AI as Your Digital Twin

What if your to-do list had a stunt double that got all the items checked off?

Personally, I could really use that this week with a pending vacation.

Not to brag.

And what are the moral and ethical implications of cloning the best parts of our work selves using AI?

And seriously, are you ready to look at the man or woman in the mirror and find out?

Hi, I’m Courtney Baker and this is AI Knowhow from Knownwell, helping you re-imagine your business in the AI era.

As always, I’m joined by Knownwell CEO, David DeWolf, Chief Product and Technology Officer, Mohan Rao, and NordLite CEO, Pete Buer.

We also have the second part of a two-part interview with Max Gabriel on why AI is eating SAS, and what that means for you.

But first, get ready for an insightful discussion with David and Mohan on the idea of using AI to create your digital twin.

So here’s a thought I’ve been thinking about lately.

We used to think about delegation in terms of who we could get to accomplish a task.

And there’s so much work that has gone into delegation and really framing up how you look at your workload and how you equip a team.

Now we’re seeing a shift.

It’s not just who am I going to delegate, but what am I going to delegate to, or what who am I going to delegate to.

And that what increasingly is AI, no doubt about it.

As a matter of fact, there is a CEO on this, on this, in this studio right now, that all the time is saying, how do we get that done with AI?

I try to beat them to the punch, Mohan, the cat’s out of the bag, who it is, but I don’t always beat them to it.

And many of you listening, maybe your CEO is also saying, how do we get that done with AI?

And maybe you’re thinking about your work and how you get it done with AI.

So today, I want to talk about this idea of AI as your work twin, okay?

Just hang with me, your work twin.

A partner that doesn’t sleep learns on the fly and sometimes, sometimes thinks faster than you do.

Are we ready for that?

You two are like, no, we’ve already been doing that.

We’re light years ahead.

Well, I like the way you described that because I think it implies something a little bit different than a twin, right?

I think a lot of times people hear AI and they think about this concept of a twin, which is how can I replace myself?

How can I replace a worker?

Undoubtedly, there will be agents that do work on our behalf, right?

I think that’s what people think of when they hear twin.

But I like to think of AI as a chief of staff, right?

Yes, it can do some work for me.

But I think there’s two ways where it can really help.

Number one, it can be an incredible collaborator.

We can engage with artificial intelligence to break through that roadblock we have when we’re processing something, to help us get creative, to be able to refine an idea, refine a piece of work that we’re working on, to improve it, to take it to the next level, those types of things.

I also think that AI can be used for gathering information and understanding information, synthesizing information.

I think this is an area that we forget a lot.

Because AI really became democratized with generative AI, which is creating content, we sometimes forget about its research capabilities, its ability to understand the world around us.

And when I think about a great chief of staff, they don’t just execute on my behalf, they also have their eyes open on my behalf, and they have their eyes wide open, and they’re synthesizing things and surfacing and bringing to me things I need to know that I don’t even know that I need to know them yet.

Right?

And so it’s both sides of the equation of both understanding the world around us and collaborating to create the world around us that I think can be incredibly powerful, not just doing work on our behalf and automating what I would otherwise be doing.

Yeah, I love the frame of Chief of Staff, because what a great Chief of Staff does is sometimes they’re a digital mirror, meaning they know how you work, they know how you prioritize tasks, and they know your unique approach, right?

So they’re totally complimentary in that sense.

But sometimes they’re also counterintuitive, because they create a partnership with you, and intuitively know your weaknesses that they can compensate for and the strengths you may have that they will not as much kind of participate in those areas.

So it is this ability to completely synthesize in harmony with how you work, but also offer the complimentary intelligence, right?

That’s what makes AI so fascinating, that you can work in either mode with these tools.

You know, the way, in more simple terms, I think of it as sometimes I’m the author and the tool is the editor and sometimes the tool is the author and I’m the editor, right?

So, you can just completely switch how you work and the ability to flex both ways is what is truly powerful.

I got a question from one of our team members today asking me about what prompt I used to generate, I was using AI to create some work and I was asked about the prompt.

And it actually turned out that the answer was there wasn’t a single prompt.

It was actually a conversation with AI.

And I think that’s a takeaway that a lot of people can really learn, is it’s not about a single prompt.

