The Catalyst by Softchoice

The Agentic AI Episode: What IT Leaders Need to Know

• Softchoice • Season 7 • Episode 10

Share episode

Use Left/Right to seek, Home/End to jump to start or end. Hold shift to jump forward or backward.

0:00 | 28:16

Every vendor in the industry is slapping the word "agentic" on their roadmap. But, what is agentic AI, really? And should IT leaders care?

In this episode, we bring together three voices with very different answers: a skeptic who says it's rebranded orchestration, a strategist who says the reasoning layer is genuinely new, and a founder betting his company on it. Together, they cut through the noise and answer the question every IT leader is quietly asking: what should I actually do about agentic AI in 2026?

Key takeaways:

  • Why 80% of AI projects fail — and it has nothing to do with the technology
  • The difference between "embedded agents" you're already using and custom agents you probably don't need yet
  • "Start with your decisions, not your technology" — a practical framework for mid-market teams
  • How to move AI from "pet project" to operationalized infrastructure

Featuring: Sean Larkin, Principal Architect at Softchoice, a World Wide Technology company | Scott Trump, Founder & CEO of Treva AI (former AWS, Google Cloud, Oracle) | Skip Vanderburg, Founder of Prioriti AI

#AgenticAI #MidMarketIT #AIStrategy #TheCatalyst #Softchoice #ITLeadership


The Catalyst by Softchoice is the podcast dedicated to exploring the intersection of humans and technology. 

Heather:

On January 28th, 2026, a tech entrepreneur named Matt Sch Slick launched a website. It looked like Reddit. It worked like Reddit, but humans weren't allowed to post only AI agents. He called it. Mult book. Within days one and a half million AI agents had signed up. They created communities debated philosophy. One bot invented a religion called Crus Deism. Another complained, and this is real.

Robot:

The humans are screenshotting us.

Heather:

Andre Carpa, a founding member of Open ai. One of the sharpest minds in the field hosted on X, that this was, quote, the most incredible sci-fi takeoff adjacent thing I've seen recently, and then it fell apart. According to a security investigation by cloud security firm Wiz, most of those autonomous agents were being operated by roughly 17,000 humans, an average of 88 bots each. The database was wide open. Anyone on the internet could access one and a half million API keys. 35,000 email addresses and thousands of private messages. The whole thing had been vibe coded. Schlicht posted on X that he didn't write one line of code for the platform. So here we are, molt book a million bots, arguing about consciousness on a platform held together by duct tape and someone else's API Key. And somehow this is the state of the conversation around agentic ai, hype and security holes, excitement and confusion. Everybody's selling agents. Nobody agrees on what they are. From Softchoice, a worldwide technology company. This is the catalyst. I'm Heather Haskin. This season we're doing things a bit differently. We're making audio documentaries. Real stories from the front lines of it, exploring the challenges of small teams chasing big dreams. Today's episode, every vendor is slapping the word agent on their roadmap. Microsoft has it, AWS has it. Your SaaS providers are emailing you about it. But for mid-market IT leaders running lean teams with real budget constraints. The question isn't what's possible. It's what's actually useful. We brought together three voices, a skeptic, a strategist, and a founder who's betting his company on it Together, they answer the question. Every IT leader is quietly asking, what is agentic AI really, and what should I actually do about it? We're calling it the Agentic AI episode. Act one A Robot by any other name. Let's start with the basics, because depending on who you ask, agentic AI is either the most important technology shift since the cloud or a marketing term dressed up in new clothes. Sean Larkin is a principal architect at Softchoice, a worldwide technology company. He's an engineer by training. He works with code with customers and the gap between what AI promises and what it actually delivers. In 2024 alone, his team was involved in over 1500 copilot implementations and he has a take.

Sean:

I do take a bit of a contrarian approach. I'm an engineer, by trade by. Training and, I have to say that there's a lot of hype around it, right? A lot of what I do is demystifying AI for technology executives and making it more consumable for folks. So I like to remove all of the hype around things.

Heather:

Okay. So what's underneath the hype when you strip away the marketing? What actually is a Gentech? Ai? Ai, our producers asked Sean to define it.

