IBM DB2 Genius Hub, AI Observability and the Future of IT Modernization

Control, this is Agent Dev. I'm in position. Copy that, Dev.

Stand by for go. Standing by. Hey, everybody.

Welcome to another episode of Agents of Dev podcast. I'm Mitch Ashley, one of your co-hosts, along with my friend and co-host and colleague, Brad Shimmin. Welcome, Brad.

Good to be chatting today again, again, and again. Hey there, Mitch. Good to see you.

Uh, it's, it's a nice almost spring-like day, one might say, at least in terms of the calendar. So I'm, I'm going with that. So in Boston, there's one...

A nice, almost a nice spring day, you mean you only have to do one shoveling duty to, through the day? Is that what that means? Yes, that is precisely what that means .

But this winter it does. This is the hottest winter on record in Colorado since 90... Since like 93 years ago, or some- Yeah ...

some absurd number. It's, we don't have any precipitation to speak of. Just a little bit, but- I, I worry about the snow pack in the Rocky Mountains.

I really do. Yeah. It's very low.

Yeah. We need, we need to get dumped on, and I mean snow. But, in the mountains particularly.

It doesn't matter if it comes in Denver as much. But, good stuff. Well, that's our weather report for today so- Good.

It's always good to keep track, I, I say. Yeah. It doesn't hurt.

It doesn't hurt to catch up. Um, so, let's kind of jump right into our, our first segment here, and I think you had a few things you wanted to chat about on the call-out. I do.

So why don't you start? Yeah. The call-out for me this week is an announcement that may, may have gone under the radar, and it shouldn't.

Uh, last week from... I, I think it's last week, depends on when this is going out, but I know for sure it's, it's out. And that is, IBM, and their DB2 database, which is a very storied database, that is- Oh, yeah ...

very important. I remember when it came out. I was an early user of DB2, for sure.

Yeah. Me, me too. Yeah.

I, I... It was... It's, like, so straightforward.

I really appreciated that about it. Um, so, so on, March 5th, they will have had launched their DB2 Genius Hub, which is, a step or a stepping stone toward full automation, but they're not pretending to have full automation of the DB2 database. So all of you DBAs out there, you know, rest easy.

It's not meant to, to displace you. It is, it is really... They're, they're...

They have five tiers that they, or, objectives, steps if you will, that they wanna go for. And they're really launching this with a, a modicum of, of what I would call pragmatic autonomy, meaning it's, it's completely in the human's hands. Mm.

Uh, it is, based upon their very extensive knowledge of DB2 installations over the course of just a few years, to help you, for example, go from what might be many hours of root cause analysis to, to basically finding out what's wrong in a matter of moments. Uh, and th- that's a lofty goal. And if you can do that with, without in- increasing risk or introducing risk, then yeah, my, my hat is off to you, and I think that they're gonna do it with this.

I think that they have the skills. They have been working toward this for some time now across all of their portfolio. But particularly focused on DB2 here, I, I feel like I- IBM is, you know, moving toward what Oracle initially, had the great idea of with their autonomous database, and that is a database that, you know, doesn't take an army to just keep the thing running.

Oh, yeah. 'Cause- Yeah ... I mean, right now, you know, if you, if you talk to a data professional, they're like, "I, I very much would love to have a l- you know, stratum of data, without the data management.

I just want the data. I want it to be accessible. I want it to be clean.

I want it to work. " So if they could do it, I, I would love that. F.

Codd and all the kind of early- Mm ... relational, relational gods, if you will. I first remember, you know, I, I still remember- Yeah.

Oh, yeah ... you know, third normal form, so help me Codd. You know?

And all these kind of funny terms we used to see. Well, I mean, and that was the problem with, you know, Oracle Autonomous Database initially was, was, you know, people... If you wanna do this, you need to sort of have your act together, and not every company does.

Mm-hmm. Mm-hmm. Yeah.

It was, this was a big change. But, you know, so, so, so can you kind of put in, in summarize, like what does it mean when we have a DB2 that's autonomous? What d- what does that mean?

It's, it's like going to the store to buy groceries? It's up to buying a new car for you? Or is it like, you know- Yeah ...

re-indexing itself and, you know, correcting broken pointers and references and things like that? So the way, the way it's gonna launch, it i- isn't going to sort of ascertain which groceries you need and then go buy the groceries. It, it's going to basically, you know, be with you as you are shopping at the store and say- Mm-hmm ...

"Hey, Brad, it looks like they're out of your favorite bread. " 'Cause there's- There you go ... there's an option over here.

So it's very much in, in the control of, of the DBA, to, to basically set the tone, to decide what they want to do, and then shepherd, which is, you know, a key word for us for this year- Mm-hmm, mm-hmm ... to shepherd that process so that they have trust in, in the outcome. So when you're walking down...

" And it says, "By the way, it's on the end cap on aisle four. Just if you go back there you'll find it," 'cause they always put in everything in two different places, right? So- They do.

They do. Yep. The mysteries of the world.

Well, that, you know, it's, it's amazing, you know, to have a career that spans, you know, the introduction to DB2 and, you know, to where we are now, with it, with AI and, my God, it's, you know, what a, what a fun- Seems like it was just yesterday. Yeah. Yeah.

