Not All AI Agents Are Created Equal: A Playbook for Hiring AI Teammates in Revenue Orgs

Julia Nimchinski:
And next up. We’re joined by Andrea Tucker, VP of Product Marketing, and Paul Selby, Director of Product Marketing at Outreach. What a pleasure, super excited for this session. They will demonstrate not all agents are created equal. Welcome to the show, and how are you doing?
Andrea Tucker:
Thank you so much, Julia. I’m doing well. very excited to be here, and, you know, if you’re unaware of who Outreach is, we’re an Agentic AI revenue platform that is purpose-built for revenue teams. So… Are we ready to go? You ready for us to go? Please meet.
Julia Nimchinski:
Sure, Selby?
Andrea Tucker:
Oh, okay, great. Paul Selby is my Director of Product Marketing on my team as well. So, we’re really excited to be here. Before we dig in, I just want to put a couple things on the table. You know, everybody’s selling AI Agents right now. This is not something new. I know that you’ve been hearing about this all day, right?
And… It’s hard to tell what is actually real. everybody uses the same slides, they have the same promise, but they’re really completely different underneath. And here’s the filter that we use when we talk about agents, and I just kind of want to set the table with this so that people understand where we’re coming from.
If you were hiring a person onto your revenue team, you’d ask 3 things before you made that offer. What can they actually do What are they allowed to touch in your organization? And who do they answer to? And so many people don’t even take the time to ask the questions about the agents that they’re looking at.
they just ask what model it’s built on, right? And… Honestly, that’s the wrong question. that’s a big part of the reason why so many AI pilots have quietly died over this last couple years. So, that’s what we’re going to walk through today. We’re going to talk about our version and our vision of what has changed.
and what it looks like to hire capacity instead of licensing it, and then Paul’s gonna talk a little bit about why it matters, our architecture, the agents, the customer proof, and how you can actually get started without boiling the ocean. If you’re good, we’ll go ahead and go to the next slide. Awesome.
Alright, so… Let me start with just saying we’re hearing from just about every revenue leader right now, and just say in the chat if this is the same in your world. Gartner surveyed B2B buyers and found that 74% of the buying teams run into what they like to call unhealthy conflict before they decide anything.
So they have conflicting objectives, disagreements on which path they want to take forward, or somebody outside the group is overruling the whole thing at the very last minute. So, when a deal stalls, it’s usually not the rep losing an argument, it’s a rep trying to referee one that they were never even in the room for.
And this causes a lot of… a lot of stress. Now, second. this second number, the same firm, 70% of sellers say that they’re overwhelmed by the sheer number of technologies they’re expected to use. And that’s from Gartner’s Sellers Skills Survey. And it was, golly, just about over a thousand B2B sellers.
And in that same study, 72% said they were overwhelmed by the number of skills that that job now demands. And last, but definitely not least, 69% of workers say that their company has taken some action on agents, but fewer than 10% can even explain or define what an AI agent is in their own words.
So… this whole mandate that we’ve got across technology is to go compete in an AI-driven market, and the workforce simply isn’t there yet. Now. One of the things that I’d like people to consider… we typically trade all of these as people problems. We need to train the reps better, we need to coach them harder, we need to hire differently.
But look at what this data actually describes. A buying process that’s too complicated for one person to track. A tool stack that’s too fragmented for one person to reconcile. And capability bar that’s rising faster than any training calendar can move.
And right now, the reps are the integration layer, and that’s a very, very expensive way to do business. And so you’re probably saying, well, how’d we get here? Well, for about 30 years, our model was pretty simple. You rented seats and software, then you hired people to use it.
we had CRM, or have CRM to track deals, sequences and campaigns that help us to reach buyers. conversational intelligence to figure out what happened in a meeting, and even forecasting so that the leaders know what numbers that they need to call, right? But look at what those all have in common, they’re reactive.
Every one of them is great at telling a human something, but not one of them does anything. You ask, get answers, and then some person comes and carries that answer across four other systems, and then turns it into actual work. The intelligence has never been the bottleneck. We’ve got all kinds of intelligence.
It’s the stitching that has been the bottleneck. And what’s different now is that software started doing pieces of the work itself, not just advising on the work, actually doing it. And that changes the math on capacity. So, what we’ve seen is that… selling had to evolve, and AI did too. So let’s talk about, you know, kind of what that looks like.
