GTM 139: AI Agents Are Changing Everything — Microsoft’s VP of AI Agents on the New Era of Work and Software | Ray Smith

The GTM Podcast is available on any major directory, including:

• Apple Podcasts
• Spotify
• YouTube

Ray Smith is the VP of AI Agents at Microsoft. Previously, he was the Global VP of Product for SAP, CRM and Sales Cloud. Before that, he was the CEO and Co-Founder of DataHug, which was acquired by Calidus Cloud in 2016.

Ray breaks down why the rise of AI agents is a tectonic shift, how businesses are already seeing ROI, and what it means for SaaS, team structure, and go-to-market strategies. He also shares real-world use cases from inside Microsoft and partner companies, plus how founders and operators can build or adapt in this new AI-native era.

Discussed in this Episode:

• What exactly is an AI agent—and how is it different from bots or automation?

• Why traditional SaaS pricing models (like per-seat) don’t work in the agent era.

• How enterprise teams are already deploying autonomous agents in production.

• The emerging architecture of multi-agent workflows.

• Which skills will matter most in an agent-first world (hint: think like a GM or growth hacker).

If you missed GTM 138, check it out here: The AI Advantage, Solving Sales & Marketing Alignment for Better GTM Execution with Jaleh Rezaei

Highlights:

06:53 What makes an AI agent different from bots and automation—and why it matters now.

15:33 How AI agents are changing the role of traditional software tools and UI.

25:15 Why legacy SaaS pricing models don’t work for agents—and what comes next.

31:12 Inside Microsoft CoPilot: Real-world agent use cases across sales, support, and strategy.

37:42 The most valuable skills in the AI era—from delegation to agent orchestration.

51:42 Why agentic AI makes customer relationships and long-term value even more critical.

Guest Speaker Links (Ray Smith):

• LinkedIn: https://www.linkedin.com/in/raycsmith/

Host Speaker Links (Scott Barker):

• LinkedIn: https://www.linkedin.com/in/ssbarker/

• Newsletter: https://thegtmnewsletter.substack.com/

Where to find GTMnow (GTMfund’s media brand):

• Website: https://gtmnow.com

• LinkedIn: https://www.linkedin.com/company/gtmnow

• Twitter/X: https://x.com/GTMnow

• YouTube: /@gtm_now 

• The GTM Podcast (on all major directories):

The GTM Podcast
The GTM Podcast is a weekly podcast hosted by Scott Barker, GTMfund Partner, featuring interviews with the top 1% GTM executives, VCs, and founders. Conversations reveal the unshared details behind how they have grown companies, and the go-to-market strategies responsible for shaping that growth.

GTM 139 Episode Transcript

Scott Barker: Hello, and welcome back to the GTM podcast. As always, you’ve got Scott Parker here and I am joined by Ray Smith. Ray, welcome to the podcast, man.

Ray Smith: Well, thanks, Scott. Delighted to be here and looking forward to the discussion today.

Scott Barker: Man, we couldn’t have a more timely and topical guest. I feel like right now. I’m sure you feel it, but everyone I talk to every CRO, every founder wants to talk about AI and even within the last, I’d say three months, they now want to talk about specifically, AI agents and for the listeners, I’ll just cut right to the chase by giving the bio and you’ll understand why, this man Ray is the perfect guy to have this discussion.

He is currently the VP of AI agents and moonshot projects at Microsoft. It’s literally in the title. Previously to that, it was the global VP of product for SAP, focused on the CRM and sales cloud. And before that he was the CEO and co-founder of DataHug, which was acquired by Calidus cloud in 2016.

I think I got to start by maybe putting you on the spot a little bit. I mean, moonshot projects sound pretty fun. are you able to give our listeners maybe a sneak peek or preview of any of the moonshot projects you’re working on? Or is that all behind closed doors?

Ray Smith: Yeah, I think it’s two years ago, it was definitely termed the moonshot project because the whole thesis was the future of AI is not going to be just this chatty interface or LLM that we’re going to interact with. It’s going to fundamentally reshape how we are. Live and work or run our businesses. And for that reason, it was termed a moonshot project over two years ago.

Now I would almost say we dropped the moonshot project. I’d probably need to update the LinkedIn. And mainly the moon now.

yeah, we’re, we’re, we’ve landed on the moon and it’s like, you know, businesses are getting real ROI and, you know, fundamentally transforming with this technology. So yeah, we’ve landed on the moon. The exciting part of my job now is kind of the constant innovation around this space and then seeing all these kinds of customers and partners building these exciting solutions. I regularly like the hairs on the back of the neck when a customer or partner says, let me show you what I built.

And then you’re like, oh, wow, you know, this kind of goes beyond even, you know, what we thought of maybe two years ago. So. Yeah. So leading the, the kind of agent, transformation at Microsoft and, that’s obviously within the co-pilot studio. So Microsoft co pilot studio product, but the real focus today is really on these autonomous agents as we talked about.

Scott Barker: you couldn’t have a more interesting and exciting role. It must be almost a little weird for you because this is something that you’ve been focused on and thinking about for a really long time. I remember you mentioning like you wrote a paper. I can’t remember what year it was in. You can maybe let us know, but you, you actually wrote about.

This idea that the future of AI agents isn’t this, you know, chatty Cathy interface, it’s going to be workflows. And that’s exactly what has kind of played out. When did you write that paper?

Ray Smith: Yeah. I honestly, Scott, I’ve been on a journey and the highs and lows in the AI space for probably over 25 years. So like, if I go back. 25 years ago, I was writing like I was writing papers and doing research on brain computer interfaces. so early neural nets, and I’ll put that as my kind of early optimistic or naive phase that it was going to change the world.

and then like, if I go to 2010 in the startup space. building data hug out, which was in the predictive forecasting, you know, pipeline management space. I thought AI was going to be the big unlock, but it really did fail to deliver on the transformative ROI that businesses were expecting from AI.

And then two years ago or maybe two and a half, three years ago now, this new technology with LLMs came through and we, as the humans, as humans, we’re running around like with hammers in our hand going, is that a nail? Can we use it here? Can we use it here in the early use cases where Content summarization, searching the web, creating creative content and so on.