It was just yesterday. I was loading the... re-indexing those files, fixing those broken-References, referential integrity, all that kind of stuff.

Yeah. But, but it's- And that's what this proposes to do, is fix- Yeah ... things like that.

To, to make that something you don't have to paint over every day. " Interesting. How do you think, how do you think IBM is addressing the, accountability, the trust issue?

Now we're dealing with data, you know, you don't mess around with data, you know, find out kind of thing. F around- Yeah. I mean- ...

and find out what you do... mess with the database. Right.

Do not mess with databases. Um, yeah, I think that it's reflected in their rollout plan. So the, the March 5th rollout is, as I was mentioning, something that, you know, is, is just a baby step toward level five, which is full autonomy.

Um, wherein the database would say, "Oh, I need to do, apply this security patch. Uh, oh, the shard I have for this region is, is it going to get oversized in, in 30 days or tomorrow? " That's level five.

This is, this is, you know, very much a tool that, is trained on their knowledge base, their institutional knowledge of DB2 installations. Mm-hmm. And that's a big deal.

And I feel like that-that's an area of IP that we're going to see increasingly, valued, within our, our industry. So those... you know, that institutional knowledge, the domain expertise, if you will, and what that means for software that, that we rely on every day.

So, yeah, I think, I think they're taking the right approach to it. They, they have the right foundation. They have a much more, you know, pragmatic, as I said, you know, approach to doing this.

Mm-hmm. So I, I have high hopes for them with this, with this endeavor. You know, it's, it's a theme 'cause m- 'cause my call out is about a couple of things.

" "We're gonna... We can replace all your IBM COBOL for you. " And you know, you know- Yeah.

and that's like, to me, that's like saying, "We don't need developers anymore. Everything's gonna be in the cloud. " You know, how many times do we do this?

The next thing will be the savior- Every year ... of everything else, and nothing will de-exist after this gets done. We're doing the same thing with AI, so you know, I, I'm a pragmatist.

Yeah. Not a skeptic, but a pragmatist, but also like a realist. I like to make it happen.

Anyway, the point being is, you know... So I, I wrote... You know, I kind of watched this happen.

" Well, it didn't. It kept going and, and IBM countered and said, "Yeah," you know, "baloney. " So you know, I...

Speaking of DB2 and all those things, one of the things if you've been, been in IT for a while, you may not have called it modernization, but we went through these series of re-engineering, redesigning, creating the next generation of whatever apps- Yeah ... you work in, claims processing, billing, banking, whatever business applications. And sometimes it was upgrading to a, to a new database technology.

Sometimes it was just replacing it, 'cause like I worked in telecom and a lot of systems came from, from Bellco, which was old Bell Labs, and, people were like, "This is great. We've run this way for, you know, 25 years, but we need something more modern. " Every one of those projects- Mm-hmm ...

virtually fails because- Inertia? it is such a monumentous scope. Yeah.

It is just... And, and I made up this axiom of never be the first or the second project manager on any modernization project- ... because the first one, the expectations are so wildly out of, out of scope, they immediately fail, so it must be a leadership problem.

The second one comes in, gets things kind of back on track, sort of going, but not quite enough to make everybody happy, and so it must be a leadership problem. And then, you know, third one, Brad or Mitch shows up and said, "Okay, I'll help you. " And we kind of...

We've, we were able to build up enough of the problems that got solved and fixed the things that need going, keep going. Now- So what you're saying- What, what you're saying, Mitch, is, is that for all the project managers that get hired, fired, as we increase, as we go down that slope, if you will, the amount of debt that they inhe-inherit decreases. The- That's right ...

it is. So, so you really want to find yourself that optimal spot on that slope of decreasing debt, and, and inertia that you want to tackle for a, a new job. Well, I live in Colorado.

It's a lot easier to ski downhill than it is uphill, so. Unless you're on a ski lift. Then it's pretty easy.

Unless you're on a ski lift and you enjoy cross-country, which I don't. Um- Yeah. Don't, don't put me on this.

So, so, so I wrote my own... You know, I can't sit on the sidelines any longer. I'd like to...

Let's, let's stop. That's... This whole you're gonna re-rewrite all my COBOL.

Okay. Writing the code is that big of a problem. It's all of the, what's the business process?

While you're rebuilding it, you're gonna much, make a bunch of changes. Well, what's the architecture- Yeah ... of what you built and how does it really function, and does that, do those people still exist?

So to your point, and by the way, my post went viral and this had, like, over 40,000 views on it and, you know, once in a while things get up into that number on my LinkedIn, not very often, but for me, that's, that's a big deal. But, it's, to your point, it's that institutional knowledge. Yes, that AI can both help, ingest and become part of it, its skills.

That, that doesn't make up for the intuition, the knowledge, the I can recognize- Mm-hmm ... when this is a problem that other people can't. Yeah.

Because- Yep ... we've, we've lived this experience, either living with the application, living with re-engineering or modernization or whatever it might be. Um, so it takes, it really takes a team.

It takes people that are good in business process re-engineering and rethinking- It's diverse Being able to do change management 'cause while you're changing, you're, you're modernizing, guess what? Other people want a bunch of other changes too, so a lot of things pile on all at the same time. So it isn't just throw it at Anthropics or anybody else's model, and now you've got your problem solved.

