Proactive AI Agents: The Real Business Upgrade

While he gets that pulled up. My name is Jonathan Hessing. This is Diego Herrera. We've been best friends since we were 12 years old.

Started this company. Very complimentary. And I actually put my presentation together. We were thinking about it last night.

And we have four agents that we built in our business. And as we're talking through this, so I had I had our agent, I tked it you asked about agentic workflows. I told our marketing agent, I said, "I want you to pull all of our social media data. I want you to pull meeting transcripts over the last three weeks and what moved down the pipeline, take the most resonating messaging, and that's what I want to do my presentation on." And so, not only did it come up with the copy in the presentation for tonight, but it actually designed the site.

And and so, as I'm listening, right, we talk about the human and the AI. The AI did that, but the human is sitting here calibrating to the conversation. And the conversation is pragmatic business, right? It's AI theory.

It's where is AI going? So, we're going to just come up here and make it as practical and turn theory into practice as possible and talk about what's been our experience, right? How did we get a phone call from our CPA client the other day saying that we just saved them during tax season with quantifiable metrics, right? How did we help a health care company boost adherence for one of their one of their things that they sell to patients which will net in more revenue, right?

We can talk about ROI, but if it remains ROI theory, it's kind of hard to grasp on. So, we're going to go through this. We're going to talk through this. I'm going to skip anything that could be redundant.

And if we have time at the end, I brought my laptop. I'll plug it in and I'll talk to my AI agents and show you what that looks like live because it is different doing that. And so Diego, if you want to add anything or if you just want to hop in at the security part, >> we'll walk in there. So, I think a lot of people are still using chat bots, right?

Everyone uses chat GBT. Who's running an actual agent? And I don't mean in cloud. I don't mean talking to chat GBT.

Who's running agents? Right? And that, okay, so that's where the drop off is, right? And that is react reactive cha chat bots versus proactive agents, right?

The guys over here were talking about, right, the AI and how you want to control it and it doesn't control you. I agree with that. I think I would add on to that though. When you make really good AI, right now you're in the pilot seat and it's co-pilot.

I think if you can build it right and you can train it on your north star, this is where it becomes agentic. It should be able to be a pilot and you're the co-pilot. And I'm not talking human in the loop approval. I'm talking it knows the strategy.

It's sending you daily reports. It's always on, always 24/7. So that's what we're going to talk about today is the difference between reactive chat bots and proactive agents. And the most important thing is int intelligent sovereignty, right?

You can't build agents if you're locked into clo cloud or you're locked into chat, right? Chat GPT 5.5 is the newest model and it beats cloud in benchmarks, but people are have their whole ecosystem built into cloud chat. You can't build agents off of that, right? Daniel Merrill was talking about that too.

And you you look at this and the calculus on buy verse build is changing for the first time in in the history of software. Only enterprises used to be able to build their own software and own their own tech stack. That's changing now like the guy up here was talking about and they were talking about I'm going to build my CRM. Well, it may not make sense to build your CRM.

Maybe you want to use HubSpot or something. That's fine. That's like the tool layer. What you need to own is your intelligence because if you don't own your intelligence, you're not collecting data.

You can't replace it with an open- source model and get rid of API costs. You there's a lot you're very limited. So, and to make proactive AI, it's got to be trained on data, structured, unstructured, it's got to build agentic workflows, you cannot be confined into these platforms. So, a lot of people we talked to have spent money on AI and it did nothing.

I mean, they go through subscriptions. Like, it's like like a pro baseball player goes through bats during batting practice. I mean, it is absurd that they do this. And look, the models are free.

Like, you can get the models. You can use the AI. Making them run your business, the expensive part. I'm not going to double down here.

This whole tonight has pretty much been about this and but this is the most important thing. So, just to double click on what everyone has talked about. This is what 90% of companies get stuck at is that they're not leveraging AI in a real material way. And everyone on the business side, the best way I'll just kind of conceptualize this is everyone wants to do AI, but they do it ass backwards.

They start with step three, which is AI enablement. You got to do the right three-step process, which is get in there. Step one, KPIs, quarterly objectives, bottlenecks, understand because if it's not measurable, like it's a science experiment. It's not a business strategy.

Start with that. Understand what are the existing systems? What workflows do you have? What does the data look like?

