Summary
This was the 188th episode of “Coffee with Digital Trailblazers,” focusing on technology budget planning for 2027 and AI investment strategies. Isaac led a scenario-based discussion where panelists were asked to provide specific recommendations rather than generic responses about AI funding and budget protection. The panel included John, Joanne, Derek, Liz, and Joe, who discussed various approaches including AI agent orchestration platforms, security governance, identity management, and business process re-engineering. Key frameworks presented included Derek’s four executive questions (value, risk, resilience, and accountability) and Joanne’s emphasis on creating differentiable value through capturing institutional knowledge. The discussion also covered how to make priority decisions between competing AI initiatives, with participants agreeing that strategic clarity and measurable outcomes should drive AI investments rather than simply cutting costs.
Speakers
- Host – Isaac Sacolick
- Digital Trailblazers – Derrick Butts, Martin Davis, Joanne Friedman, John Patrick Luethe, Liz Martinez, Heather May, Joseph Puglisi
Discussion
- What’s funded: It’s budget season. The CEO/CIO/CAIO have committed to 200+ AI agents in production by the end of 2027 — agents shipping inside our ERP, HRMS, and CRM, plus proprietary agents. Fifteen months. You get to pitch one incremental funding area. What is it and why?
- What’s protected and where to cut: The CEO wants to plan for a recession in 2027. What’s one area of the budget that you’ll fight to protect, and what’s another area where you’d look to make cuts?
- How funding decisions are made: Now you are the CEO/CIO/CAIO, and you have to decide what gets funded, what’s protected, and what you cut. How do you make those decisions?
Research
Dashboard
Transcript
[00:00:00] Isaac Sacolick: Greetings everyone. Welcome to this week’s September 18th episode of the Coffee with Digital Trailblazers, our 188th episode.
We are now almost officially at four years. I don’t remember the exact date, so I want to thank Joanne, Joe, Martin, Derek, Heather, John and Liz for being our standing guests for our discussions. Every week covering a different area focus around digital and AI transformation and our topic this week is one that you’ll all be going through over the next one to two months.
We’re going to be going through some scenarios around the technology budget in 2027, where to focus your AI bets and how to think about protecting it and security so they don’t get undercut.
To have this conversation I’m going to lay out some scenarios for my for my speakers almost. We’ll do a little bit of story and scenario planning here around what it’s going to feel like as we’re under pressure to deliver value from AI and also under pressure to find, cost and simplify or organizations.
Can I get a mic check? Can you guys hear me?
[00:01:28] John Patrick Luethe: Yeah, we can hear you.
[00:01:30] Isaac Sacolick: Okay, great. I see like a spinning wheel on my zoom screen here and I was like okay, I’m glad you’re here.
I have this week’s research slide up About Tech Budget 2027 what AI bets to fund and security and it to protect.
This is coming from an article that did really well. It came out two weeks ago on my blog. I wrote about the 10 signs of the AI bubble bursting. Seven way CIOs should prepare for it.
I’m not saying when the bubble will burst, but I do think it will. I’m not saying that it will be a huge disastrous trough or necessarily saying we’re going to get into a long multi year recession.
But I am like many others, predicted that there’s going to be some form of correction and gave 10 different reasons why that’s going to happen. You see four of them here where I brought some data into the conversation.
Enterprises respond to AI’s rising costs and ROI gaps. Some data from infosys, KPMG and McKinsey 72% have scaled fewer than 1/4 of their AI pilots. Only 7% of leaders report having established ROI from AI.
Record AI spend but can’t move EBITDA for 94% of enterprises. So when you look up rising spend and productivity as a big driver but not a lot of value, somebody on C Suite is going to say show me the money. Second reason VPC bets are concentrated in fewer companies in H126 100 million plus revenue took 87% of US venture dollars, mostly in AI.
So you’re not seeing a lot of startup betting. You’re seeing a lot of doubling and tripling down on existing Same trend happening in the eu.
AI Security insists Impact More than Reputations this is hitting the AI companies themselves the last two weeks more than others.
Lots of announcements last weekend about slowing down the pace of AI.
As the New York Times reported this week, at least six other incidents of questionable responses from OpenAI while it was in development, including releasing documents that shouldn’t have been released. I think all those are going to raise questions inside enterprises about how fast they’re going to and then we’ve covered this before here. There’s growing resentment as AI impacts the workforce. There’s resentment against the data centers that are being constructed, the impact around energy. Some data points here around that. I’ve got all 10 signs on the blog. You see the URL here that I’ll paste for all of you to be able to go access to it. And this is the forefront for our conversation today about how should we think about the 2027 budget now, you know, all of us, we do not have boards or leadership teams or businesses that we can surface as context around how we think about these answers. But clearly a $100 million business in retail is going to have a very different answer than a global company with hundreds of agents deployed that’s regulated and thinking about what their AI spend is going to look like. So I’ve already prepped my speakers and said, listen, our answers can’t be. It depends. Our answers can’t be look for where the business value is. We’re going to try to steer away from generic answers and giving you some context around how to think about funding.
I’m going to suggest this scenario, everybody.
The CEO, CIO, C A I O, somebody who really owns the AI budget, has committed to release 200 plus agents in production by the end of 2027. Doesn’t matter what the agents are or what the value is. Chances are a number of them will come from the erp, the HR or the CRM. Core applications are now providing agents directly in their platforms. So. So the easiest things companies are doing is saying, hey, what is my platform offering me and how should I take advantage of that? You’ve got 15 months to think about taking these ideas that vendors largely deployed this year and saying how are we going to not only deploy, get value from them, but also scale, govern and manage and monitor them. About 15 months from October to end of 2027.
