Annie Croner:
Welcome to the Whole Assistant Podcast, where assistants come to embrace their badassery and discover how to show up more strategically for their careers, their executives, and most importantly, themselves. I'm your coach, Annie Croner. Join us as we dive into the skills, strategies and mindset that will help you unleash your full potential. Let's go. Okay, guys, welcome back to the podcast. Today with me is Ebony Milhumer. Ebony, I don't know where to start with you because I love you so much and I think you're so wonderful, but I'd love to have you introduce yourself to my podcast audience, share a bit about your background and then also what you're doing now, because this is really fascinating, people. I really need you to hear this.
Annie Croner:
Go ahead.
Ebony Belhumeur:
That's awesome. Thank you so much for that warm introduction and the feeling is a hundred percent mutual. Annie, I also love you. I love what you're doing in the space that you've built in the executive assistant community. A little bit about my background. So I started as an executive assistant. God knows how many years ago. I got my first big girl job, as I like to say, in San Francisco supporting the SVP at Sephora, where shortly after that I was promoted to supporting the CEO of Sephora at a 10,000 plus in person company.
Ebony Belhumeur:
Had to manage a very steep learning curve pretty quickly. I had just had kind of my first kid. I was 26 years old, really green around the edges, but was helping to steer the ship in the right direction, as you say. From there I transitioned out of corporate retail and into technology, which was not an easy transition, but definitely a welcome one. I knew I wanted my career to go in the direction that tech was trending, which was a much more flexible work environment. So I currently and have for the last several years worked from home in a pre pandemic environment. I really loved the salaries in tech as well, and I thought that I was. It was a really smart way to kind of future proof and engineer my career in a direction that I would like to develop.
Ebony Belhumeur:
So from Sephora and corporate retail, I went into tech supporting the CEO and founder of Twitch. Shortly after we were acquired by Amazon for around a billion dollars. My boss at Sephora was reporting directly to Bernard Arnault and then at Twitch reporting directly to Jeff Bezos and then Andy Jassy shortly after that, who is the current CEO at Amazon and ran AWS and did a great job over there. From there I went into Blockchain, supporting the founder and CEO of a company called Protocol Labs and they invented he invented something called ipfs, which underpins most of the decentralized distributed systems that you find around the Internet. It was another steep learning curve, but really awesome. We never had a bad day. We never had a boring day, that's for sure. We did have some bad days, but we never had a boring day.
Ebony Belhumeur:
Was just really excited to be around so many competent, intelligent, really passionate people working in this space. I stayed in the blockchain and crypto space for a while before leaving to start my own AI company where I'm currently at now. Building for the future.
Annie Croner:
Yes. Awesome. Okay, so that's what we're going to be talking about today is AI and how to integrate AI into your workflows and how do we actually make sense of the inundation around all the tools with AI and all the things which we were talking about prior to hitting record. So Ebony, where would you like to start with this conversation? I'm going to open it up to you because you're obviously the expert here. What do you feel would be most valuable for executive assistants and support professionals to know moving forward?
Ebony Belhumeur:
Yeah, so I think what's top of mind for me, and I think there's so many places that we can go because I have a broader critique, I think about the folks who are at the helm of a lot of these organizations. They are, they kind of own the cluster model owners like OpenAI Anthropic. Less, less so. But you know, there's a broader sort of anti human sentiment that they give off. The posture feels very kind of antithetical to being a technology that people would want to embrace. And I think there's a lot more good that this technology can do, especially if we have folks kind of building and helping direct this technology in the service of humanity. But for support professionals, what's top of mind for me is, is learning to kind of decipher from the inundation of noise that we have and turn that into signal to figure out pretty quickly, I would say, over the next kind of 3 to 9 months what competent AI integration looks like for you. Because we are definitely out of the.
Ebony Belhumeur:
The last 24 months, I want to say, have been characterized by experiment with AI. There's been sort of this generous, I think, you know, what a lot of leaders of organizations feel is this generous sort of onboarding time of 24 to 36 months for folks to kind of get acquainted with this technology, start getting accustomed to its capabilities, and now to be able to show and demonstrate clear results. So top of mind for me in terms of this community is understanding what competent integration of this technology looks like for you within your organization between the next three to nine months.
