What does it really mean to build AI agents in the real world, not just as demos, but as tools that save time, surface insight, and change how people work?

In this episode of PROMPTED: Builder Stories, Kyle sits down with Jason Burke, founder of AllStage and an experienced product builder and investor, to unpack how he thinks about AI agents and how he actually builds them. Jason shares how his background in engineering, product management, and early-stage investing shaped his approach to designing agent driven workflows that support founders, investors, and everyday life.

This conversation goes beyond theory. Jason walks through live examples of the agents he has built, including an AllStage business workflow agent that helps analyze investment opportunities, as well as personal productivity agents he created to solve real problems in his own life. Along the way, he breaks down what makes a workflow “agent ready,” where agents add the most value today, and why trust, risk, and human judgment still matter.

You will hear practical guidance on how to start building agents, why experimenting with personal use cases is often the best entry point, and how AI is lowering the barrier for anyone to become a builder. If you are curious about using AI agents to work smarter, think more clearly, or build something meaningful, this episode offers a grounded and inspiring look at what is possible right now.

Learn more and connect with Jason:

1. Jason on LinkedIn: https://www.linkedin.com/in/jmburke/
2. All Stage on LinkedIn: https://www.linkedin.com/company/all-stage-invest
3. All Stage Website: https://allstage.co
4. All Stage in All Stage: https://allstage.co/us
5. Deal analysis agent: https://agent.ai/agent/allstage
6. Company updates summarizer Agent: https://agent.ai/agent/allstage-company-updates-summary
7. Better prices/shopping assistant agent: https://agent.ai/agent/better-prices

Subscribe for more Builder Stories and visit agent.ai to start building your own agents.

2025 to AI is like 1995 to the internet.

There is no going back.

The world is moving in this direction, or at least AI is going to play a role in all of our experiences as humans.

So it makes sense that we all understand how this works.

The challenge used to be just the technology.

Like, could you find software engineers that can build this?

You know, when that goes away and anyone can do that, do those companies need to exist anymore?

And the challenge with the technology is going away.

This is, you know, democratizing things such that anyone

can do this, whether it's with vibe coding or tooling that just makes it super easy.

It does require, still requires, product thinking around how does this stuff work?

What is possible now versus what it would have taken to build that 10 years ago?

It's unbelievable that a fraction of a percent of people in the world could have built this functionality 10 years ago, and now everyone can do that.

I sat down with Jason Burke, founder of Allstage and a lifelong product builder who's been deep in the weeds, building AI agents across investing workflows, personal productivity, and more.

We talked through how he thinks about agents, where they add real value, and where they still fall short.

And best of all, we actually demo several of the agents he's built along the way.

This episode is packed with practical examples, thoughtful takes on trust and automation, and a really grounded perspective on where all this is headed.

If you've been curious about building agents, but want to see what good looks like in practice, you're going to get a lot out of this one.

So let's get into it.

Jason, it's absolutely fantastic to talk to you today.

I'm really interested and I'm excited to get into like the whole product side of agent building, because I know with your background, you've built a whole bunch of agents in the last year in a whole bunch of different industries and using it a lot to kind of pop or build tools for your startup, Allstage.

So would love just to kind of start the conversation is like, how are you thinking about shaping and building these agents and how has that evolved over the last year here?

Yeah, so Kyle, you mentioned that I am building a bunch of agents, one for my business.

So just at the highest level, I'm the founder of Allstage, which is an invest tech platform.

So we're building products that are at the highest level, improving the way that early stage investing and fundraising works and enabling that collaboration between those investors and the founder and vice versa.

We had started about two years ago incorporating AI into various bits of the product to create incremental value.

And earlier this year, I was seeing some of Dharmesh's posting about Agent AI.

That was the first time I'd heard about it.

I had a conversation with Mike Redboard on the team.

And frankly, my motivations for starting to build agents was selfish.

You know, I'm a product builder.

I'm also an investor.

And

most companies are incorporating AI into their products.

So as an investor, I wanted to also be able to speak that same language and understand what agents meant.

And, as a builder myself, I also wanted to, get my hands dirty with this.

And I'll admit I'm also a sucker for being biased towards local companies.

Agent AI is Boston Built.

You know, I'm dating myself, but

in college, my search engine of choice was Lycos because they were based in Waltham, which is where I grew up.

So that gave Agent AI a leg up there.

But yeah, this is, this really is not the future, but everything is moving towards AI being a piece of the puzzle.

So this is, by building these agents, helps me

be able to speak that same language and create value both for my company at all stage, but also we'll talk a little bit about this, how I'm building some agents that I'm using in day-to-day life.

So you kind of mentioned your journey of like figuring out and learning what these agents are.

Like maybe help the audience.

Like how do you define an agent?

What do you think about it when you break things out of like, all right, this is something I want to, as I like to say, identify.

And

what goes into that?

Because in a lot of respects, it's kind of what we thought of as apps before, but it can do much more complicated things than an app could do easily, right?

But how do you think about agents and when do you decide to deploy or build an agent?

And how are you kind of leveraging some of them now?