So that paper you mentioned was really kind of heresy, like the future isn’t going to be these little chat windows where we just ask questions, get nice answers back, and then copy and paste that to our bosses and sound really intelligent. It’s going to be. AI capabilities that we will delegate tasks to, it will run processes in the background.

And yes, we, as humans, will kind of manage those AI, and this is what’s now become agents or agentic architecture. But that was at the core of it. People felt this technology couldn’t be controlled, couldn’t be harnessed, like we were always talking about hallucinations but really this has just become a control framework, a way of building apps or programs or solutions on top of this new technological breakthrough.

That’s fundamentally what. Agents are, but as a result of this, there’s both the brain or the models that get a lot of attention and focus on breakthroughs. But there’s also breakthroughs happening in these kinds of control frameworks to keep these solutions on the rails and deliver outcomes and reasoning where otherwise humans would be needed.

If that makes sense.

What makes an AI agent different from bots and automation—and why it matters now.

Scott Barker: So you’re only, you know, 25 years too early, but the future you predicted is finally, finally coming to pass. I would love it, you know, just as we kind of set the stage and, and go down this topic, could you define what an AI agent is? I think particularly in the last five, even 10 years, the use of AI was used pretty freely, in different software companies.

And it could have been described like basic machine learning, or just like kind of an automated spreadsheet on the back end, but they threw AI on it. Because something like that was happening. And now we’re in this new world. Like what is an AI agent in your eyes?

Ray Smith: It’s a great point, Scott is like, as I mentioned, and that cynical phase around 2010, it seemed like every startup, every business was slapping AI. So it became like a marketing buzzword. Um, And we’re, you know, you see similar patterns with any new breakthrough technology, but the big difference with agents is what we’re seeing is real ROI, real value being delivered to customers.

And there’s lots of publicized examples out there and both at Microsoft and other companies, but the key is the value and the speed to that value is, is, is what’s different from let’s say any other, you know, technology breakthrough or, marketing term or label. The biggest challenge that exists in the market today maybe is there’s a huge spectrum of agents, you know, and there’s agents like where they came from, traditional bots and automations where they’re like.

Hey, this is now an agent because I sprinkle in some LLM uses or scenarios around it. and it moves towards the RAG or retrieval based agents where it is reasoning over data sources just to give you an answer, but then it moves more towards kind of tasks and like autonomous agents where it is.

Getting entire work done or processes done in the background and only bubbling up to a human when it has an exception. So I think there’s lots of people in the market asking like, what is an agent? And various people have various different definitions, because they’ve maybe approached it from different perspectives, whether that’s historical or where their tech is today and so on.

But for me, agents at the real core of it is like the ability to plan reason. tool selection. So it’s kind of like coming up with something based on giving it a task or a goal, it’s going to reason on a plan on how to accomplish that. And that’s going to be dynamic. That’s going to be dependent on circumstances.

And this is why it’s a key on lock before we’d have to build either extremely complex programs or automations to handle all the, if this, then that, or all the permutations of life or the complexities, if this customer has this policy or has this kind of as a high network individual, then do this or do that.

So all that complexity was factored into a really complex program today, which was costly and hard to maintain with all those business rules. Or it was given over to a human to say, Hey, process this insurance claim and check this and check that and do these things where now we get to give it over to an agent that will reason over all those variables, follow some guidelines and policies, including the guardrails.

To then call on these tools to accomplish these actions. So I think the key components to an agent, yes, there’s knowledge. We’re all familiar with that for rag based agents, but really goes into the tools. The planning, the orchestration, and then the evals and evaluations to make sure it’s doing what it’s supposed to.

So there’s a couple of different components and each of those fields I mentioned are exploding with breakthroughs and innovations, all by themselves. Even like in the testing frameworks and regression testing agents and AB testing and all these things, there’s so much going on. But the key thing is, it’s really about the outcomes and the solutions.

And if someone says, Hey, this is what I mean by an agent. Just ask the question, what is the agent doing? What is it controlling? And how does it handle the variability of my business? You know, cause sometimes it could just be, that’s just an automation, you know, you don’t need to spend all those tokens or calls to LLMs if it’s just a simple, straightforward automation.

Scott Barker: Yeah. And would you say the, the main difference then between just setting up like some automation using, you know, Zapier and, and different APIs versus an actual agent is the, is the reasoning component. Is that kind of the main difference?

Ray Smith: Yeah, so they, you touched on a really good point. The connectors into the existing tools and systems are really critical. So at Microsoft, we power automate. There’s other tools in the market that you mentioned, those connectors into SAP, Salesforce, Oracle, Workday, all these tools and systems. If you can give an agent those connectors and say, I want you to update this record, or I want you to get me new leads.

And I want you to score those leads, or I want you to qualify that. And then I want you to go into this tool, pull out usage metrics. It’s very easy to give it these connectors, give it the kind of business logic, almost in natural language. You just say, do this step, do this step. That’s a real key unlock compared to say automation today, where you would have had to almost program each step of the process, and then it wouldn’t really be able to handle variability as much.

Scott Barker: You know, it’s so interesting as I’ve, throughout the course of my career, Max, my partner and I have used and leveraged, a lot of virtual assistants like over in the Philippines, super high quality work there. They’ve been incredible but there’s sometimes a bit of a language barrier. So you really have to think through the workflows and the processes that you’re giving them.

And it’s usually like a very detailed sort of Google doc, here’s the login for X. You’re going to update these fields and then you’re going to. Turn this into a blog post and yada, yada, yada. And like that exercise of doing that, I feel like it has prepared me for these agents, because it’s a very similar thing that you can now do fully autonomously and have basically unlimited of these things working for you.

And, could you touch on like multi agent interaction, and kind of orchestration and what that looks like?

Ray Smith: I think the key learning from agents early on is if you cram too much into one agent, it really kind of struggles with the determinism. Like if you put too many actions, too many business rules and kind of my rule of thumb is. If we, as humans, would have to read like a set of instructions that we give to an agent, we’d have to read it two or three times or multiple times to really understand if it is there, is it ambiguous or not?