Annie Croner:
Okay, so that begs the question what does competent integration look like? And I know for so many people we're using it like we use Google. We go in and we ask questions and we get data points out. But like what does it actually look like to actually maximize AI in our work environments?
Ebony Belhumeur:
Yeah. So I think that the reason that we see that is it's important to just take a step back and understand why are so many organizations think failing to competently integrate this technology. I think we all saw kind of last year MIT and report after report is finding not only failings of corporations to deliver ROI on whatever their spend on AI is. Right. They're allowing people access to enterprise grade AI and the ROI is negligible if mysteriously disappeared. So the reason that's happening is because this is brand new technology and they borrowed an interface for from search that was never meant for this kind of technology. Right. So we have these large language models and it's difficult for an average person to understand what this technology is, what it, what an LLM is, what is it doing? What is a gtp exactly right? Did you ask somebody, hey, what does chat gtp, what is it? What does it stand for? Most people would not know.
Ebony Belhumeur:
So if you put the same ui, which is the search interface that we find in the Google chat box, if you put that UI on top of a technology that is a hundred percent, it's wholly different than what an LLM is and we ask people to use it. When you deliver large language models as a chatbot, for us, it's probably the most inefficient way to deliver this technology to people. It becomes very unclear how we're supposed to get results and outputs from this thing. How are we supposed to finesse it? Do we need to use a magic spell which prompts have sort of taken on this sort of otherworldly thing? If I just screenshot this prompt, if I copy and paste it directly in here, it becomes some sort of magic spell that will help you to finesse outputs that you can use. And because people do not know, both don't understand this technology and don't know how to use it, we AI slop was the word of the year last year that was 2025's word of the year was AI slop. What happens is that you get slop because people do not understand this technology. So that's why most of us are failing to see ROI and organizations are the reason organizations are feeling an anxiety, I would say deep anxiety around this. We saw Facebook brought together at their year end off all hands were all however many employees they have at this point.
Ebony Belhumeur:
They brought together their entire group of employees for their end of the year all hands and said that AI is no longer optional, it is mandatory. And your performance will be now evaluated. Your performance conversations will be evaluated based on how much time you're spending on these machines, how much work you've been able to do, how quality, how high of quality you've been able to deliver as a result of utilizing these machines. So we know that it's going to be increasingly tied to our performance and then of course, to any job prospects that we have. Right. If we don't understand this technology and we don't know how to competently integrate it into our workflows, our performance is going to. Our performance evaluations are going to suffer. And then our potential at achieving or succeeding in the workforce and with evolving jobs is going to suffer as well.
Ebony Belhumeur:
So it's important to understand this technology for sure. That's its own barrier, right?
Annie Croner:
Yeah. And my question there is like, how do we get past the slop? Like, how do we actually get out of it? Something that would be beneficial for our careers. How do we actually get out of it? What would be beneficial for our organizations when we have no real knowledge? Because there again, the interface looks a lot like Google.
Ebony Belhumeur:
Yeah. And so identical.
Annie Croner:
Yeah. So then like, how do you manage that? How do you actually conceptualize what you can get from this technology that will actually benefit our organizations and ourselves? I am also a big fan of like, coming back to our organizations and being like, hey, I heard this great podcast with this gal from Dapple and she was talking about X, Y and Z. Like, what would be some things, some takeaway points that you would have them share, perhaps with their leader or perhaps with their organization in order to max maximize the ROI piece, as you said.
Ebony Belhumeur:
Yeah. So I think the two biggest blockers for individuals are mastering two really complicated and complex skills. The first is prompt engineering and the second is context engineering. They're difficult to master. And when I say master, in order to consistently get high quality returns, to be able to use AI at a very fast clip without degrading the quality of your work, you would need to master both prompt engineering and context engineering. They're complicated and difficult to learn. It's also time consuming, which the overwhelming majority of businesses are not providing a window of time. Experimentation is telling, is not teaching.