I think what you're touching on is the fact that, yes, there are many shades of gray.

Is this an agent or is this just an...

application that uses automation.

And we're in a very fuzzy zone right now.

But at the highest level, an agent is some software feature that uses different tools and takes action to complete a task.

And a lot of what we're seeing in these early days is, you know, automation that otherwise might have been done with scripts or systems like Zapier to sort of plug together different links in the overall chain.

to take an input and then, ultimately generate something that comes out the other end.

But we'll see this continue to evolve, such that agents are, there'll be orchestration agents that are ultimately completing some task, but they're using other agents to do specific sub-tasks along that chain.

And

I think we're a ways off before this becomes really common, but a lot of the dream of these agents is to, for these agents to be making decisions on their own.

We'll talk a little bit about why that is a challenge to do, but we're already seeing a lot of value in an agent, which is a tool to create, to deliver on some task and complete some tasks.

We're already seeing that happen.

But, as things move forward, seeing that agent do more things versus create, versus just being a creator?

With agents, there's certain things that it can do and it can not do, or it can do easily and it can not do easily.

Do you have certain guides or rules when you said, hey, this is something I can agentize or something like, no, it's more of a traditional software thing?

Or

It's just not ready yet.

But do you have any things that when you look at a problem or a workflow, they're like, this is something that makes perfect sense to like build into an agent versus no, we're not touching that with this yet?

Some of the indirect value that folks are getting out of agent platforms is the connectivity and the plumbing that would otherwise be required to connect into LLMs or to

automate or, to do things like, for example, suck down Kyle's LinkedIn profile and identify some of his, key skills.

These are things that, yes, could have been done in the past with code, but some of these platforms that are being

that are being created like Agent AI to help people build agents, they're just removing that.

So it's an example of just removing the friction.

So that's not really, that's not agentic, but it is a barrier for entries being lowered.

And what that means is that it doesn't limit the creation of agents to technical people like myself.

Now anyone can do that.

So that's one big benefit.

But when you start to think about how, one of the qualifiers for agent, one of the adjectives is AI agent.

So how can it use AI and what can it do with AI?

So a lot of what we're seeing come out of the agent world now is

leveraging the thinking and preparing of data and analysis and outputting that.

And that is really valuable because these are doing things that otherwise either would have created a lot of, or would have required a lot of code and limits to who can actually build those things.

And it's also allowing you to do things that may have been too complex to do with code.

And that's really, really valuable.

And in today's day, it is,

It is limited in action and it's or it's being cautious with action and action meaning the agent itself is doing something.

So, you could imagine I talked about thinking and preparing data.

So let's say you had it analyze your exercise log.

You exercise and you use something like Garmin and it could, you know, provide some analysis that shows that you tend to run

quicker on Mondays than on Fridays.

And now if it started to order food for you on DoorDash based on how fast you were running, that's an action that it's not a fatal consequence if it gets wrong, but maybe it is, it might not be right, and now you have a bunch of stuff in your refrigerator that you can't eat.

So that's just, a.

Bunch of bills you don't want to pay for.

A bunch of bills you don't want to pay for.

But we will start to see that,

there's also a matter of trust that has to be created.

And I think this is, we're in a time where this stuff is new and human beings are averse to risk.

But as we become more comfortable with this, we'll allow these agents to, we as users of agents, will allow them to take action and trust them to take action.

And as a result, we'll see builders start to

go beyond building agents that simply look at data and output data, but will also have an opportunity to take action because the users will trust what the agent is doing.

Oh, and I think what I hear you talk about that is a lot of this concept we call human in the loop, right?

Like, and everybody's like, no, we have to have humans in the loop kind of sitting at the top of the pyramid.

Agents are doing this stuff.

But what I'm hearing you say is like,

We're going to get to a point where we trust these things more and more because the success rate or the hit rate of them knowing what we want them to do, 99.9%, 100% correct.

So like, why do we need to like watch over them if they can see that I have no milk in the refrigerator?

Just go order me some darn milk, right?

Just do these sort of things and free us up to not worry about it.

Is that like, I think that's what you're, but also like, how long do you think before we get to that point?

I do think there is, there's also a spectrum on the risk or reversibility of an action.

So we are at a point where folks are, there are folks that are trusting agents to take that action.

So, if the agent does,

recognize that, there's no milk in Kyle's fridge and orders it, and the reality is that there wasn't extra milk behind the apples.

It's not the end of the world.

Yes, it stinks.

You know, that was $6 down the drain, no pun intended, but that's not that's not fatal, and we're we're we're starting to see that, you know, that that's a that's at one end of the spectrum, but if you could imagine a different action could be

Again, going back to the example of an agent that is analyzing my running logs and making, giving me some analysis, no risk involved.

Maybe it's telling me to go on a long run on Saturday, not too much risk if I'm a little bit tired.

But imagine Jason, who is a type 1 diabetic, is being told by the agent, increase your insulin level, that

is dangerous.

So this is, I do believe that, as it relates to benefits to human health, using that example, I think there are huge, huge opportunities there, but that's at the other end of the spectrum.