Well, then likely it needs to be broken out into sub agents that are effectively like SMEs. And this is a paradigm we’ve seen already in like app development or solution development over the last couple of decades. Like the ideas behind microservices where you, you build kind of like, um, Well defined atomic capabilities that can be easily maintained that can be kind of like, leveraged across the business so Obviously, this is pushing the need for these multi agent orchestration and there’s kind of agent to agent chaining, but then there’s true multi agent dialogue where they’re figuring out who’s breaking down what part of the task to accomplish the Uber goal or the Uber request.

So, we will see an explosion of multi agents, because that’s just. The design pattern is to accomplish any process. It really does need to be broken down into parse and process the invoice, then check the ERP, then do these things. And you’ll build a series of experts or, Hey, this is my, you know, I want you to help me to create a content for this topic but I want you to then call our brand agent to make sure we’re using the right language and style guides and blah.

So you’ll. Build a series of agents that may be owned by different departments and they’ll kind of be chained together to accomplish like a process. And, and that’s in the business world, even in our personal lives, we’ll all have personal assistants pretty soon. And they will be powered by a series of agents or skills or capabilities that you enable to help you shop or to help you inquire about, some local kind of.

Product or service and we’ll, we’ll stop going to the web as much as we do today because. You can just have a conversation to find out the opening hours for the local store and navigate straight there. You don’t need to go onto the website for certain bits of information or even to just replace an order you’ve already placed before.

So we’ll see multi agent architectures becoming the norm and the standard, mainly because of that kind of maintainability. The ability to break down a task, and then it really does help with the kind of reliability or determinism because you make a series of experts that are really good at doing one thing, and then you chain that together to accomplish the goal.

Scott Barker: Makes sense. I mean, this is so fascinating and things that I think about all day long. I would love your opinion on where this leaves the kind of application layer software in your eyes? You know, like there, there’s one kind of case to be made where let’s say, you know, I have an AI agent that I want to send my invoices and follow up on invoices.

So I give it a, you know, QuickBooks login and it goes, and this software is still going to make its job easier as an agent, but if I give it enough data and there’s a repository of all the deals that have closed that should have invoices. Like will these agents need to leverage software or w if you have a system of record that is complete enough, can it just go and do that without leveraging some of these applications.

How AI agents are changing the role of traditional software tools and UI.

Ray Smith: Yeah, I think that’s the fundamental shift. So like, you know, to really underscore, this is a tectonic shift in how we run our businesses, how we serve our customers, and how we deliver products. Mainly because thus far, we’ve had to say like, we want to scale, we want to grow. We have to just keep pouring people into that equation.

To be more productive, to deliver on the outcomes. Basically, now what we’re saying is agents come along and allow us to scale very efficiently, deliver very personalized experiences. Like I would even say in a couple of years, the way we scream down a phone today is like, I want to speak to an agent, a human agent in the future is going to be like human. I want you to transfer me to an AI agent. Cause I’ll remember all

Scott Barker: You’re too inefficient.

Ray Smith: Yeah, all my transactions, all my personalization, it will know exactly, and it will be like having a conversation with your best friend type thing where it just knows you.

And so we are going to see a disruption both in the experiences and how we run our processes. And then even like the shift of the people, we do have an organization because the view is it is going to be humans. An AI assistant plus these agents, that is the future of work. So even the humans that are in that process, they’re going to shift from doing the frontline work or the mundane tasks to managing the exceptions that the agent is popping up to them or the output of what the agents do.

So that is going to cause a fundamental shift in how we think about business applications today. So today, business applications, largely user license driven. So you pay for a number of seats. And so if you need 3000 people to perform a function today, maybe in the future, it’s a thousand or 500. Plus agents that will kind of infinitely scale to, to, to kind of do parts of the process even in between.

So that’s a challenge for traditional SaaS vendors or applications, particularly as the tooling is emerging, even with Kua, like so computer use agents, where it can move the mouse around on a screen, more connectors, more automations, the more tooling we give these agents, the more capable they are. And as a result, you start to ask the question is like, why would I log into that SaaS app now to move around a forms over data experience when my agent could do 90 percent of it, and maybe the 10 percent that pops out into some dynamic UX or UI experience that’s on my workflow and convenient.

So, you know, and this is what we see in the market. Like fundamentally all SaaS vendors are like, I’m building a higher order, higher value agent that sits on top of our data. No one knows our business quite like us. So therefore we’re going to give you the best agent solutions or agentic solutions. and that’s, we’re going to see more and more of a trend.

Like the, the fear with SaaS applications is they don’t want to just be a database or storage for these agents to interact with.

Scott Barker: Do you see then the opportunity for maybe, you know, founders listening to this or early stage teams or people thinking of, you know, building a company, do you see it less so in like these horizontal applications and more in these like specific niche, maybe verticals, um, where, you know, the larger LLMs are, you know, it might not.

Understand the job of a customs broker or, you know, how a telco ISP operates and like you go into those, you know, very niche verticals and start building, you know, some agentic workflows.

Ray Smith: spot on. It’s got like the old world where you build a horizontal platform and obviously you supported customization or extensibility. And you said, Hey, tweak that for you and finance versus pharma versus manufacturing and make it your own. And we all know that was painful. They were big implementation projects.

Lots of failed implementations because someone didn’t understand the data model or didn’t think about the process flow. and there were some of the challenges. Like I think what we’re seeing now is highly focused use cases or agentic solutions. You know, staying in an industry, nailing a couple of use cases and being very successful, maybe they get like 10 customers, each paying them, you know, multi million ACV contracts.

But they’re like on the back of that, getting to maybe 30, 40, 50 million in revenue quite quickly. but they’re staying focused on those use cases and then they’ll, they’ll eventually kind of go horizontal. So what I would say is. The challenge we always had in the past was customization and tailored.

Most customers, particularly in ERP, still also in CRM, would say, I’m a snowflake. My business runs differently. We do things differently here. So that level of customization is hard to do. In an agent world, you start with a template saying, I’m going to process leads this way. I’m going to score them this way.

And for the lead scoring agent, you just pop it open, change the business rules, tweak it to a different system, pull in information from a different source. And now you’re done. You’re like, you’ve started with a template, but you’ve made it your own. So I’ll see, I think we’re seeing lots and lots of startups and that’s the right thing to do is just build your beachhead with the first 10 customers.