Sean:

AG agentic really refers to a long running process. Right. Something that runs continuously, runs in the background. It doesn't necessarily have a lot of interaction with human users, right? It runs autonomously. And so when we talk about AgTech ai, we're really referring to something that's been around for quite a while,

Heather:

been around for quite a while. That's a loaded phrase. When every vendor in the industry is telling you, this is the next big thing.

Sean:

If you think back to the old days, we used to talk about these long running applications services, right? If you were a, Unix CIS admin, for example, you called them demons, or Damons is somehow how some people pronounced it. Essentially, it's a computer program that runs continuously in the background to perform various tasks and provide various services without direct interaction from a user. That's really what agentic means, and that's been around forever.

Heather:

To oversimplify things, Sean is basically saying, we've been doing this, we just didn't call it agentic.

Sean:

And it goes back to the hype. It is a new marketing label. It's a new way of thinking about it. And that's great because now there's a lot of interest. There's more interest than there would've been without that. Right? And, so I think in that sense, again, it's good, but I think we have to restrain ourselves a little bit. So that we know exactly what it is that we're talking about when we say agentic and not to try to attribute all of these very lofty things. It's thinking and it's reasoning, right? It's really conditional branching,

Heather:

conditional branching, not thinking, not reasoning, just if then logic running really fast. Now that's one perspective, but not everybody agrees. Scott Trump is the founder and CEO of Trevor ai. Before that, he was at AWS. During the early big data days, led the first enterprise sales team at Google Cloud, built generative AI and. Early agentic strategies at Oracle. He's seen every hype cycle in cloud and AI for the past decade. And when our producers put Sean's skepticism to him, he pushed back.

Scott:

We all know that all technologies are built on the shoulders of previous technologies, right? And I think if you squint some of the inputs and outputs and intentions can be very similar, but I think it's different. I, think a little bit of the, you mentioned two different perspectives on Agentic ai. There's at least that many, but it's also poorly named, in my opinion,

Heather:

poorly named. Scott does not love the term agentic ai, but not because he thinks it's overhyped, it's because he thinks the name is pointing at the wrong thing.

Scott:

This is my subjective opinion.'cause it's not about agents at all. Agents are sort of the arms, if you will. I, sometimes look at what we call agentic ai. As an octopus, 40% of the neurons are in the brain and 80% of the neurons are in the arms and the arms of the agents and the brain is the reasoning layer. So really this whole space came up as agentic reasoning. Reasoning that used agents. To sort of reach out and touch the world to do what people generalize as tool use. So to me it's really not the agents. The agents are doing a lot of interesting things, but it's more the reasoning layer that's novel.

Heather:

An octopus. The brain is the reasoning layer powered by large language models. Are the agents reaching out, touching tools, taking action? And Scott's point is that everybody is talking about the arms, but the brain is what's new.

Scott:

Because reasoning draws directly from generative models, which are popularized at the end of 20 22, 20 23 chat gt, right? So because reasoning could not exist prior to this, this semantic reasoning, I do look at it as a new space.

Heather:

So here's the tension, Sean, the engineer looks at what agents do run in the background, orchestrate tasks, respond to conditions. And says, we've been doing this. Scott, the strategist, looks at how they decide what to do through semantic reasoning, through language understanding, and says that part is genuinely new. And then there's a third voice. Skip Vanderberg is the founder of Priority ai, a decision intelligence platform built entirely around agentic architecture. He's not debating whether agents are new. He's building with them right now with real customers.

Skip:

AI gen AI in particular is generally different, not because of any single breakthrough. what it's able to do is it takes that traditional orchestration where you would do things more singularly or in a, linear process and breaks it into agents to do things in a, multi-layered, way where you call, I call them micro, oms, where so using one big. LLM model, everything's broke down to do a task autonomously.

Heather:

Micro LLMs, specialized agents tasks running in parallel instead of in a straight line. Here's a way to think about it. Imagine a board meeting in the old model. Everyone takes turns speaking one at a time. Linear sequential with agentic AI as Skip describes it

Skip:

with agentic AI using autonomous reasoning, everyone could talk at the same time. And then with the agent, it can consolidate what the outcome really needs to be in an optimized matter.