Ebony Belhumeur:
Right. So there's a lot of telling, but there's almost no teaching on the part of organizations. So I would tell people that if they have the time, focus on mastering those two skills and if you don't have the time, sign up for dapl because we've created actually the first adaptive orchestration layer that manages prompt engineering and context engineering on behalf of the user with also with several other great features that I think people will love. So those are the two biggest blockers for individuals. So you need to overcome those blockers as best you can on your own or start using products or tools that can help you overcome those. The second, I guess in terms of looking at organizations and companies is they actually have not spent any real time. This starts with leader at the leadership level. At the executive level, they haven't spent any real time considering what are, what's a framework for good, what kind of returns we want, where would it make the most sense? Depending on our business, right? Because depending on the business will really depend on where you hope to get the most leverage from large language models or artificial intelligence more broadly.
Ebony Belhumeur:
They haven't spent they've been almost no time spent understanding how we need to strategically integrate this technology into our organization depending on the type of business. So at the executive level we really need people coming in and talking to leadership who understand the business. So I would think operationalists who understand how the business works and then understand the unique capabilities of large language models and where they can best be deployed in their business.
Annie Croner:
Hey there, quick pause. Do you want to grow in your career but don't have a ton of extra time, money or let's be honest, energy to commit to leveling up. If this is you, I have got you. Empowered Seat is my intentionally affordable membership for busy executive assistants and support professionals who are done white knuckling their careers in isolation. Inside Empowered Seat, you get support that's actually built around you. You're going to get access to me inside of our VIP online community. You're going to get two monthly calls with me as well. A monthly training session you can watch or listen to on the go.
Annie Croner:
It's pre recorded so you can either listen through our private podcast or in the vaults. It's affordable, practical, sustainable growth built by someone who gets it. If this sounds like what you've been looking for, you can learn more and join empowered seat@wholeassistant.com empowered seat or just click the link in the show notes below. Now back to the episode.
Ebony Belhumeur:
Yes, awesome.
Annie Croner:
And for those of us who are not familiar with large language models just to go back to real basic, a brief overview would be great.
Ebony Belhumeur:
Yeah, so, yeah, so what, what most of us see as anthropic or Claude or mistral hugging face or of course the Most ubiquitous is OpenAI. We see that as one of two things. A chatbot that we can kind of tell our woes to or say things to, or a genie that you can use some sort of magic prompt and get a return. What these large language models do is, is they are based on prediction. So they are probabilistic, they are not deterministic. What do I mean by that? It means the, that when you, the tokens that they're using are, they're trained on billions, in some cases trillions of parameters to predict what the most likely next word is. And they are very good at this. They are better at this than probably almost any human.
Ebony Belhumeur:
They're incredibly good at this and they can do it very fast. Their speed is incredibly impressive. What happens though is because they are probabilistic, the LLMs are unable to say that they're wrong. They're unable to pause and say I don't know that information. Their job is to predict the next word, what the most likely word is. Now you might know that you have a blank spot of information where there's something going wrong, but because it's predicting words, it sounds incredibly intelligent. And for the most part it is. But this is also why you get things like hallucinations.
Ebony Belhumeur:
So these are word prediction machines. Okay. When you're using something like Google Search, what they're doing is they're searching for information that already exists, so they're not creating information. And this is why when you use a chatbot that looks like it's Google Search, you're deeply confused because you think it should be going to get true information and bring it back to you, but that's not what it's doing. It's generating. Right. So these are generative, pre trained, pre trained transformers. That's a GPT.
Ebony Belhumeur:
So it's generating information based on most likely predictions. So that's a, I guess the most elementary way that you can kind of understand this technology. So it's probabilistic. It's going to constantly and perpetually fill in what the most probabilistic next word is. Regardless. This is why prompt engineering and context engineering are so fundamental, I would say especially prompt engineering. Right. Because it's probabilistic.
Ebony Belhumeur:
Prompt engineering is how you frame the information that you are trying to extract. And in order to master something like prompt engineering, I would say you would need to. A person would need to understand at minimum around a dozen different techniques. Things like chain of thought, least to most self ask around a dozen of these techniques you would need to kind of understand and then you would need to know how and when to deploy them, depending on the type of work you're asking the large language model to do. Context engineering itself is also incredibly complex because we know that context flow drives drift and hallucination. So it's difficult to know when does the large language model need this information? How does it need this information? Should it be a PDF? Should I take an image to help download and provide context? And then how much does it need? Right. How much is too much before it starts to blow and drift? So I guess that's a very elementary version of this technology and why you need these two skills and how you would hope to deploy them in terms of being able to increase the quantity, the tick up the output that you're doing, but also across more domains than the one you've been hired in. One of the things we talk about at Apple is evolving jobs.