Then go to 20. Stay focused on those use cases and crush it because it is a really high ROI for the customer, but it’s also a very high ROI for the startups. Um, because if you say, hey, claims processing, you don’t need like thousands of people doing all these checks. I’ll give you this agent. You know, low millions of dollars, that’s still a huge win for the companies, but it’s also a great, you know, high margin contract for the startups.

So we’ll see that focus. I think over time, then there’ll be the big agent platforms where these startups will say, well, do I dock in on some of these agent platform capabilities, or do I need to build it myself? I always am worried of these startups that say, I’m going to build a perfect agent platform with all these governance and observability and evaluations and testing capabilities, and they probably need that, but that’s going to be a huge investment that they may not want to do now, whilst they can feast on like nailing use cases for customers and waiting for the agent platforms to catch up.

Scott Barker: Is it a fair comparison to, you know, we had the explosion of, you know. cloud infrastructure and you know, many thousands, hundreds of thousands of startups, you know, built on top of Azure. They didn’t have to reinvent the wheel. They didn’t have to build their own cloud. They got to leverage certain parts of that infrastructure and then build on top of it.

Is that kind of what we’re seeing again with these large LLMs?

Ray Smith: Yeah, it’s exactly the analogy. Like, I think there’s probably two analogies I say is like, you know, we kind of invented. Mobile phones or iPhone and like there was an explosion of apps. So there was a race to build apps because there was this new route to customers or usage, our new way of doing things.

And then the second thing was when we moved. From on prem to cloud, but that is on a timescale relative to what this agentic disruption or transformation. It’s like, it’s a second. It just feels like such a short comparison to the migration to cloud, but you’re absolutely right. More and more of these things.

are starting to be viewed as infra and you’re just like, well, why would I build the info when there’s going to be, you know, a cloud provider that’s going to maybe offer these capabilities and I remember sitting in my office in Dublin back in 2010 and I used to joke around the hyperscalers were like, had our, our development floor bugged.

Cause every time we thought we needed to build something, um, A week later, they would, you know, naturally just release these capabilities and services and I was like, that saved us a truckload of work. And I think it’s a good analogy. Scott, it’ll be a similar pattern to all these capabilities.

Will emerge, and will be used as ways of kind of keeping these agentic solutions on the rails, testing them, making sure they’re efficient, using the right models, interoperating really well and so on. And the startups to just focus on. The relationships, like that’s the IP, that’s the moat today. It’s not the code.

It’s not the, even the designs to a certain extent is the relationship with the customer delivering on outcomes. So your own kind of domain expertise, plus your ability to build these agents and then just speed, like time to market. So you get to a customer and say, we’ll build this prototype for you POC in weeks.

And we can get you into production in months, like less than a month, even, it becomes a no brainer. And it’d be very easy for these customers to AB test. Cause it’d be like, we’ll channel some requests through the standard process. Some will go to the agent, they do a quick diff and go, Oh, this is actually, this really works.

And then they get the trust and then they do a full production rollout.

Scott Barker: It’s interesting and hopefully we all learn something from mobile and cloud and have applied it to this next shift. I guess hindsight is 2020, but looking at apps and mobile, the pricing to me, at least just always kind of made sense. There was an obvious way to. charge folks for the value they were getting similar to, you know, the shift to cloud.

I think that the seat model was just super widely adopted. Everyone used kind of the same thing. but with this shift, it feels like pricing and monetization isn’t as straightforward. You know, we can’t with the agentic workflows. Like, we can’t really use a per seat. model, unless you’re doing like a per agent model or something.

And we’re seeing people kind of test consumption based, like how much you’re using it, even outcome based. What are your thoughts on pricing? Cause it does not feel like being overly dramatic, but it’s an existential risk to some businesses. Like I see some of the incumbents and you could. Build, I don’t want to pick on anyone, but let’s say the incumbent RevTech platform.

You could build some of these pretty quickly with agents and come in and be like, okay, you’re using X, Y, Z company. You have 500 seats, scrap that, just use us every time you book a meeting or you do X, Y, Z. And that’s pretty compelling. What are your thoughts on the kind of this disruption in the per seat?

Typical SaaS model.

Why legacy SaaS pricing models don’t work for agents—and what comes next.

Ray Smith: I think, fundamentally, the user license equivalent doesn’t work for agents because it’s kind of infinite scale. It’s not tied to a seat.

And it’s primarily driven by consumption. So customers kind of pay for what they use, which is nice for customers. They don’t have these, you know, unassigned seats or seats sitting on a shelf. So that’s the, it seems to be the approach for, for agents. and. You know, how they charge for that is usually made up by the, we talked to the spectrum of agents from simple retrieval based agents for a conversation where it checks a document is very different to an agent.

That’s running autonomously consuming. Lots of spending. Lots of tokens to reason. To pick the tools to run automatically in the background and so on. So there’s a spectrum. And what we’re seeing emerges depending on the types of ingredients or tools that you use, spins kind of the meter at a different rate.

It seems to be kind of the. The emerging kind of approach, but for businesses that were fundamentally to your question, which was businesses traditionally based on user licenses. There is a disruption coming. And I think there’s, there’s a need to. Stay close to the relationship side, the stuff that humans can only do, like if it’s just like research and pre briefs and like some data analysis, that is easy for something to be done offline by an agent and brought onto the workflow for the human to kind of strategically think around how that gets used.

So we as humans are going to shift more and more to like being the quarterback, thinking about the plays, but we’re going to be fed information from these agents. So the more we can think about in our existing apps, how we become integral to the workflow, obviously augmented with these AI or aging capabilities. That’s going to be key to, you know, as this shift happens, that you’re not just.

It’s easily replaced by a few statements of a business logic into an agent that connects to these tools and replaces this whole SaaS application.

Scott Barker: Mm hmm. That’s super interesting. Your point around these different styles of agents. So you could kind of see a world where, you know, the human is there, not unlike managing a team where you have certain job functions that you can hire someone, maybe straight out of college that doesn’t have a ton of experience.