I’m going to start with John. John is only here for our first half hour. That will go to Liz, Joanne and Derek. John, you get to pitch one incremental funding area, expand my story so you have as much context as you want.
Where are you investing and why? Welcome to the floor, John.
[00:06:43] John Patrick Luethe: Yeah, Isaac, thank you for having me on. It’s been a lot of fun conversations over the years. If I can pitch one area of funding, it’s really, it’s really going to be in, in the security space. And, and right now it’s so easy to have agents take the permissions of users and start doing activities. And so I would put funding into security so that people in the company aren’t using unauthorized AI agents. And so the other thing is, is that they’re, they’re not using unauthorized AI tools. And I would put security that, that. So it’s really making sure that people are only putting data into authorized AI tools. And I would put training in and so that all my funding that’s incremental for the year. The one ask is going to make sure that we’re doing things and in a safe way with AI because it’s so easy to use unauthorized AI tools, paste the data in. Data loss protection would stop that. And it’s so easy to also have people using unauthorized tools that have agent capabilities and they can log in with their username and they can log in with their permissions and start making changes to things and running autonomously. And it could be with supervision or without supervision. But the only way to control that is to start putting limits on what tools can be used and, and how they’re used.
[00:08:10] Isaac Sacolick: Interesting.
So we’ve already got agents available to us in our existing platforms. Let’s make sure that we’re doing it in a governed way, protecting our data and that we’re training our employees so they know how and how not to use these AI agents.
Paraphrasing that, right, John? Yeah, I’m getting a thumbs up.
[00:08:33] John Patrick Luethe: Absolutely, yeah.
[00:08:35] Isaac Sacolick: Liz, I need a different scenario from you, a different proposition. No repeats. I want to give everybody different ways to think about this problem and not that you would have gone into security angle anyway. Go ahead, Liz.
[00:08:48] Liz Martinez: All right, so first of all, part of the problem that I have with all of our AI conversations is that they talk about AI as something that has to be focused on. I’m like, what do we do with this? And I it’s completely a tool. It is a horizontal that goes across all your business problems and we have lost focus on focusing on the business problems. So if we, if we’ve done the hard work of understanding what our strategy is, we already should know how we want to divvy up those 200 agents because they should be divvied up by the percentages according to your strategy, your strategic initiatives. Number one, I 100% love what John said.
I would say take 20% off the top and put it directly to security and governance, just 20% off the top. So like from your 200 you’ve already lost 20 and then take the rest of them and divide it up by your strategy. Now, I was most recently working at a credit union where we’re actually looking at the business problem of frontline staff who are trained in different ways, they have, they are providing members with different experiences. This is a real business problem that they are struggling with and something that AI could actually be used to address. We instead of retraining everybody in every single discipline, how about we create an AI agent that actually pulls the right information based on whatever prompt that they give it so that they can properly serve their constituency and have all the right information at their fingertips. Now that’s solving a business problem, focusing on the business problem, not focusing on, gee, what do we want to do with AI? It’s like having a million dollar Swiss army knife and then walking around your house going, gee, what do I want
[00:10:42] Isaac Sacolick: to do with this?
[00:10:43] Liz Martinez: It’s just silly.
[00:10:45] Isaac Sacolick: Liz, I love that you represent the PMO and Value management office and already giving us a balanced answer to taking 20% towards security and governance and then giving us a real use case. Right. I’m going to target frontline staff. I’m going to look for agents that are going to allow them to focus their questions to make them more effective in their work. And then I’m probably going to do some kind of balanced approach based on strategy and who am I going to help, which departments with their use cases after that?
Love a very pragmatic approach, Joanne.
We’re talking scenarios. You can build up the storyline in any direction that you want. But I need to hear what is the one area you’re funding and why under that scenario?
[00:11:31] Joanne Friedman: Thank you and good morning to everybody. I hope you’re all well.
I would put the incremental funding into agenta control and execution layer.
And I would do that because first of all, 200 agents in production is not an enterprise capability. It’s 200 new actors operating across various back office systems, proprietary systems that may have partial context, they might have overlapping authority, competing objectives, you don’t know what you don’t know. So I kind of challenge that premise. But I want to continue with this line of thought.
Until you have a common layer, every team will build its own permissions, its own integrations, its own approvals, monitoring and audit trails.
In short, you’re going to be paying for plumbing 200 times and still have no reliable way to operate those agents as a system.
So I would invest in a shared layer that would give every agent an identity, a definition of what it’s authorized to decide, what it can and cannot do, coordinate its actions with other agents and with people, so that you have some centralization, not in terms of systems, the agents running, but you can create the foundations for, for example, a swarm or a team, or whatever vernacular you want to use around agency.
I would also have it maintain a traceable record of evidence of decision of authority, action and outcome.
And it should also allow permissions to be revoked. It should have bounded autonomy. All of those things go into this sort of control plane or execution layer.
And you know, in terms of challenging the initial hypothesis, 200 agents is an inventory count, it’s not a business result. And to Liz’s point, we need to deliver results.
So it’s challenging to find what those problems actually are in priority. But if you can’t deliver a solution and a financial outcome, then how much decision to action latency is being removed and falls by the wayside. And that’s really the goal. So my pitch would probably be a little bit simpler and I would say, okay, you funded the agents, now fund the enterprise’s ability to operate them as a cohesive system.
Otherwise you’re just, sorry, digital labor.
[00:14:06] Isaac Sacolick: Joanne, do you call that middleware an AI orchestration, AI agent orchestration platform, or do you label it differently? That’s what I call it.
[00:14:16] Joanne Friedman: Well, I tend to go more on the side of choreography than orchestration. I just call it a control plane, embedded execut agentic control and execution.
Because there’s two versions. There’s what’s built in and then there’s runtime.