It, it takes, it takes a lot of elements to succ-succ-successfully complete a, a modernization project. So that was my soapbox and- Okay ... people- We're jumping in.

Yeah, yeah, yeah, yeah, yeah. " And, you know, I, yeah, I agree, I agree that that could, you know, help and would help and is very helpful to use. But, you know, I think we're seeing right now in play out in real time this rediscovery of the value of that human insight and experience- Mm-hmm ...

that you're talking about, and you can see that reflected in companies like IBM, who, I think in the last week started rehiring or just started hiring newbies. You know, let's bring some people in on the ground floor and show them how this works. It might help us- I learned a new term, EICs, early in career.

EIC, that's the name we're doing. Oh, yeah, okay. Yeah.

Yeah, yeah. So we're, we're LICS. You're late in your...

Yeah, or, or, uh- We got- ... won't leave our career, whatever it may... You know, it's, it, and that's, that's the perfect point because, again, it's sort of the overro-- it's the overrotation on what technology- Yeah ...

will do and how quickly it will do it. Yes. Yep.

" It'll be the yeah, but agents. "Yeah, but what about this? " That's, that's not gonna work.

You know, all those kind of things that the engineering mindset is, "I don't say yes until I've eliminated all the nos," right? Okay. Once I know- Yes, right ...

these problems will be solved, then I'll say that's the right answer. It's kind of that mentality. " Pick, pick whatever the next generation of, of technology is, that's gonna be the thing that replaces everything else.

It just doesn't happen that way. And yes- No ... this time's different.

There's a lot of things that are different about it, but- Well- ... inertia's a big thing. You know, I don't think it's gonna solve all today's problems tomorrow.

Several tomorrows from now it will. No, it shouldn't. I mean, we, we have to learn that the hard way as a species time and again.

Um, just- I know ... ask anyone from Quantum. Uh, you know, we- Mm-hmm ...

we know that it's going to happen, but how many times have we overrotated on that? Um, it's, it's the nature of- It'll be next year. Next year.

Right. No, it's 10 years. Well, which is it?

Next year or 10 years? Which... Where are we here?

Well, it, it's, it's a generational thing, unfortunately. You know, the Thomas Kuhn wrote a book called, um... Oh, gosh, now I'm just blanking on it.

It's, it's, the, the revolutions in science are very- Mm ... much episodic in terms of they only change every 60 or some odd years. Mm-hmm.

Why does... What, what is that number, 60? Why is that magical?

Mm. 'Cause that's how long the average, you know, industrialized people, people live. And, uh- Okay ...

it is very true. You have to have a changing of the guard, for, for real change to happen. There, so I did write another paper coming, another report talking about the, the depletion of the talent, talent hiring supply chain- Mm.

This junior developer. Yep. Just like last year, we didn't need any senior deve- any developers at all because AI would do it all.

This year we're, all we're developing tools for is developers, frankly. And, and now we're starting to realize, well, wait a second, you know, that all's not gonna happen so fast. We're not gonna have all that stuff in place with AI.

We still need people. And role may change. How they develop and grow them may change.

And there was a paper- Yeah ... that came out in, Communications of H- ACM. If you're not an ACM member, it's the Society for Computer, Computer Management.

Um, that, that, talked about... It was actually two people from, Microsoft, in their, I think it was developer relations and also one of their development organizations, an Azure CTO person. They came up, they came up with a model.

It, it's called the preceptor model, which essentially is what is sort of this mentorship, it's almost like peer programming. How many s- what senior person and how many junior people can that senior person, I would say mentor, but work with, and then someone else who's helping both making sure that those people get the r- well-rounded experiences. But not just- Mm ...

on old ways of doing things, new ways of doing things, things too, so they accelerate and become senior engineers more quickly, but it's based on not just their own experience, which is all valuable, it's also based on what they learn on the job and they learn from more senior people. " Well, it's, it's important to remind ourselves of that. And as, as you just mentioned, I, I think that one of the most, you know, accelerationist, real accelerationist ideas out there is the, the truth that any discovery is built on every discovery made prior to that.

Mm-hmm. And that you c- Mm-hmm ... you're not just throwing out the old and starting over.

That does not happen. That does not work. Mm-hmm.

So I'm glad- Mm-hmm ... to see companies like IBM trying to reestablish that cadence of, you know, new people coming in and learning, and then building on that, that knowledge. Yeah.

And Microsoft too. I think they're, you know, a little more visible- Yeah, yeah ... but I think for IBM, but Microsoft at, at, at Build last year was very clear in their, "We're on this journey together with developers," right?

"You're not going away, neither are we. We are all builders," et cetera. So that, that kinda jumps to, if I can segueUm, one of the main topics that I wanted to talk about is I just released something called Observability Native.

So let me step back for a moment. Um, some of the things that have changed about, yes, we're using agents to create software, and we're using AI in the development process. Stats from our, from our research show ninety-three percent of organizations either are using, heavily rely upon, or, or considering, there's only about thirty percent considering out of that number, using AI.

And this is from, you know, last year when, when a year ago when it was maybe forty percent of people were sort of somewhere in that mix. So in one year- Yeah ... it's doubled.

More than doubled. That's a big jump. It is.