So I would say that right now we are in a place where there is trust.

There is trust to take action, but there's a long spectrum of the

the risk profile of different actions.

And we're just at the beginning of that risk profile.

Over time, people will start to trust more and more.

Yeah, there's like a whole extra or extra variable there that needs to be applied to all these scenarios, like a trust matrix, right?

Of like, how confident are we in like, one, what is the likelihood of something going wrong?

Does that hurt us?

How much does that hurt us?

But also,

how, is the juice worth the squeeze?

I guess it's kind of an analogy.

Like if it's not big enough and who cares if it's wrong, go ahead and keep moving forward.

But there's certain things that become big enough and important enough.

I don't want you doing that.

I don't care.

Whatever it is, AI, we don't want you doing those things.

Well, we've talked a lot about those kinds of things.

Let me flip it a little bit.

Like, what are you seeing now?

because it's lowered the barrier to do more and to be able to accomplish things easier that were complicated or needed engineering talent to do before.

What does it still suck at?

I mean, the bad joke I always say is like, the LLMs still suck at math.

They can't seem to do basic math sometimes.

But what are some of those challenges and things that like we're just not going to solve in the near future and we shouldn't even try to build agents around?

Do you have a category of those that you just stay away from?

Well, at the highest level, AI is

Stochastic.

So it is by design not going to be consistent in its outputs.

And this is by design to help it find the right answers.

And it's, you know, this is really deep, you know, computer science reasoning behind this.

But at the highest level for any user to recognize that when you ask one of these models a question, you're not always going to get the same answer.

So as an agent builder, you have to take that into account and recognize that, certain things, let's say you're building an agent that is going to give you some analysis on your exercise logs and you want it to generate a web page that has a certain design.

Oftentimes that design will change from run to, you know, from execution to execution.

So keeping that in mind, there are ways around this to infuse some

consistency, but it takes some work.

But I think that's one of the higher level things that people need to keep in mind.

But this higher level consideration needs to be, is really important as you start to, as we were discussing, trusting the agent to take action.

Because if there is some inconsistency, and this goes beyond simply displaying analytics to Kyle,

but it's actually taking action.

And what it does, what it, the action it takes today is different than the action it takes tomorrow.

even though the data was the same, the inputs were the same, that can have negative consequences.

So I think that that's one of those, you know, it comes back to the risk profile is that, you know, are, is this variability in the outputs something that, something where a mistake is not

I keep using the word fatal, but it is not damaging versus, is this stochastic nature of the responses something that could cause problems?

I'm curious now, because as you're talking about this, I'm sitting here thinking to myself like, all right, I get it.

Like, you ask anybody a complicated question, they're never going to give you the exact same answer over and over and over again, especially if it requires some amount of lengthy output, right?

So how does that work?

Maybe this, maybe you don't, but I have to ask, is there some like random number generator or RNG behind this that's like making it tweak and change the output?

Or is it more just this nebulous thing that all of these billions of data points kind of just don't always congeal the same way every time?

Do you know?

I don't know.

I don't know way above my pay grade.

I just tried that.

But I think the reality is that, and this is part of the secret sauce behind a lot of the foundational LLMs, is that they are using billions or trillions of data points.

These data points are changing all the time.

And the other

The thing is that you do start to see some continuity, not continuity with the outputs that we're talking about, that there is variability in these outputs, but there's continuity.

And then this is one of the really impressive bits of that we're starting to see come to the agent world or to AI in general is that it does start to remember what I've asked it.

So when I do ask it the question, where should I

go on a run this morning.

It is going to answer something different today than it did last week because it knows, I'm making this up here, but it might know how much I slept during the week.

It might know what my, you know, what my schedule is next week.

And this continuity, it's a continuity of understanding and memory.

So this is really powerful.

So these things are coming together at the same time, this just underlying variability of AI and just

the nature and just the way this magic works.

But alongside, we're layering on top memory, or these models are starting to remember what, Jason has asked and what he does and, how much he's run in the past.

Kyle's going to ask the same question and he's going to get a different answer because it has different context about him.

So just like the exact same way that either you or I asking a friend,

asking the same friend for an answer, that friend is, she's going to give you a different answer than me because she has different context about us.

Yeah, that's so true.

And there are ways we can go into the LLMs and like intentionally get that out, right?

Give it personas and roles to play for output.

But it's interesting too, because as I think more about that, it's like, it's such, it's building these mental models, right?

Like very similar to how humans do.

It's just, I don't know, I don't really have a question.

It's just fascinating how it's getting closer and closer to replicating what we do, but yet, I don't know, I'll throw it out there.

How close do you think we're getting close to kind of this, you know, artificial general intelligence or we still haven't cracked that code yet?

I think we all think that AI has

progressed so far, and it really has.

But I'll use that overused analogy.

We're just in the first mile of this marathon.

We, 2025 to AI is like 1995 to the internet.

You know, in 1995, the world thought that, you know, what else could be done on the internet?

Everything is right there.

And obviously, that has continued over the last 30 years, more and more enhancements to the internet.