Cause you’re going to get them to do, you know, a fairly repetitive task. It’s going to build them some skills and then you’ll have others. Tasks or team management that’s going to require someone that is more skilled and it’s almost like people will have kind of their own agent budget that you can’t go above and you have to find, okay, which tasks are better for like the entry level agent, which require more depth and reasoning, and you’re kind of be like a quarterback of these different systems.

Ray Smith: Yeah. And, and that’s honestly the advice we give to kind of customers, is that how to get started is not be overwhelmed by all the use cases you can apply AI, because you genuinely can apply it everywhere. Like these agents can be built in any kind of function, both back office and front office.

So it really takes that kind of focus and that’s what we see startups doing really well is like focus on a use case. Um, and with that focus, then you kind of like hone in on a use case. And with that use case, you go, okay, this is what it looks like. And not just starting with like, this is what we do today.

Cause sometimes it’s like, Hey, wait three days for a human to check this and then send an email over to this person. But actually look at it from the customer perspective of like, what does this process need to be? Break that down into a series of agents. Even if you’ve got humans, either side of that, that’s kind of building the trust and verifying, yeah, it’s doing this check, right.

Or it’s the risk assessment is accurate. Now I’m going to pass it on to the next stage and so on. And then, you know, eventually you build a series of like the multi agents and it completes the entire process automatically. That’s the way I would, um, I’m seeing customers kind of getting started. And then once they’ve done that once it becomes almost like, as you say, this agent factory, and they’re then looking to who in the business or who do we hire to then manage this process and oversee exceptions or when the agent needs approval to kind of proceed and so on.

Scott Barker: I hear you on the. The feeling of being overwhelmed. I think a lot of folks are right now. It’s like, wow, I need to fundamentally revisit every part of my, my process, my business process. What are some of the use cases that you’re using in your day to day, or you see partners, customers using, that are maybe not as obvious that have had a big impact.

Inside Microsoft CoPilot: Real-world agent use cases across sales, support, and strategy.

Ray Smith: Yeah, so I’m obviously on the kind of engineering or product development side. So you, as you can imagine, all of us like GitHub copilot, we build a lot of rag based agents using entry 65 copilots. Every time we have a product strategy day, we’re bringing all our strategy papers. And yes, people may use AI to help them write these strategy papers.

but we pull all these strategy papers across different teams. We link them to. Use cases and epics in like ADO and we put all that connection into a single agent and we call it like, this is our way of planning for this semester. And then everyone gets access to that agent to see how it’s doing, tracking dependencies and interacting with an agent or setting up notifications if they want to be notified when some team moves something.

So it really does break down all those kinds of org silos, dependencies, and shared knowledge. We see lots of agents where it’s around interacting with customers. So we’ve got to know your customer agent. So pulling in usage metrics. So anytime I’m jumping on a call. I get a quick rundown, like it’s proactive, sends me a breakdown of that customer, the usage, you know, and it could be from case history, it could be from communications and so on.

So that’s kind of in my work, but that’s in the product development and engineering space. Frankly, what I see customers do blows my mind. I’d like, that’s some of them, like from. One customer pets at home, they built a fraud detection agent in less than a week and got it into production and it’s doing all these risk assessments going into Salesforce and CRM tools into various different tools to pull all this information together.

That would take like 20 times longer for a human to do it. So, but that allows them to change what is their risk assessment or scoring? How do they approach this? How do they do retail transactions? And they’re looking at many more other agents, but I’m seeing like Dow Chemical was another great example.

They fundamentally disrupted how they did this invoice processing that involved BPLs for. Checking for overcharging and reclaiming the overcharging. So they’ve looked at a lot of these processes and said, that can be an agent, that can be an agent. And it’s not just around efficiency. It actually can, you know, impact the top line or it can impact, um, like let’s say customer satisfaction because they’re getting a more immediate response where it would otherwise tick tack over emails or to be dependent on humans in the loop.

I’ve seen a lot of sales development ones, which is obviously relevant, relevant for your listeners. So. the auto sales development rep that does all the kind of lead qualification, brief company, brief, report building, pulling it all together into a scoring algorithm. Then deciding how that gets rooted?

Does it get rooted to an AI agent that offers support over email or some synchronous chat? Or does it then go to a human with all this information for a high touch engagement? So seeing lots and lots of use cases that are almost becoming commonplace now, but real kind of ROI, tangible benefits, very quickly compared to, you know, let’s say implementing enterprise applications a few years ago, it was like,we’ll know in a year or we’ll know in six months what the ROI is.

It’s much, much quicker to implement and much quicker to see the benefits.

Scott Barker: Yeah. Yeah. Those are cool use cases. It feels like you’re just limited by your. Your creativity at this point, which is, which is pretty cool.

Ray Smith: This is the joke I have. It’s like I, you know, I, I joke when I talk to customers, I’m like, I went from being in AI and in the startup world around 2010, everyone was saying they were in AI and I became quite cynical. And then I show a picture of me and it’s like, you know, Ted lasso pointing to the belief sign because mainly because there’s no, there’s the limits.

The limit is no longer tech and I, you know, I’ve been in the development space for decades now, and I’ve always had an engineer that’s kept me grounded. So I’d be the one dreaming of like, where is this going in the vision? And my head of engineering is like, Ray, get your head out of the clouds.

That’s not possible today. Whereas now it’s like the engineers are just going, yeah, just, just tell me what you need. Like, just give me the focus and we’ll go build it. And whether that’s a mix of building some deterministic stuff with non deterministic stuff with agents. Like it’s all on the table. So you’re limited by kind of our focus or our imagination, whichever way you look at it.

And it really just about, um, people thinking around how they transform their business and getting started. Like the worst thing people can do is, you know, spend all this time and spreadsheets and time analysis and like thinking around how the strategy might impact them. I’m like, pick up the tools, build an agent today.

You know, like I’ve seen people build agents where it was like, I’m sick of also having to order pizza for our kind of hackathons or our thing. So they like to build an agent that sees who’s accepted the meeting invite. Asks them for their preferences over teams, keeps a record of the personalization. And then once it says, okay, who’s arriving for this meeting, places an order with a pizza chain for the pizza to arrive 15 minutes beforehand.