And if you don’t build what you need in runtime at the outset, it’ll fail.
[00:14:41] Isaac Sacolick: So Joanne, I actually agree with you.
I think in, in my scenario, just to continue the thought, if you’re in that 20% of our agents are making it into production and we’re somewhere in that, you know, let’s call it 0 to maybe 28 range, start projecting what your world looks like when you get from 20 to 250.
Now you’re starting to think about control planes and you’re thinking about observability and you’re thinking about uniform security and a whole bunch of attributes that you want to manage in a uniform way.
And I call those AI agent orchestration platforms. There are at least 65 vendors, we’re doing the research on this at Star CIO, at least 65 vendors who are coming out with platforms under that category, whether they’re true orchestration or choreography. I think that’s a good debate on semantics, but you’re going to see a lot of that play out. So if you’re, if that’s your space and that’s your goal, you might look at those platforms. And the reason I went with that conversation, Joanne, and I’ll let you comment before we go to Jerrick. I mean, this is, you know, going to have 200 plus AI agents in production at the end of 2027.
Unfortunately, that’s how a lot of board and CEO edicts come across.
Right. They don’t really get into exactly where to place the bets, unfortunately. And to Liz’s credit, they don’t speak to what the real business objective is. They let the teams fight and sort that out themselves. And that’s going to be the nature of our last question. How do we make progress, priority decisions? I’ll let you comment on that, Joanne, before we go to Derek.
[00:16:33] Joanne Friedman: Okay, you can do it at the beginning or you can see the result and then redo it at the end after it’s all taken place. But if you don’t have a level of homogeneity, you’re going to have agents that are running, not only running amok, but running into each other, where at some point they’re going to go, I don’t know. And I don’t know is a perfectly good answer where a human in the loop needs to come into play. But you know, think about the back office systems that you’re talking about. If your ERP has a set of agents that that’s built in from a vendor, you have no control over how those agents run. And that should be one of the priorities, by the way, in how you choose the platform. You need to be able to make that as adaptive as possible.
Because if you don’t, you’re going to not only have semantic collisions, which is really simple to describe. The word yield means different things in different situations.
Production yield is the output yield is also don’t cross that intersection without slowing down first.
It doesn’t understand the difference. So you have to put parameters around it so you can have agent collisions between the back office systems. You’ll have plumbing that has to be amended or appended in different ways. That that’s part of the reason that I said the funding should go to control like this control plane. And to your point, there are a lot of vendors in the space but it’s about really comes down to agency.
How much of your judgment do you want to outsource to AI in any situation?
And that’s not about people necessarily or process necessarily. It’s about bringing people and process business solutioning if you will or systematic ways of controlling the agents so you don’t end up in this quagmire because every vendor’s interpretation will be different, every model will constantly be changing and updated. That is an AIOps problem that you don’t want to have consistently running to Derek and John Security absolutely. But it all starts with agency. How much do you want to allow them to do to run your business or or rather how much of your business are you going to re architect to allow the agents to take over?
[00:19:10] Isaac Sacolick: I’m going to take your comment here Joanne, which is I think a question of strategy and planning. And a comment on the comment streaming on LinkedIn Raymond Lacqua is also somewhat challenging this assumption. If you don’t know what the outcomes you’re driving, that’s a problem. Maybe AI isn’t the solution.
And so you know, if you’re in that organization that’s chasing agents without outcomes or intent, that’s probably the first thing I would bet place my bet on is bringing the organization together and doing that planning and strategic thinking and going back to departments and saying we need vision statements around this from here before we start even experimenting.
And that takes time and it takes investment to do that. And before you start throwing tools out there, maybe we should be able to have these conversations say like what is our target?
I’ll let you finish the thought Joanne, then we’ll go to Derek.
[00:20:10] Joanne Friedman: Yeah, what is our what is the outcome in financial terms?
What is the outcome in business operations terms and so forth is absolutely critical. But the, the notion of multiple vendors putting agents out there is a train wreck waiting to happen. Because until you understand what in their proprietary technology, and this is especially true of SaaS apps, what will the impact of those agents be? How do you design anything that’s going to collaborate with it? Unless you want to start going into the weeds of each vendor software and saying well you know this is a better way to do that part but then it fails at this part. And so before I Start putting McP servers and APIs out there and you know, looking at new security mechanisms to govern the whole thing.
I really have to understand how my, as I call them swarms are really going to work together.
So it’s a vision statement with a fixed outcome attached to it.
And I think you’re, you’ll find across different functional groups of the organization that in many cases the vision statements can be wrapped into one larger statement. I’m trying to gain efficiency, I’m trying to gain more revenue, I’m trying to gain better performance from, from the functional group as it relates to everybody else. This is where this level of blue ocean thinking comes into play and why all this has to be done from a sort of systems level before we start throwing tools.
[00:22:02] Isaac Sacolick: Thank you, Joanne. Let’s move over.
Derek has been waiting patiently. Derek, before you give me your scenario and your answer, take 30 seconds and tell people what you’re doing now and the cause you’re supporting.
[00:22:15] Derrick A. Butts: Oh, well, thank you. Thank you for having me. Yes, I’m actually work with a nonprofit called Prostate Health Matters and we’re pushing pedals for PSAs and the reason is that September is Prostate Cancer Awareness Month and we’re really trying to help men focus on their prostate health before cancer diagnosis. I am actually going to ride 250 miles as part of this campaign for men’s health awareness.
About 147 miles in. I’m going till the October 10th. And it’s really kind of help men understand they need to know their PSA. They need to track this because it’s the number two silent killer of all men in the US and it’s also 75% of men diagnosed will be asymptomatic. So the only way you’re going to know if something’s taking place is to understand what your PSA number is and be able to track it. Thank you.