We're in that world with

with AI as well.

And I do think that AGI is, there are going to be many shades of gray with respect to AGI, but ultimate AGI, which scares a lot of people,

perhaps myself included, is way, way out there.

But we will start to see, a timeline, we'll start to see, we were already seeing this happen, and dribs and drabs with AGI becoming, you know, becoming a real thing now, but this is a, there's a lot of things that can be done there, and none of us can predict what that, you know, what that ultimate

ultimate story looks like.

Yeah, I mean, who knows?

Like, anybody that's 100% certain of anything, I'm 100% certain they're wrong about it at this point.

But it sounds like the thing that I'm taking away is like, it's not going to be a flash point.

It's kind of what I'm hearing you say.

It's probably going to be these different, these different things that happen that kind of enlighten us, kind of like a

a kid just doesn't, a baby doesn't all of a sudden become, an adult or they go through these stages and one they're walking and next they're feeding themselves and these different inflection points happen.

Maybe not always in the same order for every person, but like we'll see sparks of it.

Is that kind of what I'm taking away from you?

Like we'll see sparks of AGI and it won't happen all at once, right?

Yeah, whoa, that's I think that's the perfect example.

And that's when you'll start to see some of the

the different stakeholders in that value chain, the builders start to take advantage of that.

You'll start to see some of the vendors, the foundational LLM models and the real companies that are building tooling on top of that take advantage of that.

Once, you know, again, trust is a piece of this puzzle along the whole way.

And once the end user, whether that's an actual person or a business, once that end user starts to trust that this stuff

works and can drive value, then they'll be leaning forward more and the builders will start building towards that.

The vendors will start investing in expanding what they support for it.

All right, do you want to get us backgrounded a little bit?

And we kind of teased before we hit record talking about some agents that you'd built that we wanted to kind of show off.

So you want to come back to some real-world practical examples and kind of show everybody what the all-stage

business tool agent does and kind of how it works and kind of talk people through a great example of what a powerful agent could do.

Yeah, absolutely.

So just a little bit of background on this.

And it's one of the things that I learned in building these agents.

And it's a core tenet of product management, which is start small and then expand from there.

So what you're going to see is, you know, the first agent I built, which has a lot of things, a lot of functionality within it,

that agent on its own is much bigger than anyone should build.

But what this agent, what we'll see is that it's made-up of a bunch of what I call sub-agents, these agents that are doing specific tasks.

So at Allstage, Allstage is an Invest Tech platform that's helping investors evaluate investment opportunities and help founders share their company.

And

If you're a founder, using founders as an example, if you're a founder, you want to be able to position your company the best way so that it's attractive to investors.

And it's sometimes hard to do that.

You're spending a lot of time putting together a pitch deck, competitive matrix, an overview of the team.

And there are ways that you can be using agents to do a lot of that work that brings value for both sides.

So I'll show what this agent looks like.

It was very

very relevant to what we build at all stage.

So it sounds like as you're building, it's a little bit of, eat your own dog food to kind of like, this is actually helping you think and strategize around, your own thought partner.

And it's like, let's scale that into something that other people can use.

Is that the, is that kind of the right way to think about that?

That's a good way to think about that.

And it was also, you know, one, can we drive incremental value beyond the core products?

Two,

Some of these pieces that you're going to see within this agent are items that Allstage could have built and incorporated in.

But to my point earlier, it would have been a lot more work to do some of these things and build them into code versus leverage

a tool set, agents aside, a tool set, in this situation, it's agent AI, a tool set that naturally has all the plumbing, the work to take the underlying query essentially, and ask the LLM for an answer.

And so we'll see what this looks like.

But that was, that was, that's the other benefit.

for Allstage as a product company, how can we be quicker to market with driving value?

So this deal analysis agent can be used by a founder or by an investor.

So a founder, you know, let's say we're a founder and we want to analyze our own deal.

So what we're able to do here is plug in a code that we can pull out of our deal within Allstage.

a deal within all stages is simply a picture of the fundraising terms, the data room, a summary of the deal, the team makeup, and.

Like a snapshot of the company, right?

That you're trying.

It's an improvement to sharing a pitch deck.

So I could plug in the code I can grab from my deal and

and run that, and here's the output.

So this happens to be the output for a company called Ping.

It's an amazing company out of Massachusetts that is changing the way that coffee gets made, and you can order it through quick service means.

But what you're seeing here is the output of this agent that is taking in as an input information from Allstage.

It's using some

some APIs over to Allstage to grab a bunch of information.

And then the agent is doing an analysis of the market.

So using all that information, do market analysis.

This is something that every investor is going to ask the founders of Ping when they're evaluating this opportunity.

They're also going to ask who that competitive set is.

So the agent is

has a sub-agent that is going off using that information to put together a list of competitors, and here's a competitive matrix.

And these are things that would otherwise have taken the company a lot of time to put together.

And frankly, these are some of the things that technology is better than humans at.

Looking up information and finding comprehensive information would either take a long time

for that human being and or that human being would miss certain things.