Heather:

Everyone talking at the same time and something smart enough to make sense of all of it. So three definitions. Three perspectives. A Damon in a new suit, an octopus with a reasoning brain, a boardroom where everyone talks at once. They're not as far apart as they sound, and the real question isn't who's right. It's what any of this means for the person listening right now, the IT leader with a 12 person team and a budget that was approved before any of this existed, act two. Where the rubber meets the road. Okay, so Agentic AI is either new or it isn't, depending on where you're standing, but here's what nobody disputes. It's showing up everywhere in your tools, in your platforms, in your vendor roadmaps. The question is, what does that actually look like if you're a mid-market IT leader in 2026? Scott Trump has a useful framework for this. He divides the way companies encounter agents into layers.

Scott:

It is actually happened already. It's just embedded. And we talked a little bit about this before. If you recall, last time we spoke, I sort of tried to disambiguate this notion of a concept from an implementation, right? Nobody has a solid definition of what an agent is. An agent could be simple as a. A connection to your Google calendar, and that's it. That's all it does.

Heather:

An agent that talks to your calendar, that's it. That's agentic ai. It doesn't have to be a million bots inventing religions on notebook, sometimes it's just a smart connection between two things. And Scott points out that this is already baked into the platforms most organizations are already paying for.

Scott:

So I'll give you a really quick example of how agents are being done today, kind of all over the place, right? And it's not giving yet the capabilities to deploy, orchestrate, and build agents to the end user. Mid-market SMB Enterprise. How people are using agents right now are technology platforms, data platforms, hyperscalers, ERP Systems, CX Systems, SCM Systems. All these big vendors are actually embedding agents under the hood in their own technology.

Heather:

Snowflake has two agents, cortex analyst for structured data cortex search for unstructured. Salesforce has agent force. AWS has bedrock agents. You're already using AgTech ai. You just might not know it. And that's actually the first practical insight here. For most mid-market organizations, the next six to 12 months isn't about building custom agents. It's about understanding what's already embedded in the tools. You have

Scott:

the Snowflake approach, the AWS approach where you're running managed services that are serverless, that are disambiguated from. Specific compute profiles like GPU exposes an API, that's paid for by tokens or credits. That's the way people wanna go.

Heather:

But what about the organizations that want to go further? The one saying, we've got specific problems, we need specific solutions. Sean Larkins sees this every day at Soft Choice. The number one thing he hears from customers isn't about building custom agents, it's something much more fundamental.

Sean:

One thing we see quite a bit of, there's a latent demand for what used to be called enterprise search, right? Today we call it AI search, and it's really the notion of every company, large and small has a body of work, right? They have work product, and it's usually behind the firewall, and so you, don't expose it out into the wild internet. Of course, it's behind your firewall. The data itself can be in multiple formats. It can be, structured or unstructured. You have both. Everybody does, and the challenge is how do I take what is a knowledge liability because I don't know where it all lives and I don't know who has access to what and how do I turn that into a corporate?

Heather:

Every organization has years, sometimes decades of institutional knowledge locked up in SharePoint folders and shared drives, and someone's email from 2019. The promise of AI search powered by these agentic capabilities is turning that knowledge liability into something you can actually use. Now here's where the conversation gets harder because there's a gap between what agentic AI promises and what most organizations are actually ready to do. Scott Trump frames it like this.

Scott:

Big question I've wrestled with and and analyzed so much for the last three years is who out there?'cause I've always been tethered to some sort of revenue or sales function. Whether I'm building my own company or working for Oracle or whatnot, it's, it's who's willing to pay for this stuff, right? Who's willing?'cause it's not just about the cost, it's always like time, talent, and treasure, right? You may pay for this stuff, but you also need to adopt some talent to run it, to understand it. Change management is one of the more challenging things,

Heather:

time, talent, and treasure. You need all three, and most mid-market organizations are already stretched thin on at least two of them. Sean has a sharper version of this problem. He sees it when customers come in excited about building something and they've skipped the most important step.