Actually, where the human experience, where the human in the loop adds value, is actually looking at processing results and then iterating with the GPT that you’re using.

And that’s a great way to put your thumbprint on something and really make sure that it’s not this cookie cutter answer.

And it’s where actually the expertise plays out, right?

A lot of the research shows that the folks getting the most value from AI is not the junior employee that’s just asking it one question and copying and pasting.

It’s actually those that are already considered experts are getting ahead even further, right?

And I think it’s because of that is you have to know the material well enough to engage with the response and collaborate with AI.

You know, I think that’s a really good example of a trait of an effective AI work twin.

I, you know, obviously don’t have access, David, to your ChatGPT account, but I would guess you have a very effective, based on the work that you’ve produced, very effective AI work twin.

Are there some other traits of an effective AI work twin?

I mean, the way I like to imagine this, and the way I like to work with these tools, or I treat the tool like a human, like a…

That is true, people.

They got on to me for not saying thank you to my chat, GPT.

My twin is, requires good manners, but it’s only because David and Mohan has insisted, so.

You know, yeah, absolutely.

You got to treat these tools, even for me to unlock my thinking, what I do is I essentially use the tool as an analyst, a super awesome analyst, who is on a really phenomenal growth path, right?

That’s how I treat them, right?

And I have expectations of those.

Otherwise, I just say feedback, I give feedback saying that this is very pedestrian, there’s nothing to this.

Can you improvise on this?

And it’s amazing how well they respond to it.

Radical candor is amazing, right?

And if you treat it well, it will respond well to your radical candor.

You know, the other thing that I think is actually just emerging and becoming more and more important is these GPTs now have memory, right?

They can remember from one conversation to another.

I think leveraging that to your advantage is another very powerful way for AI to be your twin as you’re calling it, Courtney.

Because you can cross these conversations and cross-pollinate now and it remembers things about your past conversations and it learns who you are.

I mean, go ask it for a performance evaluation.

You might be scared once you get back.

I am scared about that.

I will say, just as an encouragement, because I think, it might have been a hot, was it my hot take a year and a half ago about executives not getting their hands on to these AI tools enough?

I feel like now, if you’re an executive and you’re not working towards having, obviously we’re calling it a work twin here, but there is so much memory there, so much learning that happens that if you’re not using it and engaging on an ongoing basis, you’re really missing out.

And really, the people that probably don’t benefit from it, obviously you don’t, but your team doesn’t as well.

I mean, there’s a really rich piece of data there for the team that happens over time.

So really what I hear from this conversation is, it’s important to train your twin.

This isn’t like actual twins where you come out with your shared secret language already established.

There’s some training that has to happen.

So the first thing, treat AI like onboarding a new team member.

Obviously the more information that you can give AI earlier on, the better results that you’re gonna get from AI.

And then second, feed it with clear, consistent inputs.

And then lastly, what I hear you two saying, and actually you told me this in person, but you have to reiterate expectations and tone.

You gotta have a good tone with AI and the desired outputs over time.

So don’t forget those…

I was about to say T’s and P’s, but that’s thoughts and prayers.

That won’t work.

I don’t think terms and conditions works either.

I don’t know what you’re getting to.

What would P’s and Q’s be?

What is that?

Mind your P’s and Q’s?

What does that mean?

Manors, but I don’t know what it stands for.

Does it matter?

Does it?

Does P’s and Q’s mean manners?

Yeah.

Okay, well, that was what I was going for.

Oh.

Like mind your manners with your AI.

Yeah, I just don’t know what.

Yeah, it is true.

Wow.

It’s a place to be mindful of one’s manners, behavior and language.

So mind your P’s and Q’s.

Yeah.

So the best thing I can find so far is it stands for pleases and thank yous.

I have no idea how you get Q’s out of thank yous, but I guess it’s the thank Q’s part of it.

Thank yous.

So there you go.

Yeah, that is making sense.

Thank you with the Q.

Thank yous.

Okay, so everybody, don’t forget to mind your P’s and Q’s with your AI twin.

David, Mohan, any other thoughts here?

You know, I think the P’s and Q’s, just to close the loop on that, why does it matter, right?

Obviously, these are not humans.

They don’t have feelings, right?