And, and s- the things that are different this time are we're using AI agents to build agents, right? So kinda ask the question like, if we need to... The level of accountability we need to have for AI software, whether it's what's built or the software that builds what's built is, is extremely important for getting AI into production, right?

We're starting to hear about- Mm-hmm ... a-agent accountability as well as agent governance. We're starting to hear governance.

We're hearing about agent behavior guardrails as well as security guardrails, et cetera. " So that caused me to say, "Well, maybe it isn't tilting at windmills this time, but how do we get observability built into the whole process? And isn't that really gonna be fundamentally necessary?

How do you know why it built the software the way it built it, and what caused it to... an agent to go those directions during the development- Yeah ... " So long story short, that's where the windup of why did I do this.

I created this concept of observability native. It's not a term that's very w- you know, commonly used, observability and native are, but it, it kind of says it in the name, which is, think of it as three D observability, not shift left. It's observability everywhere, it, from the...

We're working on the specs and the planning stages of using AI in development all the way through into operations, and the governance and the provenance and the agent behavior and the a- and the controllability of agents- Hmm ... in production. And I think we're at a, a th...

Again, I don't think I'm tilting at windmills, if you're familiar with that terms, which is kind of like spitting in the wind, right? Like, yeah. Good luck with that.

I think we are at a, a time where because we're using AI in the process, and we have to secure AI and control it and, and hold it accountable while it's developing software, that will con- You see vendors racing earlier into the life cycle to try to get observability- Oh, yeah ... and security and a number of products into the development cycle. Again, developers going away, but we sure are building a lot of tools for them, for something that's going away.

Sarcasm. We, we should know what those tools are doing. Yeah.

Well, yeah. Kinda wanna know. It might, it might be useful.

Right. You know what reminds me- So, Mitch... Oh, I'm sorry, man.

Yeah. I'm sorry, please. No, no, no.

Please, please. I'm, I'm... I don't mean to hog the microphone, but I am hogging the microphone.

Well, you're... I know you're excited about it, so I'm, I I, I'm with you. 'Cause I, I am as well, man.

And I, and I, see in our research this, this reflected as well on the, on the data side. And, but I'll, I'll get to that in a second. But what you reminded me of- Mm-hmm ...

is the example we were giving for IBM earlier with just trying to- Hmm ... do root cause analysis. If you only know the database and you're looking at problems that are happening in the database, like latency for a query, maybe it didn't happen in the database.

Maybe it's in the- Hmm ... front end library. Maybe it's at a gateway.

Maybe it's, you know, your EC2 instance on AWS is down and you don't know it. If you don't have all those, if you don't have transparency and observability across all of those participants in that final outcome, you're, you, you can't say with any level of assurity what's happening, let alone solving problems when they occur. So- Mm-hmm ...

this idea of, of, you know, having observable... Sorry, observability native. Woo.

It's like cloud native, but observability native. There you go. Yes.

You know? Mm-hmm. Built for observability, is, is critical, and it's, it's...

You know, we, we... It was one of, one of the two thousand twenty-six prognostications we, we had is that everything is gonna be oriented around that FinOps idea of how do I, you know, control my spend? And you can't control your spend unless you have observability.

You start- Mm-hmm ... with observability, and then you, then you could just peek at FinOps down the road. Mm-hmm.

But if you don't have- Mm-hmm ... the observability, it's, it might as well be behind a wall or over the horizon. Well, you know, and that's...

It's a really good point because a lot of what goes into software engineering, I call them engineering problems. When you're s- when you're on a project, there's certain things you don't know how they're gonna work. Yeah.

Problems you haven't figured out yet what you're going to do, whether you're using AI or not. And AI might help, might help you solve that, but one of the factors is always, and how much is that gonna cost? How many times am I calling the model per second, right?

Right. So how... What, what's my credit card gonna look like at the end of the month if you're on a personal plan or, or, you know, what's my bill to the cloud provider or the model provider at the end- Yeah ...

of the month for my company? Because I made some choices that had huge implications on the financial end of it. And that's actually, um...

So one, one of the stats from the researcher in, in s- software lifecycle engineering is AI related areas, AI observability, agent observability, cost, FinOps of AI, um-Are, are three of the four top things in AI related in, in the what am I looking for out of an observability solution? And not just in production, but in development. So, and that's gone from one thing in the top ten about AI observ-observability, mostly about automation actually from a year ago.

So that, that's, that's how quickly we've gone, "Well, wait a minute. Here's... " That's kind of inevitable, isn't it?

Because if you can collapse the time to value with agentic tooling- Mm-hmm ... let's say you work at Anthropic and you decide, "Hey, let's make something called Claude Cowork," and you do that in under ten days, you cannot tell me that you have, you know, thought about or addressed all of the Rumsfeldian unknown unknowns- ... that exist within that software.

Mm-hmm. So it's a day two problem, but you need to build it on day zero, you know, to, to anticipate it on day zero, and you do that by understanding how the thing works. Well, and to...

That's a great segue, because in creating this idea... And, and observability native is, it's not a product, it's not a s- piece of software, it's not an open source project. It's really a reference model of tr- Yeah ...

how do we understand what's happening in the market, and who's addressing what problem, and what problems do we need to have addressed to put AI into production at, with full accountability at, at scale. So I just took the agent process of, of... Because it's non-deterministic, there's some special things we have to do.