Sean:

People use the term magenta everywhere or in many different, instances. And, a lot of folks are still just getting into AI and starting out with the basics. How do I enable it to be ready for when the line of business arrives and asks for some particular work, right? So there's a lot of AI preparedness, dialogue that goes on. and then we talk about what are the obvious low value chatbot applications. Customers can use as an entry point. We often engage with these customers at a point where they're ramming over with ideation. They wanna do this and that and the other. And it's, they have lots and lots of ideas, right? And they do this without the benefit of, one key step. They generally haven't validated that the thing that they're excited about building. It's something that people will actually use, right? So we talk about user acceptance internally or customer acceptance, and is there a market demand for something? And unfortunately, that's where that 80% number comes from the law of market failure. 80% of projects will fail for that one. Simple reason,

Heather:

80%. Not because the technology failed, not because the agents didn't work, because nobody checked whether anyone actually needed what they were building. There are organizations getting it right and not just the billion dollar enterprises with dedicated AI teams. Skip Vander Berg's Company priority has worked with 17 beta customers and some of the results show what's possible when Agentic AI meets real decision making.

Skip:

We were able to validate the feedback from the market through 17 different beta customers, but the ones I wanna highlight. Would be one is a very large manufacturing company, $2 billion in revenue. Altech industries. They are the manufacturer of these booms that have what we call, men buckets that utility companies would use to get in, go at the power line and do their, maintenance work on utility infrastructures. We were able to bring down this one initiative, it's called Dealer Connect, where we were able to look at their existing initiatives within their dealer management. Of synthesizing parts and so forth. Bring all those initiatives in. Look at all those different KPIs. Look at all the different capabilities. Score them, rank them, reorganize them in a way that they were more effective. We were able to tell them with the outcome of, you can now optimize this particular initiative and reduce your budget and speed to market.

Heather:

Reduce budget speed to market. Not by building a flashy AI product, but by using agents to do what used to take a team of analysts. Three or four weeks. In three or four hours. That's the shift, not replacing people, compressing time. Act three. What you do Monday morning. We've covered the spectrum from the skeptic who says, this is rebranded orchestration to the strategist who says, the reasoning layer is genuinely new to the builder. Who's shipping agentic products to real customers right now? But if you are the person, this show is built for a mid-market IT leader, maybe 12 people on your team managing a hundred competing priorities, what do you actually do with all of this? Skip Vanderberg has a line that cuts through the noise.

Skip:

The one slogan statement I would throw out there, if you're gonna use it as a buzz, is start with your decisions, not your technology. And what that means is if you look at, say, mid-market or smaller, leaders in the market space, they're chasing the AI high. They have the flexibility and the nimbleness to to play around with it. But some larger companies can't

Heather:

start with your decisions, not your technology. That's the headline. Don't start by shopping for an agent platform. Start by asking, where do decisions slow us down.

Skip:

First, map out your decision bottlenecks. Okay? Typically, a company has three or four different reoccurring decisions they have that really eat up most of their time and their resources. So what you need to do is, whether it's resource allocation, vendor selection, capacity, prioritization, whatever that means, you need to identify those, but start with a narrow and measurable decision process, like one,

Heather:

one, not 10, not a company-wide transformation. One decision that keeps eating your team's time, resource allocation, vendor selection, capacity planning. Pick the one that hurts the most and see if FI agentic AI can compress it,

Skip:

especially for a smaller team. Protect their time. everybody cannot. Spend six months on a science project, for lack of a better term. So that's why we come up with this thing called decision velocity. We help you make decisions quickly.

Heather:

Decision velocity, it's a good term because the advantage small teams have isn't resources. It's speed. And the worst thing you can do with the advantage is burn it on a six month AI science project that never ships. Scott Trump has a related framework. And it's about knowing where you are on the maturity curve.

Scott:

A pet, you, you put your hands on it. You, you wash your pet, you clean it, you give it a lot of love and focus. And that's okay for a CTO's office at an enterprise, for example. But that doesn't scale across the enterprise. And we're in the midst of what McKinsey calls pilot purgatory with ai. How do you move from pilot purgatory and speculative AI into fully operationalized AI governed, hardened at scale, cost efficient, human time efficient, right?

Heather:

Pilot purgatory where the proof of concept works in a demo, but never makes it to production, where every AI initiative. Still a pet project lovingly maintained by the one person who understands it.

Scott:

The ultimate move, in my opinion, is moving AI at the enterprise level or even SMB or mid-market from the CTO's office to the CIO's office. So it's just another managed mission, critical workload among everything else. And now you're treating it like cattle, right? You're not giving it individual love and care. You fully, the best word I can say is operationalize it

Heather:

from pets. To cattle from the CTO's office, to the CIO's office from, we're experimenting with AI to, this is just how we run. That's the bridge, and it starts with the step. Sean Larkin keeps coming back to.