Then once you understand that, you can come up with the right AI enablement strategy. So again, the two types of AI is choose one, right? And and so they used to be like people would be like put all their prompts on X and you would see like these big illustrious prompts that are like here's how you make the perfect image or video and that's cool and there's a place for it but that's not where AI is at now. That's not what the companies who are winning are doing.

What the companies who are winning using proactive AI, they're texting their AI in Slack, in Telegram, and they're texting it like it's an executive. Here's an example. I texted one of our marketing guys the other day and I said, "Hey we've got this newsletter. We're going out.

You guys should subscribe. It's badass." Like, it's not like sell our newsletter. It's like here's some cool AI stuff. But I was like, every week I'm having all these meetings and people are on our calendar.

Are those people being added into our email marketing tool? Before he could reply, I thought maybe I should ask the same question to Atlas, our AI CMO that I talked to through Slack. I asked Atlas the question. I said, "Can you see that?

And if if not, can you add them? Here's the agentic workflow." It looked at Clavio, which is what we use for email marketing. It saw that it was not updated every week with our new contacts. It then queried my email with read access only.

Don't write access. That's how you don't nuke your inbox. Read access only to our email. My email, it read access to my calendar.

Took all those new contacts and made an Excel sheet for my marketer to upload to Claio because it doesn't have right access to Clavio. Does that make sense? Right. And so that all happened before my marketing guy even replied.

That's real ROI. That's time saved. And to the to the guy that was up here talking about ROI, when you think about that, there's measured ROI, which is time saved. But if they're if you if your team is 1099 people, time saved, more impact results in billable immediate, you could tell.

But if they're W2, are they shopping on Amazon more? So the real area to focus on is modeled ROI, right? Because of all that time saved, can I reactivate a dormant customer list? Can I think about new marketing strategies?

Can I focus more on R&D? Can I be more proactive than reactive with clients? That's where the real value of AI is. It's not in saving time.

It's not in the measured ROI. It's in the modeled ROI of what you can do with the immediate effect. It's that step two. And that's what kind of this proactive versus reactive thing is.

And it's also getting reports like every I have a good bit of my team here and they're constantly throughout the week getting reports. Whether it's task management, whether it's sales whatever it is, it's constantly analyzing information or it's like, hey, it will message me without me prompting it to. It'll message me and be like, hey, this lead is going cold. You had a hot conversation with them a week and a half ago.

Here's a potential message you could send to them. I'd recommend sending that in the next four hours. Like, it is not waiting for me to talk to it. And just like a that is what a good employee does, right?

We had a we sold a client a big B2B e-commerce client and he asked how do I know how to use this. I was like you'll know because just like a good employee it's not sitting idle waiting for your instruction. It's saying hey within the scope of what I'm here to do I've analyzed all the data. I've looked at all the patterns and I'm coming to you and saying hey should I do this?

Yeah please do it now. Like and then it does it. And that's what proactive versus reactive AI is. It's not talking to chat GPT.

It's not talking to Claude. It's the AI living in your Slack channel, living in your Telegram, living in your iMessage, and it's talking to you as if you're talking to an executive. It really it feels different. It hits different.

It's like this omnipresent like ambient like I'm not going to go crazy right there, but it really like it's it's a different feeling. I think that's the best way to summar right one and whoever I can't remember exactly who said it but someone I think Merryill was that you talking about AI Chad if you don't own the infrastructure it's a tool it's not a true agentic I distilled that down to more business language which is one's a tool the other one's an employee and it feels that way and it operates that way and you guys can see there so I already talked to that slide so we can go to the next all right. I talked to that one, too. One thing that I want to say that's funny here that I put up a tweet about the other day.

You guys have all heard you I saw that image that came up, the popular image about 8 billion people in the world, less than.004% know you can use AI to code. Who in here has heard the term vibe coding or has vibe coded? All right. Have you guys ever heard the term chat coding?

I came up with it last night. So Vibe Coding put the power of being able to write code in someone like my hand. Like I've always I've been a tech co-founder since I got my first company funded in 2016 out of my dorm room. Never knew how to write a line of code.

Always love tech and what it could do. Never could write a line of code. Then I'm trying to get away from that buzz. Then recently I was finally able to do it.