And if you think about what every, every agent does, just being real simple about it, and that's the purpose is make it simple, right? Is, is I created this four-step cycle, right? So the first one is intent.

What's the goals of what the agent was trying to do? What was it told? What was it...

What, what was the data? What was the prompt? What- whatever it was that sparked- Mm-hmm ...

this agent to, to consider taking an action, there's a reason why, and what was that reason? I wanna, I wanna know that because then I can understand what it was trying to do, right? Right.

" Well, that agent may change on, on every execution, so you've gotta know about that. Second step is the reasoning. So in figuring out what it's gonna do, how it's gonna solve that problem or complete that task, what did it consider?

Why did it, why did it decide that this is the path I'm going to go? Something in the goal, something in the response from an LLM, data, other information that it had. md file of, of whatever operating system it's working in.

Pro tip: always review your soul MD file regularly. Same with your Claude MD, your agent's MD. Don't just let that sit.

Exactly. Yes. That wasn't a made up example.

That's a real example. Right. And then speaking of which, you know, the third step is, okay, what constraints were applied?

Well, I had all these options, but those aren't practical. Those aren't doable. I'm not allowed to do those things.

Um- Mm-hmm ... what guardrails am I operating within, and what was applied to it? So it also tells you if the guardrails are working or not, right?

Then the fourth is what the outcome is. What happened? What is the cycle?

And that, that's kind of the process every agent is gonna go through. And you might say regular software does that too, but it's deterministic. You can look at the code- Yeah.

Very different ... to see what it's doing, right? Very different.

So that's one element. Yeah. And then I created these seven principles, and I'm not gonna go through all of it.

But the, probably the main thing is treating agent telemetry as kind of a first-class signal all the way through the process from beginning to end, whether it's through the software- Right ... to operations, or it's through the cycle intent through outcome. Um, we have to have all of that to really have accountability, and we can.

I think that's, that's possible to do now because of how we're building AI software, agent software, and using agents to do it. Well, can I ask, Mitch, do you, do you think- Yeah ... so this is coming from a person who used to suffer through SNMP traps- ...

trying to figure out why things broke. Um, do you think that- Version two or version three, but okay. SNMP 2 or 3.

No, I'm just kidding. Oh my God, I can't even remember. Um, but, I've, I've tried to put it behind me.

Um, so yeah, open telemetry, which I, I'm seeing show up quite a bit in agentic tooling right now. Um- Yeah. Do you think that that is, up to task to accomplish what you're describing?

Uh, do you think that it is something that is a standard that we can apply in every layer of that tool stack, or value chain, whatever you wanna call it, to, to accomplish this? Um, it, it definitely is a huge part of it. I don't think it's the only thing.

And again- Mm ... observability native isn't throwing out what we did with observability. It's like taking it to the next level.

So... And this is already happening- Yeah ... in the OTel community, where they are building the reference model for how da- AI data is shared, right?

Building how do we, how do we manage agents- Yeah ... as well as monitor, et cetera. And they're, they're step...

I mean, these are the vendors that are driving this effort primarily, and just like we talked about vendors rushing to be earlier in the development cycle, that normally would be an operations tool, suddenly an observability tools like for developers, that's happening in the open standards as well. So, I fully expect, confident OTel community will, will be right there and a huge part of the solution. But we'll also see n- new innovations too, maybe m- new open standards, maybe, of course, a lot of vendor innovations- Hmm ...

that happen with it too. You think like edge cases to, to fill in gaps, things like that initially? Things like that.

Well, fill in gaps. I mean, because even with an open standard, it takes a while for that to get baked in, right? Vendors are gonna come up with new ideas.

That's BCM. Some things- Yeah ... they'll submit back to the Linux Foundation.

Uh, they'll keep making products out of it. They'll get acquired. It gets puts in, put into larger company projects.

Yeah. They'll elevate it. It's that whole cycle will continue to happen.

I think the, the main thing is-Is the, is the impetus, is the reason to do this sufficient with financial motivations to make this happen? Yeah. And if you believe that for AI, AI agents to operate at scale in production, that you have to have this accountability in place, governance, security, guardrails, behavior, all these things.

If you believe that's true, which I do, it won't be a perfect, but that is a huge requirement for enterprises, the industry responds. So I think the, the things are lined up to move us forward towards having observability throughout the life cycle. Yeah.

I, and I think the impetus is there, and just ask anyone who's using frontier models right now, you know, API services. It's, you're just one bad bill away from colla- You're crushing your project 'cause you just didn't know what it was doing. Mm-hmm.

And it was doing something wrong. Well, it's the- Yeah ... sort of the, like, security people hate ephemeral things because, like, well, what went away?

What happened when it was here? That environment's gone. That serverless thing is gone.

Do I have what I need to know what happened? And that's the problem we're solving. So I, I jumped in there.

You were gonna go somewhere next. Oh, I was just gonna complain more about, about inferencing. And I don't know if...

Do you wanna, do you want to, transfer to, to the drop section? Drop. Yeah, okay.

It's time for- And we can- The drop. Okay, it's time for the drop. Go for it, my friend.

D-d-do... Yes. I, I want to ask a question.

