Putting AI Agents to Work, With a Human in the Loop

Lightly edited for clarity from a transcript of the final cut. Sponsor reads omitted. Solo episode: Ysi Gonzalez, Vantelira.

This episode is powered by Vantelira. We go inside a scaling operation, fix the process, and build the system that keeps it running.

Episode 15: a solo episode that closes a block

Hello and welcome to The Forward Deployed Operator. I am Ysi, and I have the pleasure today of doing my fourth solo episode, I think, where I will be walking you through the recap of the past two episodes on the show, and a little bit of what has been happening with forward deployed operations all across.

I had the pleasure to have two very special guests on the past two episodes: Susie Bulloch and Sandra Vergara, both very close and dear to my heart in their own different ways.

Recap: Susie Bulloch and Hey Grill Hey

Susie actually is a self-published cookbook writer, and she has her own Netflix series, which my daughters are obsessed with. If you want to go and watch that episode, pause here and go look at it, because there is a little pop-up in it saying hi to Susie. It was a very fangirl moment for both of us. So go and watch it.

We talked about everything operations, everything from funding a company from the start, building a community first, and building a blog website to generate some revenue for her to be able to provide for her family in a way that would still allow her to be present for her kids. And I really appreciated how candid she was in that conversation.

I am actually a user of the products, too. This is my favorite rub. And it was very interesting to hear her say the purpose, the North Star, of why she started that business and where Hey Grill Hey is at today. I think there's a very strong learning in that episode.

The lead parent, not the default parent

Especially around the term lead parent, which I am going to start using more often now. It's no longer the default parent, but the lead parent, and how that role can shift depending on the stage of life the family is in.

As an example, you've met my husband already on the show. Sometimes he's the one that has to be the lead parent, because I'm traveling or taking on bigger projects, and then the same thing happens on my side. I get to take the lead parent role from time to time, most of the time. And I really enjoy that, too.

Consumer brands and AI assistants that shop for you

In that episode we talked about everything operations: how she thought it was a headache, actually, to go into having a physical product because of all the logistics that it involves, and how they're leveraging AI for everything that is changing today.

Consumer buying patterns are changing based on how AI agents are becoming more involved and part of our daily life. Me, particularly, I already have a couple of AI assistants working for me on different tasks, and one of them even prepares my grocery shopping list and puts it in my grocery cart. So I just have to go and review and approve. And those are the kind of things that consumer brands are getting in front of, or should be getting in front of, to make sure that they're prepared for the agentic era.

So just go and listen to and watch the episode, or read the transcript on our website. It was amazing. We had the chance to talk about everything in between building the brand and then building the product, how things shift when you start growing, and how important it is to build your business plan from the get-go and stick to it. If you're working in CPG or product-based brands, you have investment in the inventory, which is something you have to be prepared for in advance to make sure that you're growing properly.

I've seen firsthand what happens when this doesn't happen. And I've also seen firsthand what happens with companies when it's properly done. Then you get into managing that inventory, managing the excess, and all of the data points.

Recap: Sandra Vergara and Zapier

Which brings us to the episode that came after Susie's, with my very dear friend Sandra Vergara. It was a pleasure to hear the story of how she started building her career, then moved to Canada, and has had a successful career here too, starting by managing inventory and calculating the excess of inventory, and how her brain works in operations.

We were just jamming in that episode. I really appreciated being able to sit down with somebody else who thinks in an operations way, so we could talk about the nitty-gritty things.

Behind the AI hype: a table with a deterministic workflow

Everybody thinks right now that everything is AI, when behind the scenes it's really just a table with a deterministic workflow and options, where a system goes and connects and makes a decision based on what is already written. And we've been doing that for 20 years already.

It's not the agentic portion that I find so important. It's how easy it is now to create those tables, connect the workflows, and have a user-friendly interface that helps the user be faster and make those decisions faster.

So also go watch that episode with Sandra Vergara. She is from Zapier, which is very dear to my heart. Most workflows could be solved with a workflow automation done in Zapier or similar software. She talks about all of that, and how she takes a step back whenever there's an automation that needs to be built, which is very similar to what I do in my day to day.

Today: building automations and agents to work for you

On to today's subject. I want to talk about building automations and agents to work for you.

Ranking the AI assistants I use

Currently, I am experimenting with a couple of tools out there. There's Instinct AI, which right now is my top tool, the one I'm very close to. In second place is Grok Bot. It is mind-blowing what you can do with it, and I'll walk you through all of them in a little bit. In third place, I've started playing with Muse, and it's okay. I haven't played with it enough to say this one is better than that one, which is why it's in third place.