The agent is doing diligence, given all the information about the company, come up with some diligence questions.

So these are questions as a founder of Ping, I can look at this and say, you know what, these investors are probably going to ask me about unit economics here.

So this is previewing what an investor might ask.

And then finally, some recommendations.

So this might be something as an investor giving possible upsides and downsides.

These are the things that an investor or a VC firm prepares when a partner is putting an investment opportunity in front of the other partners within the firm is, you know, this is the upside, this is the downside, these are the risks involved.

So what you're seeing here is a bigger agent that is using sub-agents that are doing individualized tasks.

This agent, give me a market analysis.

This one, analyze the competition, and it's pulling all of these together.

These sub-agents are also able to be deployed individually.

So within the all-stage product, our core products, we can put a market analysis within the deal within

Inside of Allstage, so reusability is also another benefit of what you're seeing here.

And Jason, I'm curious just to kind of help people understand, because obviously all the...

listeners here aren't like, founders or investors, but how much time does this save, like a founder and what is the expectation, like time it takes, but how many people do they get involved to kind of pull this information together?

Because typically this takes like weeks or months, right?

It takes weeks or months.

It's also is often a living and breathing document.

So usually this information is in a pitch deck.

One, the pitch deck might not be the best

tool to be using, just the construct of a deck that gets sent around, it's a static piece of content.

And when a new competitor enters the market, they have to update that slide.

Oftentimes, that pitch deck is static, they create it and it never gets updated, or they're constantly updating it, which means now you have a bunch of, you have a whole bunch of

pitch decks, all, a lack of consistency with different investors.

But to answer your question, yes, this is a lot of work.

And it's one of the reasons why fundraising for companies is sometimes described as just a necessary evil.

Yes, we need to raise money because that is what will pad our bank accounts so that we can hire, so that we can pay our bills, so that we can grow.

But it's a time suck.

It takes those founders off of strategizing and building the business over to, let me gather this information.

So the time savings and the benefit and or lessening of opportunity cost is gigantic.

And then on top of that, the agent and technology and product and data and the automation of this can probably

prepare information that is better than what the human can do.

So yes, there is a human in the loop.

When the Ping team looks at this, they may recognize that, you know what, there actually is another competitor in here that the agent didn't get, and they can pull that, they can inject that.

So there is a human in the loop, but the agent is also probably identifying things, identifying diligence questions or identifying a risk

that the humans might not.

So there's a lot of benefits to this, to both sides, to both the company as well as the investors.

I want to dig into that a little bit.

I imagine there's two pieces of that for, especially from a founder.

Like there's the, I don't know what I don't know piece, but then there's also the like, I didn't think of everything piece.

And this is kind of, this is kind of handling both of those for them, right?

It is, it is, you know, using using a different

a different agent to answer that question that I use, one of the personal agents that I built is an agent for product managers.

So as a product manager, you think about your product 24 hours a day.

So from the outside looking in, you're like, well, she's a product manager.

She should know everything about her product.

But all of us as product people don't

we miss things.

We're human beings.

We're not thinking about everything.

And the agent, the agent I'm referring to was creating user stories, given the fact that I am the founder of an InvestTech platform.

We're looking to remove a lot of the friction involved with early stage investing and fundraising.

I'm trying to create a product that is playing matchmaker, as an example, between investors and founders.

I have some story, I have some ideas on what some of those features are, but I'm not thinking of everything.

And that's an example, you know, that agent is helping fill in some of those blanks.

It's helping

add to what I as a product manager or the rest of my team might be thinking of, but it's also identifying other things that we're not thinking of.

Or it may identify some risks involved with some of the ideas we have.

So that's an example of 1 where, yes, we don't, we as human beings don't know everything and tooling, you know, through agents can help, you know, help fill some of those blanks.

Yeah, bake in all those best practices just so we don't skip or miss something.

I'm curious, Jason, with like this, companies like Ping, what has been the impact?

Like, clearly, this is speeding things up.

It's freeing them up to focus on the business and not all this documentation and materials that investors want to see.

But like, what has been the feedback you've gotten from the market and customers?

What have been some of the results?

And clearly, you've made a lot of iterations as you've gone.

Like, what have you kind of learned from that as you've gone down that path?

There is a real demand for innovation in this space, and it's one of the reasons why I started Allstages that I was, I had co-founded an angel network and was having trouble finding products that would help us deliver on our vision, which was enable collaboration amongst a bunch of very smart investors and recognized that there was a lack of innovation

There were a lack of products that were helping us or to help us deliver on what we wanted to accomplish.

So how do you solve that problem?

You can either pound sand and cry about it, or you can build your own thing.

So that was how Allstage was born.

But founder, entrepreneurs have been facing the same thing on the fundraising side.

They've, the construct of a pitch deck has been around, been around forever.

And

There hasn't been much innovation.

So the feedback on something like this, which is, you could argue whether this is a better view of the opportunity versus a pitch deck, that you could argue in either direction.

But the real key thing is the most valuable thing for an entrepreneur in the early days is their time.