Um, why, is the inferencing, you know, we, we have many choices now for, for, you know, hosting providers that are offering API services for frontier models. Take your pick. It can be GLM- Mm-hmm ...

whatever. Um- Mm-hmm ... and I want to understand why, why those are turning into Netflix.

Uh, a-and what I mean by that is in terms of availability, performance, quality, everything is, is geared around optimizing the spend of that provider, not of me, the customer. And that- Mm-hmm ... upsets me.

Uh, and this is- Mm-hmm ... this is, like, top to bottom. I'm not just talking about the neo clouds here.

I'm, as, as a user of, you know, a, a certain large hyperscaler's, you know, generative, you know, mo-model repository. Uh, it's been very frustrating over the last couple of weeks to see, some, some, like, degradation in quality with, you know, 503 errors popping up day and night. Uh, when- I know, yeah ...

you're like: Why is this happening? And the Reddit- Mm-hmm ... Reddit sphere is like: Well, that's just because nobody could use Anthropic now inside of their OpenClaw implementation, so they're just, you know, pinging this model, you know, to, to to get whatever their email inbox is.

So they're, they're DDOSing this, this platform. The next platform that will block OpenClaw soon. It's just, right, it's just a game that, that keeps, p-passed, you know, the hot potato, I guess, in a way.

And, but it's not just that. It's not just the degradation. It's sometimes purposeful.

You know, if you think that, you know, OpenAI or Google or Anthropic or any of them are presenting consistently the exact same model with all the parameters available without quantization to everyone every hour- Mm-hmm ... of the day in the same way, you're kidding yourself. You know, it is, it is, you know, very much a game of optimization for them, and they have to, you know, because these are, these are precious dollars in terms of watts that are available to, to spend in each data center.

And so I, I fear and I feel like I, I... it's, we're sort of heading in this direction where, you know, you would have to demand, have an uprising to demand an SLA from a, a provider because they're, they're like: Well, you know, you pay us $200 a month, but y-you might not get what you're paying for today. Mm-hmm.

How can that be? Now just, just think if I had observability into what the agent's doing trying to call those models. As the consumer.

I could hold them accountable. Yes, not as the provider, 'cause they already got that. Mm-hmm.

Yeah. Well, you know, it's, I don't know if th-this is an imperfect comparison, but it's kind of like broadband, right? You, you, you have one gig at your house not because you can run it at sustained one gig full throttle all day, every day, 24 by seven, right?

Yep. They design infrastructure to be oversubscribed, just like an airplane seat, right? Correct.

Yeah. They, they oversubscribe those. They oversell them.

And same thing with the models, right? For your $20, your $100, $200, whatever your subscription is, there's a certain amount of usage they're expecting you not to use, right? That you're gonna be somewhere in the lower- Yes ...

the middle, maybe the upper end sometimes. But you're not gonna be running, you know, redlining full throttle every month. And that's, that's a, a lot of this...

Well, OpenClaw is gonna, like, push that way to the limit, maybe past it. That's absolutely- And that's why they throttle it back. Yeah.

Yeah. So they push us way beyond, and they go, "Well, wait a minute. Hold on.

You know, that's $1,000 a month if you wanna run at that speed, or we'll just block it because we can't control it. It went out of... " So that, that's sort of the...

You know, when you're on the consumer end in a telecom side of it, you look at your bandwidth usage over time- Mm-hmm ... and you look at the peaks and the valleys, and you look how much you're consuming. So you know with...

when you're, when you're in, within- Right ... your SLA that they should be meeting. So you provision for that.

Right. Right. Yep.

And that's what you expect. The, the other part of it, I was just, talking with, you know, speaking of junior engineers coming into, to our industry, I was just meeting with someone yesterday who was working on their own OpenClaw project, and they're like: "Well, how much do I budget f-f... "For, you know, vibe coding or any kind of development is you need to ask and develop and work with AI to create a budget.

I do this for all the time. Yeah, yeah. Like, okay, now how much is this gonna cost if I have this many companies coming in, and this many orders, and this many videos or whatever I'm doing?

Expect this kind of load. Give me a cost and where the cost is and why that's gonna cost that much. And then as we test it, how come my budget was like I wiped out my 20 bucks in a day and a half?

Hold on just a second. Where did that, where did that occur, right? Well, right.

And that, that's... those are things that we can kind of control as creators. But, but, you know, when you have an asset that you rely upon for what-whatever s- you know, what do you call it?

Uh, you... I, I have an SLA that I want to adhere to for my customers that's latency of no more than two milliseconds per query, for example. Mm-hmm.

Mm-hmm. If I can't rely on the infrastructure behind that to provide that if I'm paying for it, that, that's unnerving, annoying. I, I want, I want inferencing to be more like critical infrastructure is, is my entire point of this rant, is that I, I, I feel like we're, we're turning it into Netflix sort of, you know, attention economy, and it shouldn't be that.

It is critical infrastructure, and it needs to be something that you can count on as a creator and developer. Yep. Those...

And when you get those edges, those hard boundaries of they're not supposed to be there. I remember- Yeah ... I first started using AWS.

The first week I got this message, I was asking for, whatever size survey. " Like, "Hey, wait a minute. What happened to the elastic cloud that's infinitely available?

And then, like, I can get whatever I want when I need it. " No. Right.