And in fourth place, just because of certain limitations, I have put Claude. I know, I know, I know. Claude has come to the bottom of my list of AI tools that I'm using, but it's because of the way you have to set it up. It's more technical to have it prompt things to you, and I'll explain what that means.

Instinct AI: an assistant that texts you

Instinct AI is a new AI agent, or assistant, that you can set up with your phone number. So it's always going to text you, and you can text back. You can send voice notes and it will understand. I've been using it for many, many weeks now, and by the time you listen to this episode, it's probably going to be a couple more. And it hasn't made a mistake on what I said. I have an accent, and I don't speak English the way these tools have been taught, but apparently they've been taught very well.

So you can text, you can send voice notes, and then you can connect different accounts to it, like Gmail accounts, and it's always on, going through what comes into your email.

Always on, and through do not disturb

What I appreciate about that, which is the thing I find I'm missing with Claude if you don't have another interface set up, is the fact that because it's always on, looking at my emails, it can let me know when something important pops up.

And because it's an SMS, I can configure my phone so that SMS from this AI assistant goes through immediately. It bypasses my do not disturb or my sleep mode, these types of things, which I think is the unfair advantage they have right now, and the reason why they've raised so many millions. I really appreciate how proactive it is.

The email that changed the morning of an on-site recording

I was going to an episode recording on a Saturday morning at a location, and I already had Instinct set up. The guest sent me an email that morning. Normally you have to go into your email app to get that information, and I don't go into my emails right before an episode recording, because I'm getting ready. I'm getting in the zone, I'm breathing through my notes, reading through my things, my makeup, my hair, all of this.

On that specific one, I was going to the guest's offices to actually record on site, as a forward deployed operator does. And they sent me an email that there were closures around the area and that parking was limited. I wouldn't have known this if I didn't have my AI assistant monitoring my email. As soon as it saw the email come in, it sent me a message.

I have many emails that I have to go through. There are promotions; you all know about this. There are things that come into your email that you don't necessarily want to know. So I turn off notifications with a preview of the email, because I find that disturbing. When I'm getting ready and I don't want to be bothered, my phone is on do not disturb. Nothing gets through except my agent, which knows something is important.

So that morning, it sent me a message, and immediately I saw it and I was like, I'm changing my plans, I'm getting an Uber. So I went in an Uber, my morning was flawless, I didn't have any delays, the recording was done, and it was amazing. You're going to see it: it's actually the episode right after this one, so stay tuned next week for that amazing episode we recorded on site at a guest location. And I only knew this because my AI assistant was texting me, which is why I still put it at the top of my list.

Caveats: access, privacy and confidentiality

However, there are caveats that come with this company. As of the day we're recording this episode, there is still a limited way for you to remove and add access to certain tasks. So if you are managing sensitive information, or you're working with IP, you might not want to connect it. There are privacy laws, and companies have confidentiality agreements, that you should look at before connecting this type of app. I have it logged in with my email just to read and notify me, so I'm good, but you should look into it before you get that app hooked up to your systems.

What it caught on the calendar

It also looks at my personal calendar, and it is mind-blowing what it can find. I remember one day it texted me: you're supposed to go pick up your daughter at this time, your husband is out of town, and you also have this meeting booked in your calendar, so what is it? It didn't have the context that my girls can stay at daycare for one day. But imagine having that information at the top of your head on a daily basis. It's very helpful for us busy founders.

Grok Bot: different agents for different roles

The second one I want to put on the list is Grok Bot. It's in second place only because it doesn't have the SMS connection. That's the only reason why Grok Bot is in my second place right now. However, I might look at having the notifications for Grok Bot override everything on my phone and just pop up all the time, and that might make it my number one, because it is very smart, and you can have different agents doing the work.

That's one flaw I find in Instinct AI: when it has too many things to do and too many rules to follow up on. If it's looking at your email, if you ask it to research something, if you ask it to add a task to your to-do list or close a task, and there are threads in the emails being handled in different ways, it may have way too much context. It doesn't get overwhelmed, but things fall off its plate. It's like a person: when they have too much, something is going to fall.

So Grok Bot allows me to have different agents for different workflows or roles.

An org chart with humans and AI agents

I even have an org chart now, and I'm very excited and very proud. As part of the Vantelira org chart, we have our AI advisor, who is a human. We have our developer, who is another human, and we have the podcast editor and producer, who is a human. I still work with humans. It's not only AI agents.