In legacy SaaS, humans do all the work. This is what we’ve always known, how we’ve always seen it. But in the second era, the co-pilot era, there’s a human at the helm, and this is genuinely useful. But it hands you an insight, and then it waits. The AI gets the answer, but you still do the job, right, at the end of the day.
In Agentic, this is where we see AI that acts on your behalf, but also inside of guardrails that you set. I… I want to make this jump from assistant to Agentic really concrete, because this is where most evaluations go sideways when people are talking about agents and thinking about agents.
So… Think about what science fiction promised us, because we are living in it right now, right? Decades ago, a screenwriter needed a starship, right? To make this seem believable. Somebody would walk into a room, and they would talk to a computer in plain English, and it understood. It remembered the context, and it said what to do next, right?
And most of the people on this call probably did a version of that this morning, where you got on a call, you… something transcribed it, and you summarized it and pulled the action items. We are living in an age of conversational intelligence that is no longer fiction.
But notice which half of the fantasy we are actually on, because that ship had two different kinds of intelligence in it. There was the computer. You ask a question, you get the answer, it’s encyclopedic, it’s instant, completely passive, right? But nobody… thinks about the other one, which is data.
Data was an officer with a post, real skills, reasoning, and he answered to a chain of command. Nobody on those ships ever mixed those two up. The computer was the tool But data was the crew. And Paul, are you a data fan?
Paul Selby:
Well, of course. Grains and Braun, come on, I mean, what a guy, right? He was great.
Andrea Tucker:
Yeah, me too. But we won’t keep going into that right now. But, so we built the computer. Most of what’s getting marketed as AI Agents this year is the computer, and maybe it’s got a better microphone, right? The crew part is what’s actually new, and it’s the part that forces you to answer those three hiring questions that we started with earlier.
And there’s more to consider here. Like, why is this so important right now? We know that trust and security is a huge issue. I mean, you’ve seen it in the news everywhere, right? It’s the gate. now, to what it’s going to take for us to be able to truly leverage this kind of AI.
The best agent in the world will not ship if your security team can’t explain exactly what it’s allowed to touch and track it all. Two. We’re seeing a lot of point solutions that are starting to lose to platforms that unify revenue context, and that’s imperative, and Paul’s gonna talk about that in just a minute as well.
Agents are only as good as the context that they get to. And a tool that only sees one slice of the insight? it’s gonna give you very confident nonsense. I mean, I think we’ve all had some of that experience with AI ourselves, especially with LLMs.
And three, you know, agents that learn from your data keep getting better on your data, and it compounds that experience, and that knowledge, and that understanding. So the question has never really been whether you should do this or not, it’s whether you do it before or after your competitors will do it.
So… The next generation of companies is going to hire agents. So, Paul, you see what I did there?
Paul Selby:
I did, I did.
Andrea Tucker:
Just checking. And this means adding a connected layer of intelligence to the revenue workflows that you already are using. So, agents that can read the data, they can take action, and they can keep getting better at it. So, you’ll want to think of it as capacity you can add, but you’re not adding headcount, right?
So, agents research accounts, they can prep your sellers before meetings, they can flag risk, update deals. Recommend a next step, and even coach in the moment, and give leaders a straight answer about. Really, what’s going on out there with their forecasting, right? I think we need to go ahead and go forward there. Can we go forward?
There’s a build there. And that’s a really different purchase than just a seat. You’re not buying a place for a human to log in and do work, you’re adding a teammate that does the work for you. And once you start talking about software as something you hire, the whole vocabulary, the whole way that we talk about this completely shifts.
Because you start asking, what are the skills that the team actually needs, and who or what, actually has them to be able to help our sellers get those skills. And that takes us from seat to skills. Now. For 30 years, we’ve known that enterprise software ran on one unit of value, and that was just to procure that seat, right?
you license it, you assign it, you hire somebody to go in and actually sit in that seat. And… Agentic AI changes everything about this, right? It moves from seats to skills, and you hire for skills the way that you would… let’s just keep on the same example we were using before, like you would crew a ship.
You don’t have… you don’t hire the same officers for the same… for the same role, and on your ship, right? So nobody staffs 5 identical officers.
What they’re looking at here is different areas that are purpose-built for these potential needs that you have, like account and prospect research, like meeting prep, like sales play execution, meeting follow-up, deal updates, forecast predictions, all of that.
And that task automation is the small story, but there’s a much bigger one here that I think’s really important, and that is… The performance curve for your sellers. Agents can learn from your organization. They can find the patterns that actually close your business, and then they can make those patterns available to everybody else.