And you’re like, that’s pretty awesome. You know, no one had to do that job anymore in the, in the team, for our boat bashes or, or kind of hackathons. It’s just applying the technology to a problem. So the use case is the, is the key.

Scott Barker: It’s fascinating when you let your head just think of the infinite possibilities, it almost feels like, laziness will be our friend over the next little while it’s like, you know, be lazy. Don’t, don’t do the thing. Just get the agent to do the thing.

Ray Smith: I think we have to extend our kind of thinking beyond what we would have maybe thought about automating before, you know, before we were in, like, I could probably build a program, a script, you know, a macro in Excel. I could do a little bit of this. I could offshore this to a team here that could do some BPO for me.

Now it’s just like understanding what it can do and trying to apply these new capabilities to that. So you’re right. Be lazy. That’s a key takeaway because as humans focus on the higher order, things like relationship building or the strategic quarterbacking, as we talked about, is where we’re going to have our biggest impact.

The most valuable skills in the AI era—from delegation to agent orchestration.

Scott Barker: Yeah, that leads me to another question, which is I think there’s so much uncertainty right now of like, you know, people just thinking like, what is my job function going to look like in, you know, not even five years, call it 18 months. in your eyes, if you’re a young person listening to this, or maybe you’re halfway through your career, latter half of your career, looking to upskill, what do you think some of the skills to thrive in this new agent world are going to be, I think you said, you know, focus on relationship building. I feel like delegation is a big one. thinking almost more like an entrepreneur. What are some of the things that personally people should be focused on? What do you think?

Ray Smith: Yeah, I think relationships are going to be key. And I think what we’re seeing is this shift from like, you know, the kind of the short term myopic bang in the table number, you know, hitting numbers to more of a longer term view of like value and. Kind of customers, referrals, and like all of that.

And that’s a great journey we’ve been on over the last probably five years to, to real kind of business outcomes and so on. But at the top of the call, you mentioned like everyone’s talking about agents, you know, and it’s like, it’s not just the CRO or the CIO or the technical people. It’s like, it’s coming from the CFO as well.

The CFO is like, Hey, we need, we’re hearing all these business cases now being public around meaningful transformation. How do we change, how we serve our customers, but also like at the dinner parties, you know, you know, it’s, something’s really permeated when you’re at a dinner party where no one who’s probably in tech and they’re asking is like, what should my little Johnny or little Mary doing in school or in college, should they learn to code?

Should they like it, and these are all great questions. And the short version is. It’s like, they just need to know how to leverage AI. And you talk to the delegation, I think in the future, we’re going to be like, Ray is great. Because he knows how to, he’s got some referenceable experiences with like five or six agents where he’s able to do X, Y, and Z.

And he’s a force multiplier for these things because he knows how to run and manage and you know, basically build these agents so leveraging AI and we see that in marketing. We see that in legal terms already. We see that in a lot of space it is where it’s really permeated where if you’re not leveraging any kind of AI or agents you’re kind of not doing it at a scale or efficiency as your competitor down the street to a certain extent.

So I think everyone needs to think like that and think around, just relationship building and long term kind of customer value. and just how do you tailor the solutions? Like that customer obsession is really, really key. Cause if you just like, Hey, another sale, I got the lead, I chalked on the wind.

But like, if you don’t go in there and make that agent or that solution successful and tailored for them. It’s just going to lead to churn and so on. So, you know, I think the future is going to be multidiscipline. So almost like a growth hacker had to be multidiscipline across lots of things around, like could kind of play with the tech to build scripts, the new marketing concepts, I think in this agentic world, people are going to have to be

A little bit of subject matter experts know the tech enough to be able to build these agents, even if it’s low code or abstracted away from code. and they’ve got to be able to kind of have a little bit of a growth hacking and customer obsessed mindset. So, pretty much all, all the, like, most highly sought after skills need to come together into this, but that’s really where it’s going to go.

And people who leverage. Just agents to kind of multiply their output are going to be hugely successful. And I’ve seen that already in the content creation space where people were like, my business has been disrupted. And I’m like, why not harness that? Be the editor in chief at scale, do it much more efficiently and effectively than the agency down the street.

And I’ve seen people like to see it as a challenge, but pivot and go, I’m going to leverage it. And I see that even with business process outsourcing. so organizations where this could be a fundamental disruption to them, but they’re like, let’s leverage the tech. Let’s see if we can do it more efficiently.

And maybe the customers will still want to just give it to us because they won’t want the headache.

Scott Barker: I think there’s a ton of great points there. One of my favorite quotes, I can’t remember who said it. Maybe it’s Naval, which says specialization is for insects.

And I think over the last, you know, decade, we did like hyper specialize within our organizations. And I think this will now return to the rise of the generalist who understands, you know, various aspects of the organization and can, and can effectively orchestrate these, these agents. Shifting gears a little bit, because we get, you know, a lot of earlier stage teams, you know, listening to this and I think, you know, you’re very fortunate at Microsoft to work with some of the largest, you know, enterprise companies in the world.

I would love, if you could just give some insight. The adoption curve you’re seeing at really large organizations. You mentioned Dow Chemicals, these are behemoths, and traditionally have been slower to adopt technology. Are you seeing that? Or is this one of those things that regardless of size of organization, this is everyone is thinking about this and, and trying to infuse it in everything.

Ray Smith: Yeah, I’ve, I’ve never been in a situation in my career where I haven’t felt like I’ve been evangelizing or like, you know, there’s always like, you know, back in the data hook days, when I, you know, first met Max, I felt like I was pounding the pavements, banging on doors, trying to like evangelize how predictive forecasting and pipeline management would really help transform businesses.

It was kind of pushing that boulder a little bit. I don’t feel like that’s the case anymore. In particular in this, agent world, everyone is either already investing, and maybe they’re investing in the wrong areas because they’re dabbling with like, you know, pro dev and trying to play with different frameworks and maybe fine tuning models, but they’re not really.

Focused on the use cases. I think about every business that I’ve seen, and I still engage with a lot of startups and SMB, and small, medium companies as well. Not just the big S 500 or big accounts that we serve at Microsoft, but I’m seeing the full range of people. The penny has dropped that this is happening now, and they need to figure out how they can leverage this technology.