There are limits- Only for this T-shirt size. Yes. Come back.

Come back in 30 minutes and try again, right? That kind of an answer. Mm-hmm.

Mm-hmm. Exactly. So, so I want...

The, the drop for me is, this is again kind of the evolution of where we're heading and why. Um, Google came out with, some announcements around their Google, their agent, agent development kits. Okay, sorry to stumble over that.

But it's their SDK, if you will, for agents. Yeah. And largely, ADK has been, "Here's the builder tools to build agents," but it suddenly took a different form in the last week or two.

And that is, here's an ecosystem of things that are now available at the developer's fingertips: survivability, security- Mm. operational environment, all these ticks off... tick offs of some open source things that are...

it now supports and compatible with, but also commercial offerings. And it's gone from an, an SDK or an ADK, if you will, to kind of an execution environment that starts right in the developer's hands, that can do that. " I can make that decision- Yeah, yeah ...

right in the developer's fingertips. So again, these things are pushing up earlier into the process. Those decisions can be made much, much earlier.

But that to me is, again, it's not about writing the code, it's those things that are really accelerator and are the multiplying factor of what's happening. We're gonna see this more and more, I'm convinced. I'm with you, man.

You saw... You see it with like Databricks buying Neon because, you know, most of the provisioning of that database was by agent tools, not humans. Mm-hmm.

Mm-hmm. And you want that stack, you know, to, to be... I want my API to allow me to instantiate anything I need to support my solution without me having to, you know, drop out, go do an integration, come back and, and hope that it works.

I, I love what Google's doing there. It is. It's fascinating, and y- it, it hasn't gotten a ton of attention.

You know? You, you and I go, "Oh, wow, this is awesome," blah, blah, blah. Um, but I think in the developer community it definitely does, and, you know, I look for so what's the Microsoft response with their agent framework, SDK?

What's the response from, other technology providers, whether it's an Oracle or an IBM or, or, you know... Everybody has their own IDE now, so they can put those things in their environment as well. IBM comes out with this thing that helps you with modernization because that's now at your fingertips.

Not just to analyze the code base, but process engineering or whatever it might be. I'm making that item up. I'm not pre-announcing anything.

But that's, that I think is the stage we're at, is those accelerators are starting to take form and show us where the vendors are headed. Sign of maturity. I think we're due.

I think we, we deserve that for everything that's transpired since the end of 2022. I agree. It's a lot of fun.

Well, what, what do you, um... so what are you working on these days? You mentioned, the data, your survey data, your buyer information.

Is that coming out soon? Oh, yeah. Or have you already dropped that?

Thanks for asking, Mitch. I haven't. No.

So, so we, twice a year do a survey of, for, for my group, we do, you know, data intelligence analytics infrastructure practitioners, so everybody who's dealing with the data side of things. And, you know, one of them, you know, this is getting ready to go live probably in a week or two, I think. Okay.

We finalize the survey. We have all the data, and it makes sense. It's the one thing you can ask for, 'cause I don't...

For anyone who doesn't do this or hasn't done this, you have these, these panels, and you have percentages that you target a, a, on a step-by-step basis when you do a survey, and you watch the numbers like a hawk for each one- Mm-hmm. Mm-hmm ... " And then if, if you start deviating from that median, you know, as, as you go along, you start getting more and more nervous.

Like, either I had the absolute wrong ideas about this market, or, you know, the data's bad. You know, so, so thankfully I don't have to worry about either of those horns on the bull 'cause it all makes sense. So a-a-at any rate, long story short, we were asking about, observability as well, and amongst data professionals- Hmm.

Oh, cool ... they, they, you know, overwhelmingly, sixty-six percent said that they're increasing or accelerating dramatically their, their investment in this space specific to the challenges of anything as basic as just setting up a data pipeline to orchestrating a, a, a complex agentic workflow. Doesn't matter.

Across the board, everyone wants this. Wow. Everyone but four percent.

Four percent don't, and I want to understand them. I want to meet them. Like an ethnologist out in the field.

See, they're in a cabin in Montana off the grid. Is that the four percent we're trying to- Right. Exactly.

I want to understand them. I really do. Well, that's, that's fantastic, and thank you for asking that in your, in your survey, in your buyer decision-maker survey.

Very, very perceptive, very insightful for you to as-ask that question because that connects those dots, right? Observability is not a, a software lifecycle engineering unique thing. It's across the board- It's the whole...

Yeah ... for all- Everything ... security, data, every part of it.

And to that point, it's, it's one of the th-things that's exciting about this time is DevOps elevated a lot of other practices. You s-saw kind of data start to- Oh, yeah ... come more into the development cycle, security- Let's Opsify ...

a lot of things. Exactly. And AI is doing that, I think, even on a greater basis of kind of elevating all parts of what we do to say, "Okay, how's this part of the answer?

What is the answer we need to... " So kudos to you. Excited.

So we'll, we'll definitely talk about your, your data drop, your survey- Yes. We can, we can do... We c-we might even bring slides.

Charts. Charts. Charts, not slides.

Try something... Sorry, charts, not slides. Yeah.

Charts on slides- ... that would show some of this. So yeah, stay tuned, everybody.

Oh, goody. All the, the data wonks in the audience are like, "Ooh, great. " Yeah.