Like, it’s just a computer.

Why does it matter?

I remember back in late 2023, the first time after AI had been democratized, we went through the holiday season.

And everybody started to see a degradation of performance of our artificial intelligence during the holiday season.

And we were scratching our heads why.

And it turns out the research shows actually that because these LLMs are trained on human materials, it follows us, right?

It is mimicking what happens in the real world, right?

And so, the performance degradation was because it saw in the data it was ingesting, the performance degradation in the work world during that period of time seasonally, right?

Same thing with the P’s and Q’s, right?

It is trained on data that is based on human interactions, and so, you’re going to get better responsiveness if, yes, you have radical candor and clarity, but also if you are polite and those types of things.

So, there’s a very pragmatic reason why that can be really, really helpful.

So, sometimes the why is helpful in remembering something, and I just thought I’d pause at that as we wrap up.

Good reminder.

Mohan, anything else from you?

I think it’s always pays to be nice, so why not be nice?

You know what, I think this is just a continuation of our conversation last week when we talked about this and me not saying thank you.

I actually do think I say thank you sometimes.

I just don’t always say thank you for a response.

Thank you for joining us.

Yes, Mohan, David, thank you as always and thank you for helping us really think about how we use AI differently and maybe a slightly different mindset as we engage with it.

How do you strike the right balance in your business between humans and technology?

This is honestly one of the biggest questions of this era for business leaders.

One answer is to make sure your team has the right tools and technology to help them solve real world business problems.

Knownwell is just such a platform for professional service leaders.

From CEOs to account managers and all points in between, the Knownwell platform can help you prioritize which clients need attention right now and give you the visibility you’ve always wanted but have never been able to get.

Go to knownwell.com to see your data on the Knownwell platform today.

Max Gabriel is the CEO of Augmented AI.

Here’s the second part of his conversation with Pete Buer to talk about why AI is eating sass and what that means for leaders like you.

Max, welcome back.

Wonderful to see you again.

Thanks, Pete.

Look, there’s always a sequel every time we chat, so I’m pleased we’re doing part two.

You had too much to say that was interesting, and so we’re going to share your thinking twice.

In our last episode, when we were last together, we started the conversation on AI eating sass, and in particular, we’re talking for our purposes about vertical, agentic AI or vertical AI agents as the force or the threat to the traditional sass offering.

We gave a brief explanation in the last conversation what we meant there, but I wonder if we could go just a little bit deeper.

Tell us everything that we need to know about vertical agents.

I think it’s going to be extremely difficult to tell you everything you need to know, Pete, looking at all the books behind you.

But let me take a shot.

We established what an agent is, and vertical agents are really about industry specific, which means you are operating in health care.

So the agent is able to understand the terminology, the jargons and the taxonomy of that industry.

And then it goes beyond that where actually you’re going to deploy this agent in a company.

How much of the context of the company, their jargon, their policies that it should be aware of so that it can operate in that context successfully.

And that is also part of the challenge and the opportunity here, where if I highlight one of the core problem with SAS, proliferation Pete was data silos.

Your sales data is locked in one place, your HR data is locked in another place, accounting and finance and all of that.

Where in order for the agent to make decisions and set up choices, it needs access to all that data so that it’s able to use it and build its own intelligence.

So that’s where vertical agents specifically designed around use cases can be highly effective.

And we will start to see human to agent and agent to agent communication start to play out.

Can you give us a sort of a concrete example, maybe in a space that listeners are familiar with, of a vertical agent, what it’s like to work with versus the SaaS solution that exists currently and what it’s like to work with?

Yeah, let me talk about the SaaS product, because I think we’re all familiar with it.

The vertical agent is unknown.

We’re trying to figure it out, right?

How the software is made in case of SaaS is your pre-defining set of rules, right?

And I always feel most of the software out there, clearly there are exceptions.

I’m generalizing to make a point.

Most of the software are built for an operator in mind.

It’s like a tractor, right?

It can perform certain functions as long as the user does these things.

And that’s how these software are built, right?

So you’re pre-defining the rules, you’re pre-defining the pathways, right?

Which means the engineer and the build team had to do a lot of upfront thinking in terms of designing it and then now you’re deploying it.

In some cases, you’re expecting the client to change their ways of working.