And that's vibe coding. I could sit at my computer now and talk to a tool like chat GBT and it's outputting code. That's vibe coding. Then there's chat coding.

That's when I'm on the golf course and I'm texting my AI agent, hey, make me a presentation for me to present at this event tonight and use the corpus I mentioned earlier or the workflow I mentioned earlier. Query my transcripts for my meetings my social media engagement find the topic and make a website while I'm on the eighth hole right that's chat coding no longer do I even have to sit at my computer to write code I now have an AI agent that can write code for me that's studied how to write well-designed sites how to use our branding how to talk like we do our fonts and I'm literally walking hole to hole writing saying hey take you take the transcript from this call and turn it into a a site that I could send a client as an interactive demo and then three hours later I'm getting a call saying it's the best proposal that they've ever seen in their life and I hadn't even made it to the next golf hole yet. I'm not that good. So don't laugh.

He's the one I go golfing with. All right. So that but that's the difference. Does does everyone understand proactive versus reactive AI at this point?

All right, cool. Is this a new concept to you or is everyone kind of familiar with it? I'll take the silence as it's super new. You rock.

All right. The death of the browser. This is the last cool thing that I think is kind of a new concept. And it's something we're seeing, right?

Like they say bu if you're going to build software, you need to build for agents. Building the UX for humans is something that is not going to be as important in the next who knows how many years, right? The delta on change is is exponentially changing. And so the death of the browser means what did software go to?

Software's always been a dashboard, right? It was a dashboard on a floppy disc. Then it was a dashboard on a software you download. Then it was a dashboard on a tab in a browser.

Then it was a dashboard on a mobile application. Now dashboard is the second part of the experience. The first part of the experience of interacting with software will be through chat. It'll be where it is.

And you're going to tell your agents to use the APIs of different software, all the different tools. I don't need to look at a dashboard. In fact, I put out a video on my Instagram about this the other day. The only time I ever pull up my CRM is on our Monday morning sales meetings so we can run through and I can hold people accountable.

Outside of that, I'm texting Mo, my AICRO, on Telegram saying, "Hey, who do I need to follow up with today? What should I say?" what what kind of data should I be looking at to whatever it is you would ask a human CRO is what I'm asking my AI CRO and that has replaced and it's not binary right it's not like 100% of my sales data is coming from that chat 0% from a CRM dashboard but the pendulum is swinging the spectrum is moving to where I would say in terms of where I'm getting the sales data that I'm acting on maybe it's 20% from a dashboard now and 80% from a Slack channel or from a telegram channel because that's how it is. That's how you do it. Now, >> also Mark Andre came out and said the same thing.

So, this isn't stuff that kind of lives in the vacuum of of our minds. We talked about ownership and owning your AI and all the benefits that can come with that. That's where you really become an agentic company, an AI company. You know, I think honestly Merryill, you said it the best.

Like if you don't own your AI and you don't own that intelligence layer, then you're renting AI, you don't own it, you're using it as a tool. You're not an AI company. And right now, being an AI company is very there are a lot of dividends on being an AI first company. Because a lot of people are still human companies.

So if you are a human plus AI company that's actually using AI in the way that companies who are winning are doing it, you have a real opportunity to get market share and win. But that delta is going down month over month. So there is urgency to get in AI, start understanding it and owning your intelligence. The more you own your intelligence, the more data you can collect and the more data you have, the more you can learn how to do the right things with it.

Proactive AI can go wrong if it's not trained well, if it's not built well, and if it's not secure. There's a few reasons why, and so I'll let Diego go into that now. Hello everybody. So, I just I just want to echo what Jason was saying in the in the panel earlier.

It all boiled down to not who how much AI are you integrating into into a company. It's you have to do it right and extremely targeted. So there are three concepts that I want to go over that kind of dive into that and I could talk for this for hours but I'll try to distill it as much as I can next slide. >> Skip it.

>> Yeah, skip the slide. Okay. So the three concepts that I want to go over are memory, security and the guard rails and skills also falls within that. So, at a high level, memory, everybody here has experienced an issue when it comes down to memory.

You you talk to Chad GBT, you're talking to it for hours, next thing it's it's forgetting. It's forgetting the first thing you talked about. That memory layer is something that exists on all these platforms, whether it be Claude, whether it be Chad Chat GPT. And that's why we as a company, we stress that you want to own your memory layer.

