.63 @3 It’s increased productivity and it’s just managing through that changes is what I try to help them understand. 0:28:3.63 –> 0:28:4.663 What are your part? 0:28:6.33 –> 0:28:6.513 You guys were talking? 0:28:6.793 –> 0:28:15.273 James Savage I was just reflecting on the fact that you, you remember this concurrency and safe net were the two fastest growing companies in Milwaukee in 2014. 0:28:15.353 –> 0:28:18.153 James Savage Brad and I were on stage talking to. 0:28:18.73 –> 0:28:18.673 We’ll deal with. 0:28:20.73 –> 0:28:20.83 James Savage You. 0:28:20.183 –> 0:28:21.863 James Savage About back then, what was it? 0:28:21.863 –> 0:28:22.823 James Savage It was the cloud, right? 0:28:22.823 –> 0:28:26.583 James Savage This new cloud thing and sets everything on fire, set our company’s on fire. 0:28:26.583 –> 0:28:28.943 Certainly. Good times. 0:28:30.433 –> 0:28:33.913 James Savage Yeah, and this is, this is definitely another one of those periods. 0:28:33.913 –> 0:28:35.113 James Savage It’s exciting as hell. 0:28:35.313 –> 0:28:36.713 James Savage Scary for some people. 0:28:37.63 –> 0:28:40.543 James Savage I do believe there’s an we’re going into an age of abundance. 0:28:40.543 –> 0:28:42.463 James Savage Good things are going to happen. 0:28:42.463 –> 0:28:44.223 James Savage People are going to find it. 0:28:44.463 –> 0:28:47.623 James Savage The white collar, the the SAS population isn’t going to happen. 0:28:47.623 –> 0:28:49.783 James Savage The white collar jobs aren’t going to go away. 0:28:49.783 –> 0:28:51.103 James Savage It’s just they’re they’re going to change. 0:28:51.833 –> 0:28:52.473 I think absolutely. 0:28:52.473 –> 0:28:55.73 James Savage You know, one of the point I wanted to make that I think to make was. 0:28:57.353 –> 0:29:4.433 James Savage Again, it’s an Adela comment. During the during the All in podcast he talks about the need for chain of thought. 0:29:4.433 –> 0:29:5.833 James Savage We need to know what these. 0:29:6.343 –> 0:29:12.23 James Savage With these frontier models are doing where they’re going, where the information is coming from, and how their reasoning. 0:29:13.553 –> 0:29:16.473 James Savage I really appreciate it is comments there and I hope the industry follows suit. 0:29:16.473 –> 0:29:17.473 James Savage We really need to. 0:29:17.473 –> 0:29:18.193 We really need. 0:29:18.193 –> 0:29:31.713 James Savage Audits and controls in these in these knowledge in this in these domains it’s, you know, without them things could get scary. And I noticed. I’m gonna hand it out here. The White House accord on Super Intelligence just. 0:29:31.993 –> 0:29:32.913 Yeah. 0:29:34.213 –> 0:29:36.333 James Savage Did you see this last night on the news? 0:29:36.493 –> 0:29:37.693 But you know more here. 0:29:37.693 –> 0:29:41.773 James Savage Implement more robust internal controls to monitor the capabilities and alignment of the models. 0:29:41.933 –> 0:29:47.733 James Savage Empower an internal team to ensure all of the controls partner with internal external auditors. 0:29:48.213 –> 0:29:51.133 James Savage They’re thinking about these things, and this is one way I think. 0:29:52.93 –> 0:29:52.653 We’re well on the. 0:29:52.653 –> 0:30:2.333 James Savage We’re well on the path to getting the controls right. If anything from this pods to come out as positive, it’s it’s going to be these additional controls. 0:30:2.403 –> 0:30:3.643 James Savage Which which are imperatives. 0:30:3.643 –> 0:30:9.443 Brian Haydin They are their imperatives, which so I’ll leave it my thoughts like this. I’m on way down to to Saint. 0:30:9.443 –> 0:30:12.963 Brian Haydin Louis to do in the cloud and AI Summit conference. 0:30:12.963 –> 0:30:20.563 Brian Haydin I’ll be speaking tomorrow. Tomorrow morning and Friday afternoon, and I’ll be talking about this and what I’m trying to do is give people the tools. 0:30:21.213 –> 0:30:27.93 Brian Haydin And the understanding on how to actually develop in these these spaces without putting the brakes on. 0:30:27.253 –> 0:30:32.653 Brian Haydin So don’t wait on these frontier models to do that. Implement evals implement. 0:30:32.883 –> 0:30:34.483 Brian Haydin You know proper controls. 0:30:34.643 –> 0:30:43.363 Brian Haydin One of the things that I’ll be talking about tomorrow is how to be able to test against, you know, model A versus model B and and kind of get out of that signal wise, right? 0:30:43.363 –> 0:30:45.643 Brian Haydin So hey, we’re out of time. 0:30:45.643 –> 0:30:46.963 Brian Haydin Thanks a lot for joining us today. 0:30:46.963 –> 0:30:47.763 Brian Haydin Viral topics. 0:30:47.763 –> 0:30:48.763 Brian Haydin We’ll be back next month.