In other cases, you’re pushing the software to change to fit the client’s requirement, right?

That’s been the age of friction.

Here, what you’re really trying to do is help the agent understand the problem space.

What is clinical trial data review means?

Educate it, train it, provide access to the data so that it can respond to your specific questions.

And that’s where, you know, you’re just introducing nuances and scenarios where you’re not really pre-programming it, right?

So for me, that’s the distinction, you know, how it has access to the data, thinking for itself, and starting to make these decisions on your behalf is, you know, where the complexity lies in.

But think about it the other way.

What I’m excited about is most of the software that’s out there has been tools for us to do a certain task, and we made the forms better, but we’re still filling out forms to do an action.

And guess what?

Computers are really good at filling out forms, and they can actually get the task done rather than being a tool for people to just do these tasks.

Does that help?

Yeah, totally.

And from the user’s perspective, correct me if I’m wrong, but doesn’t this take a lot of work off of the shoulders of the user?

Like, if I imagine myself loading up my favorite SaaS program, I kind of need to understand its architecture and how it works in the first place.

I have to structure my own queries.

I have to, you know, whereas I think the use of an agent is a whole lot simpler than that.

Now, look, the biggest winner of this new disruption, innovation, whatever you want to call it, the cycle has to be the user, Pete, right?

I don’t believe, you know, user wakes up and say, just sign me up for one more training on how to use the tool.

Why?

You know?

And that is the challenge we’re trying to tackle.

So you’re right.

It’s going to take off a lot of burden from them and get the task done for them on their behalf.

So they become a critical reviewer.

An orchestrator, as opposed to sitting and doing this in a mundane tasks.

That’s because that’s how far automation has gone in so far.

At the risk of flogging our own product, I just think about how the Knownwell platform gives, you know, answers about account relationship health in an instant that, for the conversation we had previously, would take an account manager literally hours to track down.

It’s going to be a completely different way of doing work.

Correct.

Look, I mean, throughout my career, I’ve seen enough of these implementations, Pete, to realize, I mean, this is, we cannot understate this problem.

You know, a user, how many systems do they have to go in and double enter it, and review it, and all these inefficiencies that’s been built in, that’s where I feel the user never wins.

Meanwhile, the SaaS companies and software companies are just building great valuation, but the user, all the burden is put on the user.

And the integrators, yeah.

That’s where I feel like agents to the rescue, to come and rescue the users.

Okay, so we’ve been talking kind of magically about the vertical agent and all the wonderful things it’s going to do.

I’m sure it’s not as simple as we’ve been saying.

What’s hard about building vertical agents?

If you ask me about, let’s say, a month ago or two months ago, I would have had very boastful, glorified version.

You know, I’m an optimist, as you know, Pete.

So I would have thought, this is a piece of cake.

Because the first 30% is, like I said, it’s so seductively easier.

What would take three months?

You can get it done in three hours.

No exaggeration.

Zero to 30, you can get there easily.

Which will actually tempt you to say, oh wow, this is easy.

Which is where the next tranche of 30 to 60 is when you start to sweat.

And this is multiple dimension, which we are experiencing right now, right?

Which is, you can’t think like a builder.

You can’t think like an engineer.

You got to think like, what would an agent do?

And that’s a different mental model altogether.

Yeah.

Because particularly people with, I’m including both of us in this, Pete, who’ve had long career and rich experience, we do carry a lot of baggage along with them, right?

So, we tend to use these tools just enough for what we need to do.

You know, if I’m being crude and has glorified Google, you know, saying, hey, how do I do this?

As opposed to, no, you can actually state the problem you want to solve and let the agent tell you what a plan of action looks like.

So, designing for that, the 30% to 60%, takes a lot of effort in thinking through that.

The last 30% or 40% is the hardest in terms of, because now you’re going to build all this and deploy it into a new environment where you have no idea about the jargons and terminology and taxonomy.

So, that’s where domain experts are super valuable for this.

They are going to be the partners.

I certainly envision there’s a real onboarding process for these agents, just like how you have an employee onboarding.

They get ready, they get smarter, so they can become more effective in an organization.

So, it is obviously a lot harder than what we thought.

When we did the 0 to 30, we were celebrating, then we realized the next two laps are extremely hard.