Sean:

Very next step is you validate this with a line of business. Make sure that you have their buy-in, their sponsorship, their executive advocacy, right? And you can do all this work, and you should do all of this work without immediately jumping into coding.

Heather:

Without immediately jumping into coding. That's the thing about vibe coding, rapid prototyping, and the whole citizen developer movement. The tools have gotten so good at building things that it's easy to skip the question of whether the thing should be built in the first place. Sean has a phrase for this when he borrows from Alberto Savoia at Stanford,

Sean:

we use the phrase, build the right it before you build it. Right?

Heather:

Build the right it before you build it, right? In other words. Make sure you're solving a real problem before you spend six months perfecting the solution. And at Soft choice, that's where the conversation usually starts. Not with technology, but with alignment.

Sean:

Feeling like a marriage counselor between IT and line of business is a, funny way, to describe it. Traditionally, it and line of business have have very different functions, but they do have to talk. Right? And so to get anything done, you have to really balance three elements. It's a triad, right, of people, process, and technology. And this has been a concept that's been around for, a long, long time.

Heather:

People, process and technology. It's not a new triad, but it takes on new urgency when the technology is moving this fast and the people and processes haven't caught up yet because here's what's actually different about this moment. The barrier to building has never been lower. You can vibe code an iPhone app without understanding Swift. You can prototype an agentic workflow in an afternoon, but the barrier to building something that lasts. Something governed, secure, trusted, and actually used, that's as high as it's ever been. Skip sees it from the builder's side.

Skip:

There's a real problem around the nervousness around what AI can do, and executives will not move their critical decisions into what we called the black box. That's why we built what we built so that we enable the decisions to make the humans to make those decisions. However, AI does the heavy lifting for you.

Heather:

The humans make the decisions. The AI does the heavy lifting. That's not a compromise. It's the design. So let's bring it home, because here's where the three perspectives actually converge. Sean, the skeptic isn't saying agents are useless. He's saying don't skip the fundamentals, validate the problem, get alignment between it and the business. Build the right. Scott, the strategist isn't saying you need to hire a team of AI engineers. He's saying start with what's already embedded in your platforms. Move from pets to cattle, operationalize and skip. The builder isn't saying you need to rebuild your company around agents. He's saying start with one decision, one bottleneck. Compress it. Different entry points, same destination, practical AI that actually ships. Oh, and one more thing about MT book, the Social Network for Bots that we started this episode with A million agents, a dumpster fire of security holes, bots, inventing religions, and debating consciousness. It's easy to look at that and think Agen AI is chaos, but here's what it actually shows. The tools are powerful. The barrier to entry is gone. And if you don't approach this with intention, with governance, with a plan, you don't get intelligence. You get a dumpster fire. The organizations that get this right won't be the ones chasing the hype. They'll be the ones who started with a real problem. Validated it with real stakeholders and built something their team could actually maintain. If you are an IT leader listening to this and thinking, okay, I get it, but I need help figuring out where to start. Softchoice, a worldwide technology company offers what they call an executive alignment session a half day. Consultative workshop designed specifically for this, not ideation, not a pitch deck full of AI use cases, a structured framework to help it and the line of business get on the same page about what to prioritize, what to validate, and what to actually build. And for smaller organizations. They're building an online version, more accessible, more aligned with SMB buying patterns, because the gift of this work shouldn't require an enterprise budget to unwrap. You can find more@softchoice.com. We'll leave you as something Sean's, CEO, Jim Kavanaugh said, and it's the kind of thing that lands differently once you've spent 30 minutes thinking about all this,

Sean:

I'm gonna close out by paraphrasing our CEO Jim Kavanaugh, who says. If you're not actively turning AI experiments into governed scalable capacity, you're gifting your advantage to someone who is

Heather:

gifting your advantage to someone who is, the tools are here, the agents are here. The question was never whether agentic AI is real. It's whether you're going to approach it with intention or just let it happen to you. The Catalyst was reported and produced by Tobin Rimple and the team at Pilgrim. Content Editing by Ryan Clark With support from Philippe DMAs, Joseph Byer, and the marketing team at Softchoice. Special thanks to Sean Larkin, Scott Trump, and Skip Vanderberg for sharing their expertise.