So, not as an enablement PDF that nobody ever opens, because that happens all the time. I don’t know if you guys have trouble with enablement and being over-enabled like we do, but it’s one of the things that sellers struggle with the most, right? But this happens right in the flow of the work. So, watch what that does to the rep performance curve.
The top will barely move, your best reps, the ones that you know, can always… you can always count on. They’re always going to bring in the right deals at the right time.
they’re already great, but the bottom on this comes up, and that’s the whole game, because that’s where most of your headcount actually lives, not in your top echelon of your sellers, but the middle, right? And… A few points across the middle of your team is worth a heck of a lot more than another hero quarter from your top three.
So your people still keep the judgment, the relationships, they still focus on their strategy parts. But the agents would take off that repetitive, data-heavy execution stuff. And Paul, I know that you’ve got some proof to talk through on this a bit later. Don’t spoil it or anything, but tell them, is this curve actually accurate?
Paul Selby:
Yes, it’s accurate, and it’s in production.
Andrea Tucker:
Awesome. Alright. So here’s how it all comes together with how we do it at Outreach, right? You hire an agent, it shows up with these great pre-built skills, packaged workflows, it can run end-to-end. it’s not just a blank canvas and us handing it over to someone and saying, hey, there you go, good luck, have fun.
But it’s a really pre-built, thoughtful way to be able to engage in these different workflows. You buy capacity and credits, so your cost tracks the work that gets done. It’s not just about a seat. more skills, more workflows, an agent can own end-to-end, and then you get more capacity and more productivity per seller.
So, the teams add skills, or they build their own agents. The agents keep learning from what’s working across that different org, and every single action then runs inside enterprise trust, security, and governance guardrails. Now, remember, we talked about that just a minute ago, and that last part is in a footnote.
It’s actually really, really important. That’s the whole difference between a teammate and a liability in your company. So, to close the loop on my earlier point, science fiction, it really gave us both. It gave us data, but it also gave us hell, for those of you who might remember.
Hal was an agent with a clear goal, he had real capability, no human in the loop, and unfortunately, no audit log. So one of those you deploy, a crew member like Data, and the other one you write incident reports about.
And if you’re too young to possibly remember Hal, you might want to look up the movie 2001 A Space Odyssey, because, Yeah, that was more my generation, I’m thinking. Kind of… kind of silly, but there you go. Which is exactly why it’s time for me to hand this over to Paul. So, Paul, over to you.

Paul Selby:
I get to take over after the comment about the murder bot. That’s really cool. So thanks, Andrea. Alright, so that’s the pitch everyone’s hearing and getting pitched about right now. So AI Agents, Autonomous execution, it really sounds great on a slide.
Andrea opened with a stat about, really around how sellers are having a hard time around AI, and, let me ask you, Andrea, when you’re in front of a revenue leader, is the catch the technology, or is it the people?
Andrea Tucker:
You know, honestly, it’s neither one on its own, the catch is that most organizations, they bolt on a process that was never designed for. And they’re supposed… they’re surprised when AI has nothing really real to act on, because they don’t have their data set up right. It’s not a smart model problem, it’s a plumbing problem, so to speak.
So, not having the right context.
Paul Selby:
Right, right. And that’s exactly where I want to go next. Because the catch really isn’t whether the AI is smart, it’s whether it has anything to actually act on. So let’s continue here. So, here’s the uncomfortable truth. Every revenue organization is drowning in signals.
It’s got emails, it’s got meetings, it’s got CRM updates, call transcripts, buyer intent, product usage, the list goes on and on. Nobody’s short on data anymore. The problem was never collecting all those signals. The problem is actually turning it into coordinated action. And that’s where the architecture really matters.
Revenue orgs are dynamic systems. Every interaction, every deal update, every external signal creates context. And that context should shape what needs to happen next. Without that shared context, agents are just isolated automation.
They can summarize something, they can recommend something, they can trigger something, but at the end of the day, they’re still disconnected. They don’t know what the other agent did just 5 minutes ago. With shared context, like revenue context, agents become more coordinated.
They can understand the account, the opportunity, the buyer, the history, and really what that next best action is going to be. That combination of data, context, memory, and orchestration, that’s what’s turning signals into action. So, let’s take a look at the architecture behind the scenes that can put that into practice.
Now, I gotta admit, this is a busy slide. This is the busiest slide you’re gonna see today, alright? I’m gonna fly through it. So take your screenshots now if you want, because we’re not gonna read every word here.