[00:23:01] Isaac Sacolick: Thank you. Derek.
[00:23:02] Derrick A. Butts: Yeah. Do you go ahead your bet.
[00:23:06] Isaac Sacolick: Where you, where are you making your investment and what’s the scenario?
[00:23:10] Derrick A. Butts: So my investment is I’m going to look at enterprise, AI agent resilience and governance capabilities. That’s what I’m looking at. I’m not looking at adding another 201 agent. I’m really looking at how can I make this cyber resilient argument in the organization.
When you look at this from a human perspective, you know, AI agents are moving to take the place and become digital workers and take actions on the people are doing. But the question is now is what are the agents acting for? Some of the comments that Joanne mentioned and John and Liz mentioned. You know, I want to know who can see the data, who’s making the decisions for the data, how is the system changing the data? I’m looking at how These agents and APIs are working together. I’m looking at the interoperability of what I currently have and now these agents that are being introduced into my ecosystem. These are all things that I’m looking at from a perspective of I want to have a control plan around agentic AI. We’re understanding my inventories, my identities, my permissions, my data access, my observability, my testing, and most of all human oversight. There’s a lot of things that are taking place, but if something should happen, what’s going to be my recovery strategy, what’s my third party dependencies and what’s my business continuity planning? Because the planning that I had before is not going to fit my structure. Today I’m looking at funding.
How can I find AI autonomy but also have fun AI accountability as part of this process. When I look at this across the board, I see a lot of companies that are going through this today.
Even though they don’t have a vision statement, they still are dealing with shadow AI and they don’t realize they’re doing that. People are using these artificial intelligence technologies, these services unbeknownst to the company, because they don’t have the capabilities and yet to monitor what’s taking place. And it’s scary because when they do get it now, and the people that have actually started using it now, it’s kind of coming to the business culture. Help them understand how to do it in a format that’s going to give them an roi, give them the return that they’re looking at as far as the capabilities and services that particular business as you’re looking to invest, the whole goal here is how can I make money. But a lot of times I see companies are not looking at the risk associated with the money they need to make and with spending that I see if they’re spending money on how to generate revenue using agent AI agents, they need to be spending money, how to maintain, monitor and hold those AI agents accountable. And too many times the risk is an afterthought and not a forethought when it comes to putting things in place to help make the company more valuable, make it more marketable, make it more valued overall across the board from all the different things they’re trying to work with.
[00:25:36] Isaac Sacolick: Thank you, Derek. Our last to speak today is Joe.
Joe, why do I feel like you’re going to go in a different direction.
[00:25:43] Joe Puglisi: Well, everybody has really covered everything that I would expect to be in the planning phase of this effort and so outlining the business objectives, making sure that we are training both the interior, the IT staff as well as the employees so they’re well educated.
Investing in an orchestration layer, as Joanne has suggested, attending to the security and the risk and the readiness of the data. All these things have to be elements in the plan. I’m assuming those are in the budget.
So if you’re asking what’s the incremental ask, it’s in the vein of extraordinary governance. I think, as Joanne and Derek have suggested, there’s tremendous risk here and we really don’t know, as Raymond says, how to make agents behave yet. So I would want to add a little extra measure of monitoring for two reasons. One is to mitigate risk, but the other, and something that I haven’t heard mentioned yet, is what are the metrics by which we’re evaluating the success of this collection of 200 agents? You know, Liz said you should have started with a business objective or set of business objectives. How do we measure against those objectives? How do we weed out the agents that aren’t performing to standards or, or have created some sort of risk or data leakage? How do we know what success looks like? I would want to put a little extra funding in there to make sure we had the right people and processes in place to monitor the outcome of our efforts and to tune, fine tune and correct as we get toward the end of 2027.
[00:27:37] Isaac Sacolick: Thank you, Joe. I’m going to take my break here with one caveat. I’m going to play out Martin’s role.
Martin’s not here this week. And say, and maybe to some extent Heather’s role. And say, look, if my objective is to take advantage of what we have in our ERP, HR and CRM systems and get our staff using as many as 200 AI agents, I’m going to simplify this. I better have training and change management up front, right? That’s going to change everybody who is involved with these agents in finance, in hr, in sales and marketing and how they’re doing their jobs. If I want to have a healthy adoption, I better have a change management and training program. And maybe I alleviate all the needs around security and resiliency and things like that by saying, you know what, we’re going to keep our agents primitive in 2027 and prevent the orchestration that will require some of those platforms and some of those capabilities. So Maybe there’s another scenario for you to consider as we think about our 2027 AI budgets. We’ll go into our next scenario after the break. Folks, thank you for joining this week’s coffee with digital trailblazers closing in on our fourth year of doing this. Just want to thank all of my special guests, our panels, our experts here, and to all of you who have been joining us consistently, session after session, whether live or on demand. If you want to watch a program that you missed, go to drive.starcio.com Coffee and our episodes are there for you to listen to and watch.
On the 25th, we’ll be talking about getting hired in the AI era, how to stand out, signal judgment and land the role.
On October 2nd, we’ll be getting into building agents and whether you should vie bye or automate, deciding how to build AI agents. And then on the 9th, talking about change management, why AI adoption stalls and how to measure progress. I’ll be announcing all of October’s lineup at next week’s session, so tune in for that.
Some of my upcoming dates. I did an event last week with Camunda here in New York City. Had a couple people who joined that one. If you are in Houston, I’m hosting a roundtable on infrastructure leaders on AI’s hidden data debt. So if you’re in Houston October 27th, it’s a breakfast roundtable. I’ll paste the URL into the common stream. Do let me know that you want to come and do register if you’re in Houston to this great event. Okay, I’m going to give everybody another scenario here. John’s raising his hand so he can give me a quick answer. We’re going to go in the opposite direction.