That notion of onboarding to the business is fascinating to me.

How similar or dissimilar is it to the L&D team bringing a new employee?

And is this going to be a new capability of learning and development?

Well, it should be.

You know, this just reminds me of a throwaway comment my 20-year-old said the other day.

She was meeting with one of my friends who was HR director, and she made this point about, isn’t Shad GPT the new L&D?

And I said, that was just very interesting insight in terms of actually, people are not going to learn through subscription of courses.

It’s actually they can learn what they need to know by using these agents in the future.

So you’re absolutely right.

I envision, I don’t think you need a system integrator to come and deploy this or implement it.

You really need the domain experts to onboard these agents.

You know, feeding the policies, providing access to the data and all of that.

And that is where the lot of unknown speech, right?

Well, there’s a lot of claim to say it’s ready to go when you actually embedded, how do you understand the style and the substance of the company is where the success of these agents lies.

Okay.

So we’ve got a feel for what the vertical agent does and what it takes to build one, increasingly.

Each new third increasingly generating more sweat from the effort.

If I’m a buyer, if I’m a company who has identified a use case, do I build this myself?

Do I hire someone to build it?

Will they be on the market just to be purchased?

I mean, you know, this is an age old problem.

I’m not going to give a very generic answer there, because obviously, it’s completely dependent on the culture, whether they’re a buy or build culture.

I think what I want to highlight is, it’s our own revolution here, is build has a very different meaning, right?

If you’re signing up to build something either by yourself or through a partner, I think build in the software world had a very different connotation than in the agent world.

Case in point, we’re dealing with this as a small startup, is I learned Google’s 30% of the codebase is AI generated already.

And the Anthropic CEO is saying in six to nine months, 90% of the codebase is going to be AI generated.

So what does that mean when you want to build it?

What kind of team do you need?

And the reason I want to highlight that is whichever path you take, we are feeling the friction where the 20 year olds who have never built a product, who they’re jumping in and they know how to use this to generate code.

And then the 20 years experience professionals, there is a bit of a friction.

They know how to write code, generating code, they almost feel guilty.

They feel like they’re cheating, right?

So we are going through this notion about what does build really mean going forward.

Because guess what?

We spent past 20 years learning all these languages so we can tell the machines to do something.

Actually, machines are pretty good at instructing themselves.

So I think code generation is going to be the future.

So whichever path suits the company, whether you work with a partner or yourself, that’s a completely different build philosophy.

If I’m an incumbent SaaS provider, do I have experiments underway right now to build vertical agents?

Am I going to cannibalize myself in order to remain competitive?

Yeah, no, it’s interesting.

And I would love to hear your thoughts on this as well.

When I explain vertical agents, I see these things play out in three categories.

There’s a fourth category, a lazy one, I’ll talk about it as well.

The three categories are retrofit model, which means use what you have, and we’re going to sprinkle this AI agent on top of it.

Don’t worry, we got you.

We have a new model, and we’ll do that, and I’ll give an example of that.

Then there’s a second category called replace, which is what you have is completely useless.

Because I’ve got something, either it’s going to replace the people, you don’t need 30 people to do this, my agent is going to do it, or replace the system.

This is really, really hard to do, but there are some claims out there.

The third one, which we are definitely bullish on the Augmented AI side is re-imagine.

And I don’t mean that in a superficial sense.

If you have access to a capability like this, it’s a fantastic opportunity for you to re-imagine how your workflows internally and externally can work.

It’s not about making the form simpler, it’s really about getting the job done.

A lot of those interrelated system integration, maybe they’re not needed.

So it’s an opportunity to re-imagine that.

So you can see incumbents where they play today.

Now, unsurprisingly, they’re on the retrofit model, right?

I mentioned the fourth category, they simply rebrand it.

They just rebrand it and say, we are an AI agent company now, right?

That’s the easiest thing to do.

But the retrofit model, where it hurts Pete, is you got your existing cost, and then guess what?

Pay for co-pilot on top of it.

Right.

Or pay for agent force on top of it, right?

It’s not really taking your cost down.

It’s just add the layer on top of it.

So it’s tempting for the executives to make the safe choice, but you’re not really solving the underlying problem.

Which is to your point, it’s obviously they got locked to lose.