And if you can reduce the amount of time they need to complete a task, that's a big win.

But also,

it's really important.

It's a very competitive market out there for fundraising.

So anything you can do to evangelizing your opportunity or fill in any blanks that may be of concern for a, answer any questions that may be going around in the head of the investor is a big win.

And entrepreneurs have unfortunately lacking innovation in this space.

So there is a lot of interest in

movements like this.

I imagine a lot of speeding up the decision making.

But to your point earlier, like the entrepreneur, like they don't necessarily have a big team, right?

Time is of utmost importance.

And I imagine the more time they can spend solving the core problem that generates revenue is the optimal use for their time.

And not saying this isn't important to like extend their lifeline to do that, but like

Yeah, it gets them to frees them back to focus on the priority where they could still meet these other objectives and check these things off to provide the information needed.

But they're not necessarily an expert at creating a pitch deck and all this, and these tools can help them do that.

I imagine that's a huge win.

So we've talked about super powerful chained agent here, doing a lot of stuff that kind of started as an MVP and it's kind of expanded into being able to do a lot of stuff here.

How did you go about being able to do this sort of complicated stuff?

I guess where I'm leading is kind of talk to me about some of the play that you do, some of the fun personal agents that you've also built to kind of solve personal problems.

And how did maybe some of the ways that led into, I experimented with this and it was able to lead into some of these bigger things that I could build with that?

Yeah, one of the things I advise a lot of folks

whether you live in this world of product development or you're just a human being like all of us is, there is value that can be, that is being created through agents and products like this.

So one of the things I often encourage folks to do is, well, one, there is no going back.

This is, you know, the world is moving in this direction or at least AI is

is going to play a role in everything, all of our experiences as humans.

So it makes sense that we all understand how this works and also figuring out how we can help ourselves with this.

So the advice I give is start to play around with this.

Start to get your hands dirty and do this by focusing on something that is part of your life.

what is a task that you do all the time that either, takes a lot of time or, maybe you do this task and you realize, what, I know I'm not doing this perfectly because I just don't have all the information I need.

and solve a problem that is just part of your life.

And it's much easier to think about that than building for a business like what you're seeing with this Allstage agent.

For me personally, the first maybe 10 or so agents I built were for Allstage because I was, I happened to think about Allstage 25 hours a day.

So this was, yeah, this is my startup.

And it was creating value for Allstage.

But I started to recognize, all right, there's some value.

Like I could take the same approach and be building things that were helping solve problems in my life.

So this better prices shopping agent was something that came up when I went to the Apple Store with my 12-year-old son to get a new MacBook.

And

That's why I said, if you find a better price in the next, several months, then we'll match that.

So I told him, if told Simon, if you can help, I want my kids to know, about this stuff.

And I said, if you can help,

create an agent, you and I can work together that will monitor prices.

Because if you think about what you have to do to find a better price, you're, searching the internet all the time and it's a torturous process.

And chances are you're not going to find it.

But this is something that an agent is great at.

One, the agent doesn't.

And that's part of the bet they make, right?

Like they tell you that because they know

you're not going to go spend the energy and effort to catch them on that.

But AI could.

Exactly, So that's an example of 1 where AI is giving you incremental value beyond what you, what, you know, one, helping you do something you

couldn't do, or if you could do it, if you never slept and you did look around, it would take you a lot of time.

So that was the origin story for this is, you know, Simon, let's, one, let's work together to do this so that you can see how this stuff works.

And then two, if we can build this, we can, you know, maybe we can find that cheaper, that, you know, cheaper MacBook and, you know, get that money back.

And I was going to share some of that.

share some of that savings with him.

So, with this agent, very simple, and I think this is one of the one of the things that all agent builders should think about is make it super easy.

So, plug in the name of the product.

So, I, for this example, we'll use, I just bought an espresso coffee machine and the, I have to buy pods, and these

I want to be able to buy these for less than what I might be able to get retail.

So, that's what my task is.

I'm trying to buy a product and I want to find it for, anything less than $65.

So I can say go and now the agent is going off and what it's doing is it's, looking up what this product is.

It's trying to figure out, this happens to be coffee.

the user is trying to find it for less than $65.

And let me give back some results and come over here and see what this looks like.

So this is the output of the agent.

All I did was plugged it in and got some cheaper options than $45.

So now I can bounce off, maybe purchase it on the website.

It's also giving me alternate options.

Maybe the Nespresso pods are too expensive, or maybe my Nespresso machine broke, and now I can see what some other options are.

It's also giving me some related items.

So if I like coffee, then maybe I also like tea.

And so these are the sorts of things where if I was just doing this on my own, one, it would have taken a lot of time.

Two, I wouldn't have been thinking beyond just I want to buy this particular product.

From an agentic point

with you.

We were talking earlier about providing data versus taking action.

So what you're seeing on the screen is just providing data.

So it's valuable data.

It went off, it did a bunch of work that would have otherwise taken me a lot of time, and I would have missed a bunch of these things.

And it's providing data.

No risk here.

It's just showing me some stuff.