So that is happening. And, obviously for all of them, probably more for the bigger ones they care about the security, the governance, the observability, like the platform of choice for all this future agentic disruption or transformation. so they’ll push us on all of those controls. And that’s obviously a strong differentiator we have at Microsoft.

Whereas you go to an SMB, I’m like, once this outcome is delivered and it runs each and every time, I’m like, they’re more fixated on just getting a solution in quick, getting it running and delivering for them. Um, and they’re less. Just a little bit less concerned around all of the, the kind of enterprise grade controls, you know, for a CISO and so on and so forth.

So that’s kind of what I’m seeing, but the key thing is, starting with focus in the early days when we gave early access to a number of customers of varying size, there was the same learnings. They were kind of starting with the blank page and I was like, here you can build an agent by just describing it.

And they would say, I want an agent to run my sales department. I’m like, no, that’s not really instructions. And you’re kind of like, you’ve got it in the same way. You would have to get very precise instructions when you mentioned offshore BPO. You have to treat agents like that. You see, you have to be very specific, unambiguous.

Well defined and not making them too kind of broad and too, so you start there and you get a use case and you kind of build on it from there, but there’s huge reception from these large organizations. To entertain anyone who will help them solve these problems. And I’m seeing like, particularly in the insurance space, lots of really successful startups going in going claims processing, very manual today.

We’ve got these agents that we’ve done down the street and A, B and C. Can we help you? their challenge will be, can they stay in those like big accounts when these agent platforms? Allow them to build their own agents. So they have to still remember to be specialized, offer that kind of customer value and the relationship that keeps them kind of sticky or embedded in those accounts.

But the transformation is already happening. It’s like, every organization is either already investing or already deploying production agents, to, to transform both their front office and back office processes.

Scott Barker: Yeah. And what, what, percentage of conversations do you have? ‘ cause and you talk about this all day long, where folks are looking at the landscape and like, oh wow, this means we could internally build all these different applications that are custom to our business. Or do you feel like there is a real opportunity still that people are looking for partners and they’re like, okay, well we could make X part.

More efficient in our business. Let’s go find a startup, someone to help us do this. Like, what is that? Cause I think there is a school of thought out there that, you know, at least large organizations, they’re just going to start building all of these things themselves internally.

Ray Smith: Yeah, I think the key comes down to kind of like the sponsor, the speeds, the ROI and skills probably like across all of those variables. So you can get into a large account with a key business sponsor who’s been tapped on the shoulder by the CFO saying, I need you to. Fundamentally hit this gross margin or change this process.

And I’ll maybe give you some CapEx, but I need OpEx to go here, or we’re doing it to drive cross sell. So like revenue growth, we need to embark on this project. And because it’s such a big strategic initiative. They were like, yeah, I’ll absolutely talk to a startup if they come in and show me that it works or I think the key thing is, is like, we know this is going to be such a big transformation of future that people are thinking around platforms and interoperability, security and governance, and that’s why they will start to Thanks forever.

Look more and more to, companies like Microsoft to say, Hey, we want to be able to make all of these people happy, across how we’re going to manage this new agentic, estate, like there’s going to be hundreds, if not thousands of agents across the typical business and we as humans will interact.

Maybe with the one assistant that taps into this pool of agents, but with there being that many people will push to saying, okay, maybe I want to choose a platform, but so there is a balance, but we will still see almost like the way we had point solutions in the past. We’ll still see point solutions or point agents coming in and we’ll interoperate with these bigger platforms as needed.

So it’s really just a case of getting to your sponsor, nailing the use case, quick kind of POC or time to value, and building that relationship is just, that’s just the tantamount importance, before let’s say all of these companies figure out, Hmm, maybe it’s not that hard. Maybe I can build it myself, but there’s still a lot in that.

It’s still a lot of skills and there’s still a lot of them. I think BPO firms will still have a healthy business, but they may just. Deliver solutions differently, not just through people, but through agents in parallel. So I think speed is going to be key. And yes, maybe in the fullness of time, people will have all these kinds of agentic skills in house to build agents on a platform.

but the change management and the skills part is the hardest one. It’s the hardest one for it to move fast. So that’s where it gives startups a real opportunity to, to kind of nail some pointed use cases and make hay while the sun shines.

Scott Barker: Totally. Yeah. I feel like you could almost make the same argument people are making now. With just like traditional software, you know, you could have built it yourself, but there’s speed, security, domain expertise, and deep knowledge of a very particular problem. So yeah, I’m with you. I don’t think we’re all of a sudden all going to build every application we need.

I have so many questions, but I do want to get to the final two questions. But before, you mentioned, we’re all going to have like thousands of agents in our businesses and I see that future too. It is safe to say then that the agents are going to be using our software applications more than we are.

And if you’re building software today, does it need all the buttons here and there, and the UI and the UX that humans are used to? if an agent is going to be predominantly interacting with it.

Ray Smith: Yeah. And then maybe if you’ve rolled that thought forward even further, Scott is like, if it doesn’t need the UI and like the traditional forms of a data app experience, and it’s just basically a database with some business logic. Well then maybe it’s an agent and you build an agent that plugs into an ecosystem that is then invoked by another agent.

And particularly when we start to see dynamic UI and canvases emerge. Where, you know, let’s say you’ve got this big cumbersome form of all the information to support like 40 processes all into, say, one contact or opportunity form. in this new world that we just, it’ll only be dynamically rendered for what you need at that point in time or what’s missing and so on and so forth.

So I think you’re spot on, but I think this is where agents calling agents, and, you know, as these, you start to stack these agents, you know, it’s like, hey, maybe have a, A data access layer agent that interacts with these systems of record. Then I build an agent to handle, say some business logic to do some research or, um, analysis or lead scoring or whatever it may be.

Then I start to build more agents on top of that, ultimately too, you have a customer or brand agent that you expose to your customers that then is set up to call on some of these other internal agents. To serve their needs, whether that’s pre sales, sales, post sales and support. and that’s how you can imagine all these building blocks starting to accrue to a fundamental transformation.