Ebony Belhumeur:
So a lot of people talk about the despair and destruction and doomerism. Jobs are going away, they're being destroyed. Normalcy is being destroyed, industries are being destroyed. We don't think of it as actually destruction. We think that what's happening is an evolution. And evolution can look a lot like destruction. It's easy to mistake those two because you're not seeing the advances. You can only see what's changing, what's breaking.
Ebony Belhumeur:
You can't see what's being constructed. And so that's what we're trying to help people see at DAPL as well, that like it's actually not destruction, it's evolution. So jobs are actually evolving. Industries themselves are evolving. They're transforming at a rapid clip too, too. So the pace of change is also disorienting for folks. Their jobs are evolving. That includes strategic reconstitution.
Ebony Belhumeur:
So what value does a job actually add to a place of work? Understanding that is critical to being able to position yourself in your existing role and in any future role that you hope to have. The other thing is we're seeing an expansion of your operational purview. What do I mean by that? You might have just come in with a super static job title. Those are going away. Like when you wore a hat that said, I am a marketer expert. That's going away in five or 15 years. We'll laugh at people who kind of were like, oh, I just wore. I was hired as a data scientist.
Ebony Belhumeur:
You're now going to be moving across domains a lot more often. The work that you do will encompass multiple domains or directly bleed into other domains. Yeah.
Annie Croner:
And from my vantage point, like on a micro level for support professionals, I mean, there's been a lot of talk around how AI is going to affect our roles and actually showing up more strategically and actually bringing a human element and actually like the cross functional piece of what we do and how important that piece is going to be in this new era with AI. Because I mean, obviously what has traditionally can't even believe I'm going to. These are words are going to come out of my mouth because I firmly do not believe this is true regardless of where you are in time and space. But like the traditional role of executive support, managing the top three, which is like travel calendars and meeting, I mean all of that is just going to be potentially outsourced to AI and we can either get on board with that and we can learn how to use the tools and we can learn how to like effectively leverage the tools that are presented to us or we can step aside. We're just going to be obsolete if we don't actually learn some of the skills around AI and how to leverage that. So.
Ebony Belhumeur:
Yeah, and this is why. Yeah, yeah, this is why I think people should be excited about the courses that you are launching. So in my course, how AI is shaping the role of the ea. How AI is shaping the EA role, we talk specifically about this. Two things. One, I share the good news and I try to share this good news all the time. The good news is that we are uniquely equipped executive assistants and support professionals are actually uniquely equipped to move very successfully into a fast paced, dynamic, evolving workforce for a couple of reasons and I'll get into all of them in this course that I hope people have an opportunity to take. But one of the main reasons is that we've always existed and thrived in ambiguous domain environments.
Ebony Belhumeur:
We are the people who take ownerless projects and drag them across the finish line when no one owns anything, when we are constantly wearing multiple hats so we understand broad domain work. And this makes us so well equipped to succeed in this evolving workforce. Yeah.
Annie Croner:
So just so you guys know, EBONY is going to be coming into empowered seat as of the time that this recording goes live. It should already be up in the portal for you to all go in and take. And I highly recommend that you do. Ebony, can you share a bit about DAPL and how, how support professionals may Utilize dapl, how they can leverage dapl. What DAPL is exactly? I know you touched on it briefly, but giving a bit more context and a bit more in depth description of what it does and how it can really help people level up and also hopefully move them in their careers forward as well.
Ebony Belhumeur:
Yeah. So the primary focus of DAPL is to make sure that the average person can do three things. We like to think of ourselves as the CANVA of AI. So CANVA does not make people better at Photoshop.
Annie Croner:
Right.
Ebony Belhumeur:
I was a person who was just like completely and totally broken when it came to like figure out Photoshop, figure out figma. Like my brain does not work that way in any way, shape or form. CANVA allowed people to deliver sort of Photoshop level outcomes with a layperson skill without the Photoshop learning curve. That's exactly what Dabble does. So our job is to make sure that everyday people can leverage artificial intelligence to increase the pace of their work, the domains that they're working in, without degrading the quality of their work. So we do things, we have a couple of things, new pieces of technology that we've invented. One of them is domain directed execution, or dde, which is a reasoning chain which tells the model exactly what context that it needs to work in. And then we have dpci, which is something called dual phase context injection, which basically shifts what you're typically doing and it hones in on the 1% of relevant context that you're sharing and allows basically, if you think about it as like inference capabilities on what models are doing.