We’re going to talk about the CIO. The CEO wants to plan for a recession in 2027, expecting the bubble to burst. Not sure what happens next.
What’s the area of the budget that you’ll fight to protect? I want one area that you think that someone is going to say, you know what, maybe this is where we need to cut from. What are you going to fight to protect, John?
Then what’s another area where you look to make cuts? Can you answer that, John, before you run off?
[00:30:56] John Patrick Luethe: Yeah, Isaac, thank you. And I just wanted to make one comment just, just on the, on the previous question.
I actually some of the biggest agents are, are driven kind of top down. But I actually, I see most of the agents, the deci, the, the need for them being identified and the decision to create them being, being done Organically at the companies I’m at. And these, these are, these are my friends companies, you know, big, big telephone companies, big big tech companies. The agents are, are kind of like, it’s, it’s kind of like the, the people that are in the trenches are building these things to make their lives better. And so that’s, that’s why I see so much of the growth, especially by the numbers. The percentage of them is, is so much, is so much driven by, by people that are in the trenches doing the work, they’re building agents to make their life better. So by volume I really think it’s, it’s kind of like bottoms up growth.
The area that I’m going to protect, it’s going to be AI governance, like looking at applications like Portkey and things like that and security so that they’re only using the blessed corporate tools.
So if you want to use an AI tool, it goes through the corporate thing, it’s all governed. We can track costs. That’s the one thing I’m going to battle really, really hard for. And the area that I’m going to look at to cut is I’m going to go look at all the IT spend in cloud and I’m going to make sure that we’ve done the housekeeping to make sure that the cloud spend that people have is tagged, it’s being used, it’s not just sitting there. So housekeeping on the cloud costs and then I’m going to go look at the SaaS and make sure that we’ve done really good housekeeping, make sure that we don’t have duplicate SaaS products, maybe competing SaaS products and then I’m going to go make sure that, have somebody look and make sure that the seats are actually being used and I’m going to go through all that SaaS spend to fund the stuff I really want to do.
So yeah, that’s where I would put my focus. Isaac.
[00:32:44] Isaac Sacolick: Perfect. Thank you for joining this week John. I know you have to run off. Go ahead Joanne.
[00:32:50] Joanne Friedman: Okay, so what would I fight to keep?
It would be agentic execution and the control plan that I just talked about, what I would look at cutting to John’s point, definitely cloud spend where it’s not being used.
But I would also look at cutting token spend because one thing that people often forget, yes there’s you know, big brand name frontier models out there but there’s also a lot of open source and open weight models that can be used just as effectively. So I would look at cutting waste in token spend because we don’t. You don’t get a token back when you get a hallucination as an answer.
So optimize, optimize your use of the most expensive tools for what they’re geared for.
And remember that these are natural language processors that are only beginning to do math better and some of them are getting way better than one would expect. So you could optimize your budget by cutting your token spend on unnecessary usage. Teach people not only how to do prompt engineering better for an LLM, but teach them what, what kind of agents work best in what environment.
It’s the harness that counts, not the model
[00:34:19] Isaac Sacolick: optimization. I’m just going to create a word here. I love that. Joanne. Go ahead, Derek. What are you protecting and what are you cutting?
[00:34:27] Derrick A. Butts: So I would fight to protect when it comes to identity data security, AI data detection and AI cyber resilience.
When I look at these particular things, they’re all part of the necessities of keeping the business operating in case something goes wrong. And the biggest thing in looking at this is if the. I think John mentioned this earlier is those approved AI applications. So there’s a lot of redundancy I see in companies. They’ve got a lot of applications, some of them overlapping and duplicating the services that they already have.
Streamline it. I find that at least 30% of those applications can be cut if we cut. If the business entities would learn to work with the applications they have and utilize some of the capabilities that they offer. I would also look to make sure, as we’re looking at this, the, the identity, so forth, moving forward with the products is the monitoring. That’s going to be something I want to make sure is, is front and center. Because with AI moving at the speed that it’s moving and it has the ability to go into different applications, we need to be able to track that. We need to be able to understand what is it doing it looking at how is that going to impact the company or the data in which it’s trying to extract. I would also look at the, the, the meaningful use of what people are paying for. And I like what Joanne said about the tokens. A lot of tokens, you know, it’s a lot of money and a lot of companies found this early on. The budgets are being busted because the people are using more tokens than what they expected are budgeted for. So if I’m looking at the systems and I’m looking at how they’re utilized and I would actually look to make sure that people utilize them in a better manner to streamline the way the funding is actually being allocated. But also make sure that the software, the cloud systems, what not exactly what they need and they’re not overabundant. If we don’t, if we have something and it’s redundant from a resilience point of view, we have another one that’s just nice to have. Let’s cut out the nice to have until we get something that’s more solid, that will allow us to continue to improve our business processes, to improve our measurable KPIs, but also improve the accountability of owners that are taking place and utilize these technologies throughout the workforce.
[00:36:30] Isaac Sacolick: Interesting.
Very interesting, Eric. In terms of, you know, protecting identity management, if anything, I’d probably double down on that.
You know, you know, the simple idea that treating agents like people in your, in your authorization statement I think is under thinking the permissioning between what we want people to be doing and what we want agents to be doing.
So actually double down on that. And then you sort of said this, but I’m going to say it a little bit differently.
When you get into the IT ops and security ops areas, you know, maybe cut down some of the monitoring tools that are less effective or underutilized.
[00:37:15] Derrick A. Butts: Agreed.