Andrea Tucker:
We love our staff slides.
Paul Selby:
Yeah, well, bear with me. So at the very bottom, at the foundation, is our revenue data. Everything is there. First-party product usage data, customer insights, CRM, emails, calls, meetings, all of the stuff. Including public market data. The next layer up is that persistent revenue context that I introduced in the prior slide.
This is the part people skip past too fast. We’re not just piping raw data into a model.
We’re building reusable context, things like a context graph, semantic search, memory, all these technologies that you don’t need to know the names for, but they’re building this structure around accounts, buyers, opportunities, and that institutional knowledge that becomes so critical.
This is actually what lets an agent remember what happened last quarter. Next layer up, we’ve got the governed agent execution layer. This layer decides what agents can access, what they’re allowed to do, where a human has to be in the loop, and how execution is actually controlled.
For you technical folks out there, you’d think of this maybe as the policy layer. Then we have the Agents and Interfaces. Outreach Omni, Outreach Agents, third-party Agents, these are all brought into the flow of work via channels like web, mobile, Slack, Teams, email, and calendar.
And then at the top, where the work is being done, those are our revenue workflows, right? The point about all that is it isn’t just one layer, it’s that trusted AI execution that requires the full stack. Data, context, government execution, agents, workflows, they all come together.
And that’s what separates coordinated revenue execution from a disconnected AI feature bolted onto a dashboard, like Andrew was talking about. So, what is that architecture power?
So this is our universe in outreach of the agents that can show up across the customer lifecycle, doing things like generating pipeline, managing deals, retaining, and even expanding accounts. And the important distinction here is that these agents don’t just assist.
They act, either autonomously, or with human in the loop, based on the governance controls you are setting. That’s not a minor detail. That’s the whole difference between a co-pilot and an actual teammate. You’ll notice our product, Omni, our conversational AI, it sits across all of it.
That’s that conversational agent that works across every other agent. One interface into the whole system, into all the insights that are out there. Looking specifically at how some of these agents can be operationalized that we have in the outreach platform.
If you wanted to look to generate some pipeline, that’s where we would turn to Research Agent. It gives instant access into the insights without a lot of manual digging, so your reps can engage the right accounts and prospects with much more confidence and context.
The Revenue Agent works to identify high-intent buyers, source new existing contacts, and initiate the timely engagement that you need so your team is never leaving money on the table.
The personalization agent takes all of that rich data that we talked about, all of that context, everything that’s rolling around inside the platform, and makes that available to use to draft relevant and personalized messages across every channel and at every part of the life cycle. So, emails, LinkedIn conversations.
And even call steps with custom scripts and voicemails. And if you’re managing deals, the meeting prep agent makes sure reps never walk into a meeting unprepared. It provides a custom brief that pulls together relevant attendee context, account history, open opportunities, and even the recommended talking points.
Deal agent, that takes the context from your customer conversations, so those calls and meetings, and recommends or automates updates to your critical opportunity fields, like MedPick, Next Steps, and provides custom AI deal summaries. We talked a little bit earlier about retention and expansion.
That’s where a Revenue Agent can come in, perform some personalization, and drive those renewals and cross-sells and upsell motions, again, based on that rich context.
And if you need something really specific to a seller, that’s where we offer Agent Studio, where you can actually orchestrate these different sales motions and customize agents to specific seller needs. You can configure triggers and automated actions across those agents within that full revenue cycle from a single visual canvas.
Andrea Tucker:
And we have customers using all of these now.
Paul Selby:

We do, we do. So, to drive the point home, we really want to mention 3 things that set outreach apart. When people ask, it comes down to these three things. First of all, Outreach AI acts. It doesn’t just analyze. Most AI in the space is telling you what’s wrong or what to do next. You, right? Our agents actually do that work for you. They execute.
They enroll the prospects, send the follow-up, trigger the play, update the CRM. That’s the line between recommendation engine and a true teammate. Next up is scale. Outreach knows your business. Again, callback here to all of that context that we have. We are not general-purpose AI bolted onto CRM.
We’ve built years of proprietary revenue data, first-party engagement signals. pulling information out of your CRM, enriching things with third-party information, your specific playbooks, all of that. Our agents know what good looks like for your company specifically. And then finally, control.