Ebony Belhumeur:
And we built a tunnel through a mountain right of the context that you're doing. We build a tunnel, we drill a tunnel through and we're putting a high speed train through it so that you get to your results not only faster, but at higher quality, without hallucinations, without drift. So there's two main ways that people interact with artificial intelligence today, and that is artificial intelligence as a coder. Those are people on GitHub, Copilot, OpenAI, Codex, Claude Claude code, so you can use it to engineer and program. And folks who are engineering and programming already, they're doing an amazing job and it's giving them massive amounts of leverage, particularly because engineering languages are already prefixed and that's what large language models do really well. So it's a unique, there's a unique symmetry between engineering. I know everyone's like, oh my gosh, if the engineering engineers are losing their job, we're next. Engineering is actually a really unique use case that has Specific and unique, high synergy with large language models and what they do.
Ebony Belhumeur:
There are other areas where it totally fails. Right. You saw this recently. The attorneys were censured by the district judges for hallucinating context in law, citing, fabricating cases. It does terrible in that. And that's not to say it doesn't make any mistakes in coding, but it makes far less. So our job is to make sure is to create a third option. So, so not AI as a chatbot, not AI as a coder for non technical jobs, but AI as an execution layer.
Ebony Belhumeur:
Right. That increases the click that you're working at without degrading your quality and allows you to work like an expert across any domain. So one of the things, I'll just give a smidgen of background here. I know we're probably running short on time and I apologize if you have to cut this. Feel free. One of the things that happens, as I mentioned with prompt engineering and context engineering, if you're really great at marketing, you're probably going to be okay in your prompt engineering, likely in the marketing domain. But let's say you're in the evolving jobs that we talked about and now you're having to manage data, but you don't have a background as a data scientist. You don't know how to say what you want to say or what to say.
Ebony Belhumeur:
So you're lost for those two skills because you're working in a domain where you don't have experience or expertise. DAPL allows you to work with that experience or expertise because of the unique modular logic blocks that come preassembled. So it's kind of a borrowed expertise that you get to leverage on the Dabble platform.
Annie Croner:
So cool. I love that. That's really cool. Okay, so tell people first of all where they can find you. And guys, I'm going to link to all of this in the show notes below. But I do want to give EBONY the opportunity to actually tell you where they can find you. Ebony. And then I also want to make sure that if you're interested in DAPL that you have that information as well.
Annie Croner:
So please.
Ebony Belhumeur:
Yeah, so you all can find me on LinkedIn. Please do. I'm @EbellHumor. It's a little bit confusing to spell, so check the notes. It's French spelling. So you can find me on LinkedIn evalhumor or you can email me@ebonyappleai.com and then to sign up for our waitlist, which is growing, we're at over about 1500 folks on the waitlist. We're only going to be taking the first 150 though, for our private alpha, so get onto the waitlist as early as possible. You'll find us at www.dappleai.com so please check us out there and be sure to sign up for the waitlist.
Annie Croner:
Yeah, guys, I feel like, especially where we do serve so much cross functionality, like, I feel like a tool like this will be very helpful for you and your careers moving forward once it's more broadly available. So please sign up for that. It's going to be awesome and helpful and I'm going to link to all the links that EBONY just shared with you. So thank you so much EBONY for coming on. Thank you for sharing all your wisdom and all your expertise. I found it fascinating. I hope you all did as well. And that is all for now.
Annie Croner:
If today's episode gave you language clarity or just a quiet oh, it's not just me. I am so thrilled you're here. If this podcast has been helpful for you, the easiest way you can support the show is by taking 30 seconds to rate and review it on Apple Podcasts or Spotify. This helps more assistants find this work and if you're ready for ongoing support, guidance, community and growth that actually fits into your life, Empowered Seat is where we continue this work together. It's affordable, flexible, and designed for assistants who are done white knuckling their careers in isolation. You can learn more and join@wholeassistant.com empowered seat or click the link in the show notes below and until next time, go embrace your vetassery.