We're actually at, at Futurum, we're, we're, I think, very transparent in not just the methodologies we use, but with the actual data itself. We- Mm-hmm ... you know, feel very strongly that the-these are...

You know, the value isn't just in the data, it's in the interpretation of the data, and that's, you know, where our, our value comes in. Uh, so we- Mm-hmm ... we don't keep everything, everything locked down.

We, we open up a lot of what we do here. So I- We do, for sure ... I encourage you guys to, to check out, you know, the, the site and what Mitch and I are doing and, and our colleagues as well, 'cause they're all Opsifying and, FinOpsing and, and, observability-ing, verb, their, their practices, and there's a lot of interesting stuff happening right now.

There's a lot of treasure chests there, and we definitely get to see a good bit of them. There's still more. There's still more- Yep ...

for people who are subscribers. So the, the thing that I'm, I'm just kicking off, to, to preview a little bit, and you're a big part of helping me get this off the ground, is doing a signal, our AI-created- Mm ... report on agentic orchestration.

So that'll be out in a couple of months or so. And, you know, this is just coming around to, to do a refresh on the software development platforms, but I'm doing this one first. And first of all, massive kudos to you and the platform, Futurum Intelligence platform team of how far this process has advanced in six-ish months since I did...

I was the, I was the unofficial guinea pig, the second one to do a signal report. You know, so I'm getting code drops from a-from Brad. You, you were that- Every- ...

that second manager that was hired. Exactly. I'm like, "Don't change anything 'cause it's gonna change, so just kinda keep...

" Brad will say, "Oh, by the way, I already solved that. Here you go. " You know, all the undocumented things that aren't there.

You know, that's the fun part. Okay. Here's a new problem.

Being that early bird. Exactly. So, you know, I was u- more using AI to, like, "Tell me what this code does so I understand.

" Well, now it's all plug and play. It's gone, it's gotten to a much easier... There's the things with the companies and products and some of the axioms and the announcements and data and information that we feed into AI to help, help come out with this product, this report.

So I'm excited to do this. It's like redo- doing it a whole new way. It's...

Underlying, it's the same principles. Yeah. Um, s-same kind of effect that we're getting of what the output of it is.

But this will be on agentic orchestration, so a new space that we haven't... We talk a lot about... Keith Kirkpatrick talks about it in his work- Yeah ...

you and I do. Um, Nick, Patience, you know, and we all talk about this, so it's good to have kind of a starting place to look at agentic orchestration. So you can tell I'm excited about working on that too.

Plus, this coming week, I'll be introducing yet another, a second framework called the Agent Control Plane Framework- Ooh, yes. Oh, yeah ... which builds on observability native to say, "Okay, so once you can see this stuff, how do you r-make it accountable?

" Again, another framework reference kind of model. Um, but every vendor I talk to about either or both of these are like, "Yeah, let's get together. I'm gonna see what you're doing.

We need... " So I'm- It does ... equally excited about that too.

We need to think- And, uh- ... very carefully about, you know, and be very intentional about how we build software. Just because we've collapsed everything to 10 days now, doesn't mean- ...

that that's the right approach to take. We should think really carefully about what we build. Our new unit of measure in software.

Ten days. Ten days, yes. And, and by the way, huge kudos to you too.

Brad's been peer review and contributor to helping me with the Agent Control Plane framework and getting that in place. So just like, Fernando, Montenegro was with the observability framework. Oh, thanks, man.

Observability native. I appreciate it. This is the fun part of...

One of the biggest fun parts about doing this is the funnest part is collaborating on stuff. Nice. I'm still working through 10 days.

I'm not quite past that. That word using there, I, I, I appreciate the word usings. Yes.

I have lots of good words. All the best. All the best.

Absolutely. Well, I think we've more than overstayed our welcome. At least I know I have.

You're probably- No, this is a long one, wasn't it? Um- ... still in good graces.

We had a lot to say. Thank you, everybody, for sticking with us. If those of you who are here, thank you very much.

We appreciate you. We do. Yeah.

If you don't listen to all of it, we'll carve it up into three different episodes next time, so you'll, you'll listen to it. And, A, B, and C, episode 10 or whatever it is now. So we'll do that.

But not to belabor the point, thanks so much for, for sticking with us. Whether you stayed for 10 minutes or 20 or 50, we're glad that you're here with us, and, we appreciate the opportunity to share with you and talk with you about what we're doing. We always wanna hear from you.

com. com/brad-shimon, as well as mine at mitch-shulman, as well as all the other analysts at first name-last name at, Futurum Group. And be sure and subscribe, follow, share with your friends.

Plus, what you don't like, if you think we're, like, spending way too mi- much time talking about, you know, ERD diagrams from, you know, 1982, okay, that's fine. You can tell us that too. We'll, we'll, we won't listen to you, but, but we appreciate that feedback.

Yeah. Thank you very much. We're not gonna do that, but thank you so much for the feedback.

Snarkity snark. All right, take care, everybody. Thanks again for being part of Agents of Dev.

Thanks for our producer, Corey. Man, you're, you are number one tops in the biz. And our, audiovisual team for making this happen.

We'll see you all next time on Agents of Dev. Control, this is Agent Dev. I'm in position.

Copy that, Dev. Standby for go. Standing by.