[00:37:15] Isaac Sacolick: Double. Double down on the ones that are giving you a more holistic picture agreement tools use better is how. I would probably paraphrase what you’re saying, Derek. Thank you, Joe.
What are you protecting and what are you cutting?
[00:37:28] Joe Puglisi: Well, I have in the past and I would continue in the future to protect education and training.
[00:37:35] Isaac Sacolick: Thank you.
[00:37:37] Joe Puglisi: I think the most critical thing is to make sure that the people using these new tools know how to use them and that the IT staff know how to build and maintain them. And I would even cut external consulting spend at the expense of progress so that we drive the progress through internal resources and therefore increase our ability to maintain it properly going forward.
[00:38:08] Isaac Sacolick: I think that is smart thing. I don’t think that’s very.
I mean, I think that’s where a CIO is going to look for first is doing that. I think where they’re ineffective is that bridge gap between just training people to being able to absorb some of the leadership and management responsibilities that you’re relying on your vendors to do for you. And if anybody needs help with that, it’s a core competency of star CIO to help with that.
So if you’re struggling with it, let me know. I also left the URL for our upcoming research in the chat. If you want to sign up and get access to that we’ll be talking about the 66 platforms now that are doing AI agent orchestration. And then one of the examples Joanne mentioned in terms of being able to dynamically change models is not just I’m shifting from Claude to OpenAI and back and forth. It’s this notion of recognizing when you can use a smaller model than a frontier model to keep costs down and sometimes even to improve accuracy. Joanne, before you go, I want to hear from Liz. Liz, he said you were going to answer this question, but I’m not giving you a buy.
What are you protecting and what are you cutting? And you can lay out any scenario.
No, no free rides here. You’re on the panel. I need an opinion.
[00:39:37] Liz Martinez: Okay, so it’s so funny because I was just typing in the, in the comment chain. I think that security is obviously something that has to be protected, but I like the idea of protecting token optimization.
I think that a small amount of effort there could really help free up money that could be used for other things. Right. So I think that that’s a place to protect.
It’s funny that training, training dollars, I guess it’s just me, but I feel like training dollars, it seems to needs to go with the effort and the program that you’re trying to roll out. I think of it in terms of organizational change for significant shifts.
So depending on, you know, what you’re actually putting your money on, it needs to kind of go part and parcel with that.
Unless we’re talking about just generic, you know, AI familiarity, which think could probably be handled pretty easily by, you know, people just showing up to your coffee talks.
And that’s free.
[00:40:51] Isaac Sacolick: There you go. Thank you, Liz. Joanne, your hands raised. And then I’m going to give you my answer for protect and cut.
Go ahead, Joanne.
[00:40:59] Joanne Friedman: Yeah, one of, you know, to Joe’s point about training, one of the things that seems to have fallen through the cracks but is absolutely essential if you’re going to do AI and get a financial outcome that you’re happy with, is the notion of business process re engineering. And I think it’s coming to the forefront. I mean, we see a lot of, you know, there’s a lot of ads on LinkedIn now about system engineering and taking the bigger picture view. But business process is really where a lot of optimization in token costs, cost of AI risk and security comes into play. If you reorganize, if you start looking at your business processes from the perspective of now, I can use a new tool to get greater efficiency and really impact ebitda.
What you want to do is look at how you can use business process modeling and business process optimization to get you farther, faster. Because if you look at the processes, you’re going to find a lot of inefficiencies in them. You know, a lot of them emanated either from a system like an ERP or even before that, a paper system. And it’s evolved over time and so forth and so on. But the role of AI as being a ubiquitous system that all companies will adopt and all companies will use across organizations and to something Elena said a long time ago is people’s roles are changing. Therefore, the business processes that they use should change with them.
And we should look at less of a workflow orient orientation and more of a business process orientation and model those processes, because that’s where token cost increases, because you’re actually overlapping another process. No process is an island. They all have codependencies, interdependencies, cross references. Start looking holistically and you’ll find that the outcome that you’re trying to achieve is much easier to get to because you won’t be siloing the agents, which is something that’s coming up very, very quickly. We’re reinventing the wheel with a different technology, but there’s still silos and they’re still fragmented.
[00:43:26] Isaac Sacolick: Thank you, Joanne. I just left on the whiteboard what I would protect and what I would cut. I’m protecting data governance.
It sounds like something that your CFO or, you know, somebody else is going to cut because traditionally it’s required a lot of people’s involvement to make it work. It’s really hard to have an AI governance and strategy when you’re not thinking about the underlying context and knowledge that you’re building up.
So I’m protecting my data governance budget. The areas I’m looking to get money from, I would say traditional rpa. This is also sort of dovetailing off of Joanne. What you talked about is this process reengineering. I have a broken process and I threw RPA on top of it to automate parts of it.
I’m going to unwind some of that and look to say, you know, what’s our new process with AI capabilities? We talked about some of that with Camunda a couple weeks ago.
You need to think about AI enabled processes to other areas that are related. I might go up and say, if my company is still stuck on spreadsheets, I’m killing spreadsheets.
I want those assets counted. I want to look for different ways of doing things. It’s just A data issue, it’s a process issue, it’s a time issue.
And there’s a lot of automation technology and data AI capabilities that can replace them. I’m going to use Liz as my helper to help figure out what the value is. It’s not a holistic. Every spreadsheet needs to go, but I’m going to find the ones that are taking too much time and talking about time, you know, if I want people to use the training that they are getting around AI, I need to free up their time to be able to do that. I’m going to kill poorly run meetings.
And every organization has them and every organization underspends effort in helping people run very basic meetings that deliver value. So those are my three areas I’m going to cut.
We’re now going to go into our final stage. We’re going to all be the CEO.