Scott Barker: So cool. Well, it’s an exciting time to be alive, an exciting time to invest, an exciting time to build companies. And I have a final 2 questions. We’ll just rapid fire and they are intentionally vague. So you can take them any way you want, but the 1st question is what is the 1 widely held belief that revenue leaders believe to be true that you think is bullshit or no longer serving us?

Why agentic AI makes customer relationships and long-term value even more critical?

Ray Smith: Well, I’m not sure if it’s being consumed already, but you know, back when I was slinging my wares, selling software back in 2013, 2014 in San Francisco, the belief was it was all around being kind of short sighted on your numbers. So you’re like how many dials into leads into qualified prospects and, you know, first appointments.

Getting through those and getting them on the board. and yes, obviously the top of the funnel still matters. And yes, your first 10 customers and love bombing them to make them successful still matters. But I do think there’s been a shift towards how do you get sustainable revenue growth? And that comes from like, Really understanding your customers, building that trust and delivering those measurable outcomes.

And I think that’s going to be even more so in this agentic world, where the ACV will be big. So you’ll get like a healthy kind of gross margin, you know, even in a very, good position after 20 customers, but not letting the big valuation or because you’ve hit like, say 80 million ARR or super fast, that that kind of pushes you too fast because you need need to scale that sustainable growth and that’s either by continuing to build that beachhead or moving into another use case before you kind of, because the biggest challenge you have is, is overselling and under delivering where. You don’t meet the customer’s needs, they churn, and then there’s just that kind of diminishing brand reputation.

So for me, I think that’s the biggest shift I saw before it was all, you know, pound the phones, you know, you know, almost like the spam and machines that we had previous decades, whereas it’s just shifting to more, I don’t know, sustainable growth model, a more relationship focus, which I think is going to be key for agents.

Scott Barker: Yeah, I mean, it’s a great shift and, you know, focus back on, you know, customer centricity and, you know, time to value and really proving again and again that you can drive these, these great valued outcomes. I like it. Final question. I call it the silver bullet question. You know, we all know there’s unfortunately no silver bullets, but we try and talk about them anyway.

What is one, go-to-market tactic or strategy that is working today, for Microsoft or any of the companies that you’re partnering with?

Ray Smith: Yeah, I remember, years ago when I kind of connected with Max, it was all about sales hacks. What’s the latest sales hack? You know, it’s like, what’s the latest trick? It’s like, hey, dial in on leads that are in this, you know, in this state after they just won some big match and like, Everything was about like a hack or a tactic.

I feel less so because of what we talked around, which is that kind of longer term view, tailored solutions, industry expertise, all the hard stuff, relationship building stuff. So I do feel like the only tactic and not to just bring it back to agents, but I do think using agents as that multiplier effect for doing all the mundane stuff to give you more time into the relationship building.

So like where you can offload to an agent to be that multiplier effect for you. Like, it’s like we’ve always had these situations where we’re like, Oh, I’d love more resources. If I had more resources, I’d hit my number. I will do more. Well, now we’ve got this capability where you say like, well, you can have agents that can almost have infinite scale for all these things, and that’s going to free you up or your small team.

To go focus on the real things that will matter for that kind of thing to deliver growth and revenue ultimately. So I don’t think there’s a tactic per se, other than just say leverage agents to help you scale.

Scott Barker: I like it. Yeah, we’re officially in the age of no more. Sales hacks, it’s all the hard stuff now go build, you know, if you’re going to spend time, you know. Trying to figure out a hack, like. Go learn agents, go focus on your domain expertise. Like I feel like those are the things that, that matter now, um, more than, more than ever.

Final, final question, even though I said that was the final one, you’re really at the forefront agents. Where do you kind of get your information from? Or, or maybe put a different way. If I may. Go-to-market leader. I’m listening to this right now. This all sounds great, but it’s very overwhelming.

Are there any courses, newsletters, podcasts you would recommend that are great building blocks as people try and figure out this new world?

Ray Smith: Yeah, so we, we created a kind of a team channel across people where we share blogs and we share content and then we are able to point an agent on it to kind of query it because sometimes you see a blog or a podcast or something and you’re like I’m not quite ready to digest what that topic is.

But then later on, I want to be able to query it and say, so being able to keep up. With the AI trends is a key thing we all have to do, quite frankly, not just those who are building agents, just understand the opportunities. So, honestly, there’s a long list and half the time I’m not even sure the source it’s as in, like, that was put into this kind of team’s channel.

But yeah, it’s just finding the time. to research, to read up about a particular topic is really important. I don’t, I don’t think anyone gets to escape from being an AI expert. even the people who say I’m not an AI, you’re like, well, everyone’s going to need to be an AI expert in the same way we were like, Hey, do you know how to use Excel or PowerPoint?

It’s like, you kind of, you know, we went through a couple of decades where no one could use that as an excuse. And I just don’t know how to use those tools. I think the same will be true for. AI capabilities and agents specifically. So, yeah, too many sources, to even mention, to even list on this one.

Scott Barker: I love how your brain has already made this shift in this leap and I can feel it in your answer. You just answered, where should I go to learn about AI agents? And you answered by using an AI agent to digest all of the information and bring it back. So that’s the way we all have to start thinking.

I love it. Well, Ray, I really appreciate the time. I personally really enjoyed that one. I could have asked you 748 more more questions, but I really appreciate you jumping on and sharing your deep knowledge and expertise. Well, we’ll have to have you on again for sure, because, I think we’re just.

Everything’s happening so quickly. We could have this discussion again in 18 months and it will probably be vastly different.

Ray Smith: Yeah. That’s the joke I have at the moment. When people say, Hey, you’re speaking at the conference. Can we get what you’re going to speak about? And it’s in nine months, who knows what the topic will be, because everything will have moved along. It’s all moving so fast, but it was a real pleasure, Scott.

As always, the great work that yourself, Max, and team are doing. It’s a big fan, so happy to help out at any time.

Scott Barker: Awesome. Appreciate you. And to our listeners, thanks for rocking with us. I say it every week. Listening’s one thing, executing something totally different. Hopefully we inspired you to go and leverage more agents in your business and we’ll see you all next week.

The post GTM 139: AI Agents Are Changing Everything — Microsoft’s VP of AI Agents on the New Era of Work and Software | Ray Smith appeared first on GTMnow.