So they’re not going to cannibalize the business model, which is probably in a fab spec that that’s what’s going to work against them.

You started answering the next question I was going to ask, which is, what does this do to market economics?

Five years from now, are companies paying more or less per person for software slash agent solutions?

Well, what would we like to do?

I think that should be the answer in terms of we should pay less and get more value.

And valuation should be based on the value they create for the customer, not for the investors.

Investors should get the return as well.

But I’m seeing a trend which is encouraging, right?

I encourage you and people to look at this leanaileadaboard.com.

Fascinating, there’s about 30 companies listed there.

Most of them on average have about 20 member teams, right?

You saw that, right?

On a big revenue base.

On a big revenue base.

What does that mean?

Incredibly low cost, high revenue, and actually if you dig in to the price point, they’re pretty low, to the point we’re talking about, right?

So what does that mean?

If they can be lean, they don’t have a need to pass their inefficiency to the customer.

That’s been the challenge with SaaS.

I’m quite encouraged by that in terms of if you can keep the economics working, I’ll give you an example, right?

So there’s a company called Lovable, not a free plug, but Lovable.

We started using it.

I hope they stay Lovable.

They remain Lovable for a little longer because success is never a good thing, but they do really well.

20 member company, 30 million ARR, and the price is $30 per month.

I can live with that model, right?

So I think that’s the future in terms of this will force fit and reshape the economy.

That’s going to be the downfall for the incumbent retrofit model because if they’re just bolting on yet more technology onto the technology you’re already paying for, how can they compete with the lean structure?

Yeah.

Yeah.

All right, Max, this has been great.

I’ve got a dismount question for you.

Look into your crystal ball for us or read the tea leaves.

What’s your advice to business leaders having listened to our conversation and thinking about what the future holds for them?

Yeah.

Look, I don’t know if I have advice, but what would I do if I’m in the position, as I’m looking at these trends, is the point we just discussed.

I would say three things.

One, pause any long-term contracts.

If you’re about to sign up long-term contracts, definitely, I’m not saying don’t do it, definitely take a pause and think about the implications.

So that kind of works counterculture.

Number two, an obvious one, spend time with your IT leaders and business leaders together to prioritize workflows, which can be candidate for agents.

I would look at both external workflows, customer-facing workflows, as well as internal workflows.

And I say prioritize because I think it’s going to be a multi-year effort.

So rather than saying pilots, just do that.

The third one is all about pace, Pete, which is there are things which companies need to speed up significantly.

I can see even as a small company where we’re realizing we’re not fast enough in some areas.

Doesn’t mean there’s a mad rush, but set the pace for the organization in terms of where you need to speed up, where you need to slow up.

Particularly start upskilling your people.

That’s a combination of your people as well as machines.

Because if you can start to liberate the data that’s locked in the enterprise software and start to train the models, you’re going to be in a much better place to make use of all these vertical agents that’s going to come in.

Wonderful advice and on number two, the notion of prioritizing workflows.

I love both of those words.

It’s workflows that we’re talking about, not technology or software solutions.

Start with the workflow.

And the prioritization piece, I agree, it’s going to be a multi-year effort.

And also we can touch anything at the end of the day and make it better.

So we need to prioritize.

Where are the biggest problems?

Where are the biggest opportunities?

Where can we create the most value for customers going forward?

Yeah.

Love it.

Max, as always, wonderful conversation.

Thank you so much for joining and sharing your wisdom.

A lot of fun, Pete.

Thank you.

All the best.

Thanks as always for listening and watching.

Don’t forget to give us a five-star rating on your podcast Player of Choice.

And listen, we’d really appreciate it if you would share this episode with someone that you know would enjoy it.

At the end of every episode, we like to ask one of our AI friends to weigh in on the topic at hand.

Hey, Claude, this episode, we’re talking about how a digital twin or digital chief of staff might help professionals in the AI era.

So what do you think?

I think a digital chief of staff could be huge for busy professionals.

Imagine an AI that knows your priorities and handles routine decisions without constant handholding.

The key would be making it truly personalized, so it feels like an extension of your thinking.

Now, you’re in the know.

Thanks as always for listening.

We’ll see you next week with more AI applications, discussions, and experts.