Okay, we’re all going to be the CIO or the caio. One of those people owns priorities or at least has to finally weigh in and say, you know what, we’re going to start off Q1 and Q2 next year with a bank focusing on X. And we just got way too much to fund.
We’ve gotten way too many areas to protect. And even when we’re cutting something, it traditionally means that we have to invest in something else to replace it.
Derek, what are you thinking about in terms of how do you make those decisions around where priorities lie and we can’t use. You can’t go back and say, you know, it depends on strategy or value. Give me a scenario and say under this scenario, here’s what I would do.
[00:46:26] Derrick A. Butts: I would use four executive questions that could be run across any application, any AI agent. The first one be, what is the value? What’s the outcome that we’re buying and how does this affect what we’re doing? Does it solve the problems that we have better, faster and at less expensive? Number two would be, I would look at the risk.
What could this create that we don’t have today? And I look at the risk from a financial customer. Risk, operational privacy, reputation, all these different things come into play.
I would also ask about the resilience. What happens if this particular system fails? How does that impact us when things go wrong? What’s going to be compromised? Will things be unavailable on Monday morning when my team comes in? How can I make sure that humans can take over if this anomaly should take place? And our last question, I would ask, what’s the readiness? Are my people ready for this particular application? We put in place. A lot of times it gets overlooked. They just push it out to the organization, they’ll adopt it, they’ll figure it out. And the cases, we find it out, that’s not the case. Before something gets funded and it’s scaled to get funded, I need to understand, do the employees understand how this AI fits into this work? Do they understand if they can trust it or not? Do we have managers and understand accountability associated with this rollout to my business culture. But most of all is someone accountable for the outcomes when this thing is implemented and it goes sideways. I’m looking at this from a people’s process, governance, organization, readiness point of view across the board. And if I can get all four of those questions answered, it’s going to make it easier to figure out what gets cut and what doesn’t.
[00:47:58] Isaac Sacolick: There’s your framework, everybody that I think Liz would agree on. Because when you look as a CIO or cio, maybe not so much the CEO, but you start looking at these things and saying, what am I going to put at the bottom of the list? It’s the one that doesn’t have a clear value statement around it. It’s the one that says the risk is worth more than the reward.
Derek, thank you for laying that out.
Liz.
He, Derek stole your thunder so you could have to go somewhere else.
[00:48:27] Liz Martinez: Listen, what Derek said in terms of a framework for, you know, validating that this is a viable initiative, 150%. I love it. I would venture to say that the answer to this question is actually that’s nothing to do with AI, and it has everything to do with the strategy and if the hard work. So listen, executives get paid a lot of money to actually figure out where they’re going to take their, their business.
That’s their job.
So they have to think through what initiatives are going to hit us. If top line is the most important thing, what initiatives, where. Where are we focusing for top line? What is the segment that we’re going to actually target? What is. How are we differentiating ourselves there? Then you start picking initiatives.
That’s when you lay out your initiatives across the year.
Then you go back to, and I, I don’t mean back, but you implement exactly what Derek was saying to make sure that those initiatives have solid legs and can actually execute appropriately.
But if you’re not actually thinking about, if you’re worried about how can I leverage this new AI tool and this agent that somebody came up with, it’s just the wrong place to start. It’s a good place. If in Fact, it was some of your frontline workers, the people who are in the trenches, who are doing shadow id, who are creating these agents on their own. Clearly they’re trying to solve a problem that needs to get surfaced. That’s where you can start. And then you say, okay, well how does this map to, let’s say our top line initiatives that we’re trying to go and how can then we use this pilot or this example and then broaden it to make it actually have legs across the organization?
[00:50:14] Isaac Sacolick: Folks, you’ve gotten two frameworks for the price of one. Derek’s around how you make priorities and Liz, unpacking what should be in your business’s strategic statement. Am I growing top line? Am I grow market growth? I’m going into new markets, unit cost improvements, lots of ways to express strategy.
And yet what you end up happening in many organizations is they don’t or it’s all of the above.
[00:50:42] Liz Martinez: Right?
Too many number ones. Right.
How many thing is number one all the time. And if you’re interested, I actually have a product offering to help your executives talk through that and make sure they’re clear on what their strategy is.
[00:50:53] Isaac Sacolick: Or sometimes, Liz, you do the other thing, which is if you are after top line growth, here are the one or two things I would do. If you’re after market growth, here are the one or two things that you can do.
Give them the answer, then make them make a choice.
Couple way to think about it. Joanne, are you going to lay out framework number three for us?
[00:51:15] Joanne Friedman: Well, I could, but my number one would be not looking at markets, industry or segments or even product offerings. It would be how do I create differentiable value for my organization so that I am more valued by my customers and more valuable to my suppliers.
That’s where I think people are starting to go and they’re looking at Agentix as a means to get them there. Because what disappeared way back when at the turn of the millennium was differentiable value. The minute we started democratizing data, differentiable value fell out the window and companies want it back.
Their ease of doing business with them or whatever made them different in their market. That’s their new competitive advantage. Again, they want it back and they’re looking at every which way to do it.
So from the C suite, I would say if the ca, if the CEO or the board even determines that differentiable value is where they want to go, then there are plenty of frameworks that one could use to get them there. But the priority would be talk to your people and start capturing their Institutional knowledge, because it’s not just in manufacturing, but across the board.
As much as 40% of industries will have retirees over the next two years. So 40% of your workforce retiring means you’re going to have a huge knowledge gap. And that knowledge gap is your differentiable value. You know, people are your first, most important asset, data is your second. So capturing their knowledge in data is the way that you’ll capture that differentiable value to reinstate it and also to support it for resilience. And AI is a very good way to get you there because of the speed at which you can do things.