So, go ahead. So what can go wrong? Every session starts from zero. Amnesia on repeat.

Anybody here who has used Claude again, if they hop into a new session, it tends to forget things. There's a memory that lives within Claude's ecosystem, but it's not as robust as you want it to be. Contradictory decisions from sale context. That goes back to what I said about Chad GPT.

Things that you were talking about at the beginning. All of a sudden, an hour in, it's telling you something completely different and you're having to now argue with Chad GBT saying, "Wait a second. I literally just told you this. Hallucinated patterns from bad retrieval.

If your retrieval layer is not set up correctly, you are going to get extremely bad data. That's how bad decisions are made and that can cascade throughout your entire business. How you do things, right? You need persistent memory across every session.

We stress that you need to you need to own every single input and output that goes within your interactions with the AI. You always want to have your own data lake database that everything goes into it. And you want to connect your systems. So APIs are called whether it be your email, your CRM, your project management system.

All that data that you're using should be going into a database and it should be structured properly so it can be able to pull that business context. Now, now I'm talking about historical context at this point because businesses shift, businesses change. So you also add real-time integrations and that's what Jonathan mentioned about calling APIs. You want to give it as much context as possible and that's that's something that's called context engineering.

You want to be able to recall the past the past historical information and you want to recall what is right now. So for example, if I if Jonathan is chatting to one of our agents and he says, "Where are we at at at the stage of this deal?" Well, what's going to happen in a properly set up memory layer is that it's going to pull every single conversation that Jonathan has had with the client, transcripts, emails, everything. And then it's going to pull APIs on where we are right now. So the more context that you can give it and the proper memory layer that you have, that's how you can actually achieve the issues that most people who interact with AI face of it constantly forgetting things.

You're having to remind it and even when you remind it, it forgets it 30 minutes later. Security and data ownership. A lot of people here touched on on security. That is a that is a massive thing that most companies are hesitant to do.

AI governance, security, am I feeding my data to an LLM? Like is it training on my data? That is something that constantly people are complaining about. The biggest risk your agent has access, CRM, calendar, financials.

You give it unscoped access. You're runaway processes with no kill switch. Well, what you need to do at when to be able to counter that is you need to have approval gates with any sort of external action. And also when you first set up access to these systems, whether it accesses access accesses my CRM, I want to have view only.

I don't want it to move things throughout my pipeline. I don't want it to hallucinate things. And you need to have a full audit log of everything. You need to see why it did what it did and you can reiterate on those things.

Skills and guard rails. Who here has heard of agent skills? Okay, so not that many not that many people have heard about agent skills. An agent skill is essentially instructions that sit on top of instructions and actually tell an AI agent, this is this is an exact process that I'm going to do.

And this prevents things like scope creep. An agent does tasks that it shouldn't. You know, that's how what Jonathan mentioned. Next thing, it's wiping out my entire inbox when I said, "Hey, can you delete this one email, one bad instruction cascades everywhere." No versioning, no way to roll it back.

The way you do it, right, is that you sandbox these skills. So, I can create what's called a skill that is typically in the form of a markdown file that lives within my agent's ecosystem. And I can essentially say when I ask you to check an email, you check this email. You do nothing else.

You set up those guard rails. That is how you prevent these catastrophic things from happening. And you also version control on every single instruction because if something doesn't doesn't work out the way you should, you should absolutely reiterate it and refine constantly. And this is something that you absolutely test in staging before production because you don't want those kind of things to happen.

I mean I saw an article recently that a company's entire entire code base was wiped out versions everything. That's because they had given their agents unscoped access. They had not fine grained their their API accesses. These are critically important things when setting up any sort of agenic workflow.

There you go. Oh, there you go. Perfect. >> All right.

Thanks for that. Yeah. So, we wanted to go through the technicals. So, I just wanted to put a couple examples.

And so, I'd say scan my inbox and calendar for new contacts. I told you guys about that one. So, I got my AI CMO, AICRO, right? Every morning at 9:00 a.m.