But now I can also plug in my e-mail address back at that previous screen and have it e-mail me when it finds something less than.

my gold price, and take that one step further.

At some point, maybe I'll say, what, if I can find it for less than $40, make a purchase.

Make a purchase for me, and here's my credit card information.

So that's how you can start to see how these agents can progress from a functionality point of view to first, step one, what you're seeing on the screen, make it easy for me to find this stuff.

Step 2,

ping me and let me know.

So there's a human in the loop.

It's telling me that it just found it for $45 on some particular website.

Now I can go over and purchase myself to ultimately let me provide my credit card information and when it finds it for a certain price, just automatically buy it.

So these are the sorts of things like we all have these stories in life around things that we're trying to do and we do all the time.

But when you start to open up your brain to what is possible with technology, we can improve, we can improve the way we run that part of life.

I'm just looking at this thing and thinking like, you could build a whole business around this.

could be a product that I could see a lot of value for a lot of people, right?

And I love the whole alternative and related because you're kind of...

I threw it out there earlier, but you're literally answering here, like, I don't know what I don't know.

What am I not thinking about that I could buy instead of this?

Which is really cool too.

But yeah, like if you've got these repeat things that you know you need to buy instead of just Amazon Pantry or Subscribe and Save, you could really try to find deals on stuff.

And gosh, you also got me thinking, if I know I need to buy this about once a month, and if it finds a deal and it picks it up now, well,

It could have its own queue in there.

It says, don't buy anything for five weeks now.

So you could have, you could simply build all of that in here.

But I'm curious, like, how long did it take you and your son to, Simon, to throw this together and build this one?

Yeah, I mean, this too was a phased approach.

It's like, first, let's just build this to simply search.

You know, you could argue that you could do that same thing with the Google search, and you could.

But, you know, Google might not

be returning, might not be providing all of the options for, what Simon and I were building for was searching for that MacBook, but might not be looking across all of the sites.

And the agent is now able to do this across multiple different places.

But that was step one.

And then, start to extend that to, given this description of the product, what are some alternates or what are some related items?

So it was a phased approach, but for that first version, this was a, it was an hour-long project.

And an hour-long project that Simon and I worked on, iterated back and forth, adjusted the interface, said, you know what, I only want to see prices in US dollars because we, we're here in the US.

But those are the sorts of things that, just fine-tuning it and ultimately presenting that information to the end user.

was a really quick thing.

And again, if you think back to, if we flip the clock back 5, 15 years, 10 years, or even last year, this sort of stuff would have just taken so much longer.

And it wouldn't have been something that could be done by a 12-year-old who's not a software engineer.

So clearly this has changed A lot.

What have you learned along the way?

What are some of the, like you mentioned some of the iteration and stuff, what are some of the biggest failures?

And kind of what did you learn from that about what to do, what not to do?

How has it changed the way you're thinking about building these things now?

So we talked about the variability of these outputs.

And you're seeing this, you're seeing this on the screen with the output of the search for the Nespresso pods, is that there is variability in the outputs and a UI may change from time to time.

And this can start to be,

with a tool like we're showing the agent for finding prices, it's not the end of the world if the font changes from run to run.

But if this is a core product within a business, that consistency is really important.

And that's one of the things that you start to need to build around.

And, you know, there are approaches to address that, you know, providing a

template code, this is what I want it to look like, versus a vibe-coded description of a, I want a tabbed interface that shows cheaper options, alternate options and related items.

Because that Stochastic AI is going to just take your instructions and say, I'm going to provide tabs.

and not necessarily, it may, provide different outputs from time to time.

So then that's one of the things that you start to realize and accept.

Or extra work to get the consistency, the repeatable consistency is what I'm hearing you say.

That's exactly correct.

And then that's just when the UI, you also get that on the outputs as well.

You know, you don't want to have different, completely different outputs on some of the stuff.

The other thing, I mean, you joked earlier around LLMs are bad at math.

That problem will be solved soon.

But LLMs are also bad at providing URLs.

So in that

product search agent, it provided a link to, for the cheaper options, it provided A hyperlink so that you could bounce over to the website to buy those things.

A lot of times you'll find that this link is broken or this link is out of date.

Or hallucinated.

Or completely hallucinated and it completely made-up one.

So, you know, people often like to describe the AI as that

that over-aggressive intern that is sometimes sloppy, super smart, but sloppy, I think it is a very good way to describe things nowadays.

But we are living in a world where the progress with AI is living in, you know, they talk about dog years, these are dog weeks.

The speed at which this stuff is moving is incomprehensible such that these problems we're talking about, you know, here in mid-December of 2025,

maybe solve by February of 2026.

But, that's the way of the, today's world.

What do you believe is possible now that you didn't think was possible a year ago or kind of like even playing this out of the future?

How do you, like, do you have any bold predictions of like how we will be using these things?

12, 18 months, can we even go two or three years out?

You know, kind of feeding off that last comment.

What Dharmesh Shah said at, I believe it was,

inbound about 16 months ago was that we all, everyone that was in the audience, be using AI agents, but we'll be building them as well.