[00:53:22] Isaac Sacolick: I love that, Joanne. And starting with differential value, particularly around AI, and then looking at a people first strategy, particularly in organizations where the negative of it is tribal knowledge. The positive of it is you need deep subject matter expertise to make consequential decisions. And that’s happening in everything from manufacturing, construction and health care. To lay out some examples of that.
And so now you’re saying, well, what’s going to help make my people more effective? I think that’s a great one. Before I go to Joe, I listed out two of them. Here I am looking for more AI and customer experiences. And that’s not just in support centers.
When I think about what business is going to look like in two to five years, you can’t tell me what you’re doing today for customers is going to look identical to what you’re doing today. And yet most of AI is still focused on productivity and back office process. So if you want to get ahead and really think about where to make bets, now it’s one that’s going to Joanne’s point differential value.
And my two bets around that are AI and CX customer experiences. And then looking for what I call force multipliers. This means the Liz’s question find things that impact multiple strategic benefits. Growth has to be one of them because with all the cost economics optimizations you’re going to do, you cannot afford AI just by cutting costs. It’s not going to work. We’ve seen this before. Joe.
Joe, your thoughts?
[00:55:04] Joe Puglisi: Yeah, I think you’re exactly right on your last point before I answer this question.
We’re all looking for, I think we’re all looking for the Uber moment. We’re looking for when companies recognize that AI is not just cost cutting or optimizing their current business models, but rather it’s reinventing the business and perhaps creating new businesses from the ground up.
But to go back to your question, Isaac, I have a pretty simple model. I think it’s along the same lines as what Derek articulated very well.
First and foremost, does it have measurable outcome?
What does success look like? I want to know what the objective is and can we measure it? And by the way, that isn’t only CFO measures. Everybody talks about roi. There are lots of different ways to get value.
Joanne is expert at this, and so is Liz. You know, I, I recently heard a story about a, a nonprofit where the value proposition for having done some AI implementation was better retention of the workforce.
You can measure it, but you know, that’s not, that’s not one that comes up high on the CFO’s list.
The second is what are the risks? And I think Derek always speaks to this risks in terms of a potential failure. Do we have the data?
[00:56:26] Isaac Sacolick: Do we have the platform, do we have the expertise?
[00:56:29] Joe Puglisi: And then also the security aspect of it. Am I going to lose data to the, to the outside world, or am I exposing myself to, to a cyber risk?
So do I have a measurable outcome? What are the risks? And then the third leg of my stool is the probability of success.
What is the likelihood that we can pull it off? And someone has to convince me, if I’m the guy making the priority decisions, somebody has to convince me that we have the right objectives, that they’re in line, as Liz would say, with the business, that the risks are minimal or containable or addressable, and that we have the right resources, both data and expertise, whether it’s internal or external, to pull it off.
[00:57:15] Isaac Sacolick: Of course, Joe, you land with the MIC dropping moments. Number one, defining and measuring success and a lot of different words around this, but, you know, posing it, that question makes you think about, it’s an easy question to get at, what value or intent or outcomes. Those are just, I, I just find you just get better responses from people around that.
And so, you know, if you’re struggling to get those answers, start with what you know, how, what does success look like? What does the headline test look like? You know, those are some things. And then your question on risk, Joe I translates to what are the new risks that we are ill prepared to address?
And, you know, that’s not just an AI question. There’s a lot of business risk happening today that if we don’t have a plan for and we think it’s important enough, we better put that at the top of our priority. Joanne, your hands raised. Last, last comment for today.
[00:58:21] Joanne Friedman: Last comment, very quickly, because it hasn’t come up irrevocability.
You know, people look at AI and Go. Oh, it’s an on off switch, right? I can. I can stop an agent at any time before they go rogue. It’s. Etc. That’s the wrong analogy to use. It’s a gate and you have to look at the most severe irrevocability case as the baseline.
If something happens to a human being, clearly you don’t want that to ever happen and you have to mitigate the risk around that. But irrevocability is I can’t take this back. Am I damaging my brand? Am I causing business process to go out of whack?
That should be one of the priority issues that’s discussed with the C Suite in terms of your vision statement is great, but what is the irrevocability clause?
How can I take it back if I can’t take it back if it’s going to cause that much damage? It’s not about risk mitigation. It’s about just don’t do it.
Be aware that there are those levels of variability.
[00:59:28] Isaac Sacolick: Love it, Joanne, and what a great session. Geez, if you were lost enough, we just gave you 10 new avenues of thinking and how to think about it when you’re planning your 2027 budget. So thank you John, Joanne, Derek, Liz and Joe for your insights and for playing this little game. We’ll have a little conversation after and see if we should do something like this again.
Folks, our upcoming episodes. Next week we’ll talk about getting hired in the AI era. Standing out, signaling judgment, landing the role. The second we’ll talk about how we’re building apps and agents. Vibe Buy or Automate? Automate is in there, meaning we’re going to choose to do something that’s not AI oriented, deciding how to build AI agents in the future.
And the ninth we’ll talk about change management, why AI adoption stalls and how to measure progress. I’ll be announcing the other October events at next week’s look for it.
My upcoming travel I will be at Web the Cisco’s Web Summit coming up in October. This is.
I’m losing my language around it, but I’ll be at their web event in Austin if you’re going to be there. I will be at work day the following week. I finally main ends meet. Somebody asked me this week. I think it was Kristen. If I was going to make it. I think I will be there at workday that’s in Vegas and then my event with Hitachi vantara in Houston October 27th.
I left you the URL starcio.comai houston-roundtable if you want to sign up for that executive roundtable, folks, have a great good weekend. Take it easy, and we’ll see you here next week. Thank you for joining.
Leave a Reply