But then we have the stream of AI agents. Right now I have Lillian, who is a chief of staff AI agent, and she has two other agents under her. One helps me with my content creation. I do still need help with those things. I write, and I like to be polished and to follow a certain train of thought. Whenever I am writing something, I want to make sure there's a content creator agent looking at it and refining it properly, so I get to review those. It also creates my little cards with the quotes from the podcast. That is the content creator, and his name is Andy, like Andy from The Devil Wears Prada.

The other one is Taylor, whose only job is to look at all my open actions and match them to see if I have closed them, based on my emails, my meetings and things like that. If I haven't closed anything, it's Taylor's job to let me know. And that is the one thing I find all these AI assistants are getting better and better at: prompting things.

Where Claude fits, and why scheduled tasks fall short

Claude, on the other hand, which was on my list... I think I have to revise my order. I think Claude goes right after Grok Bot, and then there's Muse, because I haven't used it that much to actually have a strong opinion on it. I used it, I didn't like it too much, so I just kind of put it on hold. I might get to explore it more in the following weeks.

But Claude, for example, if you want it to prompt things: I've tried the Cowork environment, Claude Code and the regular chats. They all work very well. But most of the time, unless you set up a Cowork scheduled task, which has to run through your computer, your computer has to be up for it to run. If your computer is off, it doesn't do it. It doesn't work for me, because I need to be prompted.

Why founders need tools that prompt them

As the founder of Vantelira and co-founder of the Forward Deployed Operator School, I have many, many things on my plate. If you're a founder, or if you're an operator in a position where you have to manage several things... I'm also a mom and a wife, and I am an individual who likes to do things on her own. I have a very long to-do list. I need to be reminded of certain things, because, as I've said before, I don't remember anything. If you have too much on your plate, things are going to fall.

Roles and responsibilities for AI employees

And the beauty of these AI assistants and agents is that you can draft their roles and responsibilities, their job descriptions. I now have it hooked up to my Notion workspace, where I have roles and responsibilities defined. So as time goes on, I know I'm going to be adding more and more AI employees to this whole organization, because it's proving to be very efficient.

Enterprise level: humans managing agents

The next level would be the enterprise level. Once revenue is large enough and you have many, many positions, how do you leverage having your humans just managing the agents that are doing certain tasks, and then getting the human to approve? That's the human in the loop portion.

Human in the loop: who approves

The human approves what needs to get approved, and what can be handed off is a very deterministic table with decisions that can be made by the AI agent. But when an authorization or a money spend needs to happen, that's when the human comes into place.

For example, Instinct AI does my grocery list and puts it in my grocery cart, but it doesn't have authorization to go and click the button, because I want to be the one putting in my credit card and not have it just push whatever it wants. So that is the human in the loop: I put myself in that position, to go and review and then approve.

So that's where I wanted to get to in this episode: to give you an update on the forward deployed operations world, what's happening out there with AI, and what we are doing to leverage it on a daily basis. You can use it for your personal use, or you can use it for your enterprise, the companies you're working with. These latest AI assistants are bringing in a whole new level of getting on top of your work.

Security and guardrails

What these tools need to get their hands on is the security and guardrails aspect, which I appreciate about Grok Bot, because you can really select specific access for each one of the official connectors they have.

And I would doubt that a year from now there won't be more enterprise-level organizations adopting AI agents like these, because to be on top of all the tasks a human needs to be on top of, having this sidekick is a very, very good solution.

Test the tools, then look at the process first

So I want to leave you with that for today. Take time to test these tools, so you can make an informed decision on what you're using for your own benefit. Don't be afraid of looking at it and experimenting with it. Go to meetups and hackathons; they happen all around the world. There are people experimenting with AI who can guide you through it.

And whenever you can bring it into your organization at the right place, make sure that you're looking at the process first, not just adding the AI agent, or whatever you think needs to be added. Look at the process. Do you have deterministic tasks? Do you have decisions that are made? Is it a yes or no? What are your inputs? What are your outputs? What does your process look like? What is repetitive about this process that can be handed off to an AI or a workflow automation, whatever it is that rocks your boat?

New to the idea of a forward deployed operator? Start with what a forward deployed operator is.

Closing

And with that, I say thank you so much for watching and listening to The Forward Deployed Operator with Ysi. I am Ysi, a forward deployed operator myself, and I really appreciate you tuning in and taking your time to listen to me rambling about AI and operations. I'm very passionate about it, if you don't know this by now. So thank you so much for watching. Until the next episode. Bye bye.

That's it for this one. Follow the show so the next episode finds you. And the newsletter goes out every Tuesday. The link is below. See you next time on The Forward Deployed Operator.