And we will see more and more of that.

will come with trust as everyday folks start to use these agents.

And people won't even realize they're using agents.

It'll be like the internet is, or be like your mobile phone.

You just open it and it works.

And it's agentic technology behind the scenes.

But

The really powerful thing now is that everyday folks will be able to build these things and not just build the, agents.

There'll be some folks that build agents that are just doing very fun things like, you know, bring me a tic-tac-toe game that I can play.

but creating agents that are improving the way they run their lives.

And again, if we look back to, if we compare what is possible now versus what it would've taken to build that 10 years ago, it's unbelievable that a fraction of a percent of people in the world could have built this functionality 10 years ago, and now everyone can do that.

So I think that's one of the things we'll see is we'll see a lot more builders.

Let's go a little bit reckless prediction here, because I'm hearing you say it, and I'm also sitting here thinking like, how low does that lower the bar to software development where anytime we need something, we don't need to buy it anymore, we just go build it ourselves.

You know what I mean?

Like, yes, there will be some core platforms that still exist, but if we need something to go,

I don't know what we want something to be a call transcriber and listen to calls and give me a call transcript.

We don't need to buy a solution for that at some point in the near future.

We'll just code our own that has its own specific need and output from it and can plug it into our own system, right?

How fast do we get there?

And that seems like that's going to be rather disruptive to a lot of things, right?

Building this stuff, the challenge used to be just the technology.

Could you find software engineers that can build this?

And the question you were asking is, when that goes away and anyone can do that, do those companies need to exist anymore?

And the challenge with the technology is going away.

This is democratizing things such that anyone can do this, whether it's with Vibe coding or tooling that just makes it super easy.

It does require, still requires product thinking around how does this stuff work.

Just because you can create functionality, mash it all together, does not make a good product.

So that said, we're seeing more, we're seeing AI come to the product space to do the more subjective strategizing on, you know, what should we build and how should it be done.

So I think the scary thing for product people like myself or software architects is that those jobs are still needed today.

But what is the date in the future when a

AI is doing some of those much higher level things, not only just the advanced plumbing and optimization of which tools to use that a software architect would do, or strategizing on the direction a company should build and how to prioritize the roadmap, and also use sort of regulations on what are the risks involved with using data and that sort of thing.

And as this stuff continues to progress and

your question earlier around AGI, what happens when software is able to do those things that was only possible through very experienced people?

It's not tomorrow, it's not next year, but it's not several generations away.

It's crazy fun, exciting things, but there's a little bit of scariness in there too because of the uncertainty for sure.

Chase, this has been fantastic.

Let me

what's the best way for people to connect you?

How could people help you and the community and the listeners out there?

What can they help you with?

Yeah, so I mean, I think about ways to improve investing in fundraising.

So for all the entrepreneurs out there, check out Allstage and you can set up your deal and be able to take advantage of not only the products that are helping improve the experience when you share your opportunity with investors, but also take advantage of some of these agent features that we've started to layer on top.

I'm accessible on LinkedIn, do a lot of the posting I do to announce things with Allstage, I do through my personal LinkedIn, but you can find me on LinkedIn and love to engage with everyone who's building something as a founder.

Also those that are investing in those companies since I sit on both, I'm not only an investor and a founder, but also building product that's bridging those worlds.

I'll make sure we get all those links in the show notes and

Kind of in closing, what did I not ask you that I should have, or what have we not talked about or shared that you want to kind of leave everybody as kind of a final kind of parting thought?

We touched on this earlier, but one is everyone should try to get their hands dirty with this stuff.

We are early, and even some of the tooling that exists out there, platforms like Agent AI and

and other agent building tools, they're in their early days as well.

But that said, that is no excuse to sit on the sidelines and wait.

If we all sat on the sidelines and waited until perfection in e-commerce, we would still be sitting on the sidelines.

There was always going to be an evolution in the tools and the functionality

But that's no excuse to sit on the sidelines, get your hands dirty and start building things.

And like I said, the way to think about this is just figure out something that you do in your life, whether it's an everyday thing or maybe you're planning a holiday dinner and you need to put together a menu.

These are the sorts of things that you don't have to think about

You don't have to think really outside of the box.

What do you do all the time?

And how can I take advantage of technology?

And we all need that aha moment of, wow, I didn't realize that this could help me with what I do.

And we all start to benefit as everyone starts to use these tools.

And I think you even inspired me a little bit, like, hey, what can I go build with my kids?

Right?

So anybody out there listening that also has kids, like,

Get here, talk to them and y'all figure out a project to build together and learn together.

I think that's a really cool way to kind of bring them along and share that shared experience in something that's going to be an even bigger part of their future.

Yeah.

I think the benefit is bi-directional in that example, in that old people like you and me will benefit from some of the outside the box thinking and perspectives of kids.

So, but this is the future for adults and for everyone on earth and everyone should start to become comfortable with this.

Well, Jason, thank you so much for taking the time to have this conversation.

I hope everybody got something from it.

And until next time, everybody, keep building, keep learning, and go out there and build something with your kids.

Take care, guys.