This is the one that matters the most to our security-conscious people in the room, and that’s… there’s no shame there, right? That’s an important element to doing business in the 21st century. Every agent action is logged, explainable, and governed by controls you set. You can see exactly what an agent did and why. And override it at any time.
So, Andrea, you know, this ties back to what you said earlier about that trust thing. Trust is becoming the gating factor for adoption. What are you hearing from prospects when trust and security come up?
Andrea Tucker:
Well, I guess the same three questions every time. What data can it actually see? What can it do without asking me if it can do it first? And can I prove, and this is very important, after the fact, exactly what it did and why? And then, of course, legal always has something in there, which is, who is actually accountable when it acts?
And if you can’t answer all four of those in very plain language, you’ll end up dealing with the security review.
Paul Selby:
That’s exactly why control is in a checkbox for us. It’s core to our architecture. And to that point, we hold many different certifications, ISO42001, ISO 27001, SOC 2 Type 2, GDPR, CCPA, and HIPAA as well.
Andrea Tucker:
All the acronyms.
Paul Selby:
All the letters and numbers, yes. But don’t just take my word on all this. I’m not gonna… I know I have another busy slide here, I promise you only one, but I’m not gonna go through all these logos. Only 3 are really worth calling out for our purposes today. Our customer Siemens standardized forecasting across 190 countries.
That’s the clearest proof point for what operating at scale really looks like. Freshworks saw an 87% increase in close-one revenue attributed or influenced by AI. There’s your strong business impact number for AI. And then ServiceNow saw a 330% increase in rep prospects contacted. There’s your productivity number. So 3 different categories, right?
Scale, revenue, productivity, same underlying pattern. It’s never one feature driving the metric. When data, context, and execution come together, the impact shows up across the full revenue life cycle. But none of that happens by accident, it requires a model to get there.
Which is why… we like to ask people, where does your org sit today in AI adoption use? We have an AI maturity model that identifies four stages. Traditional, which, you know, is reactive execution, mostly manual. Connected. So you’ve got partial automation going on, but data trust is still fairly low.
Consolidated, so think connected workflows, think reliable visibility, and then finally, AI efficient, with predictive, scalable, planning and execution against your revenue engine. Andrea, based on the conversations you’re having, where are the most prospects placing themselves on this spectrum?
Andrea Tucker:
You know, it’s really interesting, Paul. Almost everybody puts themselves at Connected, which, they’ve got some automation of some sort running, right? But they don’t fully trust their own data yet. They want to be consolidated, but that’s actually what’s blocking them. It’s not ambition, it’s not budget, it’s data trust.
Paul Selby:
Gotta keep the house clean, gotta keep the house clean. So I’ll give you one more second to scan the QR code, to learn more details about our maturity model. And from there then, let’s talk about how you get started. How do we make things real?
So, we have a process for this, and the way it works is you pick one use case, either from our recommended list, or propose one of your own. During a one-month pilot, you get unlimited AI credits and full access to all of our agents, with credit usage included while you evaluate.
It’s a fast and focused evaluation, and it proves that impact very, very quickly. It’s backed by our hands-on help from professional services. They will support you in the configuration, launch, and evaluation every step of the way.
Most of our pilots run through outbound prospecting, and they combine the appropriate agents that we talked about earlier. A couple of the proven starting plays are something like a vertical play, where you do focused outreach on some key industries that you want to land.
A persona play, so tailoring that outreach to specific buyers, a product-specific play, where you want to promote a specific offering, or even a compete play, where you’re targeting accounts that are based on competitor stacks that you have knowledge of. Each one of those…
Andrea Tucker:
It’s all a very reasonable price, too, to do this. I mean, it’s not something that’s gonna break me.
Paul Selby:
Absolutely, yep. And each one of those pilots that you would choose from for your use case, they’re designed to help you move fast, test real impact, and build confidence before, you know, scaling all kinds of agents across your broader GTM motion.
So hopefully that sounds interesting, and you caught the QR code earlier so that you can measure where you are and figure out what your next steps are.
So, alright, we’ve covered a lot, I know that, we blew through a lot, but it was everything from why now, what it takes architecturally, to how you can get started with all this, and I hope you’ll definitely take us up on our offer.
Julia Nimchinski:
Thank you so much, Andrea and Paul. Phenomenal presentation. And yeah, Paul, I got you off.
Paul Selby:
Perfect. No, you’re fine. I was just gonna say, thank you, everyone.
Julia Nimchinski:
Thank you.
Andrea Tucker:
Thanks very much.
Paul Selby:
Thank you.