Pull active deals, flag unanswered emails, check meeting trans transcripts, score momentum, draft follow-ups, report to Slack. And then the score momentum, it's like how does it know to do that? You have to build a skill for that. If anyone's played video games like Sims or Madden, when you're building your agent, it's like giving it attributes.

It's giving it a personality. It's giving it things that it can do. So if you want to visualize it, that's what it's like. And you've got to be really kind of pristine about that.

You know, Marlin, one of the best things about Marlin, who's our PM, our AI PM, is we've got a great team of AI project managers. And this helps them analyze behaviors and insights so that we can help them identify hey here's an area opportunity where you can expand into right and upsell them on that or help them avoid certain things and that has been really great in that regard and then multi is my kind of he's my guy that's our AI CEO and he's for lack of better words goated but he he he runs the orchestration layer for all of them Right. And so they all run different models. They all have different tool chains.

I think I talk about it in here, but if I don't, we've got 50 softwares or 40ome softwares connected to them. Each one of them, I think like Atlas has the most skills. He's got 60. Mo has 30, Marlin has like 27 and Multi has like 40 something.

So that's kind of how it goes. So this is this is a skill. These are skills. This is this is how we talk to our agents.

I'm not putting a prompt a 40line prompt on how to make skin look more realistic on an image model. I'm talking to it like it's an AI executive on our team. And then it looks like we're out of time for the live demo, but go to the next slide. We do do a free workshop where I do go into this stuff and I go deep into it and I'm sitting there and I'm talking to our agents live.

I'm talking about more in depth. Tomorrow morning, for example, I with Cobb County School District, who's one of our bigger clients, we're talking at the Cobb Chamber of Commerce of how we are building a virtual newsroom for Cobb County School District. And again, to kind of emphasize that, we went in there and we step one, what is the ROI? What is the KPI?

Well, we want to tell more. We we want to tell more stories. Okay. What will telling more stories do?

It'll give us more time to do other things. What What will that do? It'll increase attendance in the school district. That's what we needed to hear.

Okay, great. What tools do you use? Where does your data live? Boom.

Got it. So, now we know we need to build a virtual newsroom because the school district's job, the marketing department at the school district, their job is to tell more stories. They're like a news station for the school district. And that is their one goal is to tell more stories and bring more people to the school district.

So, how do we help you go from seven stories a week to 65 stories a month, right? Measurable, attainable, achievable, AI enabled. These are the kind of case studies and things that we dive deep into on here and then I jam out with y'all as I build agents live. So, I would suggest if you want to see me do that, scan that QR code, get involved and then is there another slide?

I don't I don't even know. Yeah, again important AI models are free. What's not free is the last mile. That's the part that's going to have the last human job is the people that are best at using AI at the last mile.

And if you're good at the last mile, the last mile is infinite. >> But the guys the and gals, the people who are best at figuring out how to use AI in the real world, the last mile, those are the people who are going to win. Those are the people who are gonna jobs will be the least to go. And at the end of the day, someone else said it, but I'll reiterate it.

The best definition of intelligence is adaptability. And I hope that tonight we were Diego and I were able to give you guys some of the tools to adapt at this time. >> Open call. >> so it depends on what Yeah, that was open call.

Yeah. But we've built our, if you're familiar with Nemo Claw, which is what Nvidia, so Nvidia built their own version called Nemo Claw on top of OpenClaw with their own guard rails and their own stuff. And what they are to enterprise, we have a product called Shoefly Workmate that is to the SMB. So Workmate is our guard rails and our structure built on top of OpenClaw and that's what that that is what we use internally.

So we we eat our own cooking. Can you give us a commercial of your company and what you do? >> Yeah. Yeah.

Yeah. So, all right. Action. Now, here you go.

We build we build private ownable AI infrastructure for mid-market companies. That would be the shortest way to say it. 7 to 75 million businesses. Anything less, we try and identify workflows and productize them that there.

Anything up they can hire people or have internal resources. IT teams use cloud code. That mid-market spot is perfect because they want to take advantage of the AI wave. They have capital even though the calculus on build verse buy has gone down.

There is an entry there is a barrier to entry but it's so much smaller than it used to be and they can afford to do that and they know they want AI. They don't know where to start. And they they don't know how to measure it. They don't know how to do these things.

So that's what we focus on is building private ownable AI infrastructure for 7 to75 million businesses. All right, let's go.