In this episode of Innovators Inside Podcast, Ian Bergman sits down with Andrew Brooks, founder and CEO of Contextual.io to explore one of the biggest challenges facing companies adopting artificial intelligence: the gap between an impressive AI demo and a system that actually works inside a business.

Andrew calls this gap the AI hype tax.

Companies can give employees access to ChatGPT or Copilot, build a prototype over a weekend, or automate a single task and quickly feel like meaningful AI transformation is underway. But turning those experiments into secure, scalable, reliable business systems is much more difficult.

The conversation explores what production-ready AI really requires, why full automation is often the wrong goal, and how founders, operators, and innovation leaders can make better decisions about where AI creates real competitive advantage.

Here are the five key takeaways from their conversation:

1. A Great AI Demo Is Not a Production-Ready AI System

One of the easiest mistakes companies can make is assuming that because AI can perform a task once, it can reliably perform that task across an entire organization.

A clean demo might involve giving an AI model a document and asking it to extract a few pieces of information. In production, the reality may include hundreds of different document formats, incorrectly rotated pages, missing data, unexpected attachments, and thousands of other variations.

That difference matters.

Production AI needs to work across the messy reality of business operations. It also needs governance, security, scalability, monitoring, and a clear process for handling exceptions.

As Andrew explains, even a system that performs correctly 99% of the time still leaves a critical question:

What happens to the remaining 1%?

For leaders evaluating AI projects, the goal should not simply be proving that an AI use case is possible. The more important question is whether it can operate reliably under real-world conditions.

2. AI Transformation Requires More Than Giving Everyone ChatGPT

Many companies begin their AI strategy by rolling out tools such as ChatGPT, Copilot, or AI assistants across the organization.

These tools can create significant individual productivity gains, but access to AI is not the same as business transformation.

Andrew describes another hidden cost of AI adoption: employees can suddenly produce far more content, reports, documents, messages, and analysis. But if other employees then need AI to summarize all that AI-generated content, the organization may simply be creating a new layer of inefficiency.

The real opportunity begins when companies move beyond AI as a personal assistant and start redesigning workflows around it.

That means asking:

• What business process are we improving?

• What data does the AI need?

• What happens when the system is wrong?

• Who is responsible for the outcome?

• How will we measure whether the system is actually creating value?

AI adoption should ultimately improve how work gets done, not simply increase the amount of content a company produces.

3. The Best AI Systems Amplify Human Expertise

The conversation also challenges the assumption that successful AI transformation must lead to full automation.

Andrew argues that in many cases, the better goal is to give employees what he calls superpowers.

One example involves a company managing maintenance for forklifts and other warehouse equipment. Employees previously reviewed invoices manually to identify potential savings for customers.

With AI, the company can analyze every line item, compare historical costs, review vendor rates, and check contract terms at a scale that would have been extremely difficult for a human team alone.

The employee is still responsible for the customer relationship and the final judgment. The difference is that they now arrive with far more intelligence.

This is an important framework for AI leaders.

Instead of asking:

How can AI replace this person?

Ask:

How can AI make this person dramatically better at their job?

Human-AI collaboration can create value precisely because each side contributes something different. AI provides scale and pattern recognition. Humans contribute context, judgment, accountability, and an understanding of situations that models can still miss.

4. Build AI When It Deepens Your Competitive Moat

As AI makes software faster and cheaper to create, companies face another important decision: when should they build their own AI capabilities, and when should they use an existing product?

Andrew offers a simple framework.

If the capability is generic, rent it.

If it strengthens something unique about your business, consider building it.

For example, extracting basic information from an invoice is not necessarily a competitive advantage. Many existing tools can already do it well.

But if a company can combine that invoice data with proprietary information, customer relationships, operational knowledge, or historical data to generate insights competitors cannot easily reproduce, the calculation changes.

Andrew calls this moat deepening.

Custom AI makes the most sense when it improves something that already differentiates the company, such as:

• Proprietary data

• Unique workflows

• Operational expertise

• Customer relationships

• Specialized industry knowledge

• Better decision-making

Building software may be getting cheaper, but operating, maintaining, updating, and governing that software still creates long-term costs.

The question is not simply whether you can build something.

It is whether owning that capability gives you a meaningful competitive advantage.

5. AI Can Be Confidently Wrong

One of the simplest and most useful lessons from the episode is Andrew’s warning that AI can be confidently wrong.

AI systems often produce answers with the same level of certainty whether the answer is correct or incorrect.

That becomes much more important once AI moves from generating text to interacting with business systems, financial information, workflows, customer data, or operational decisions.

Andrew argues that production systems need guardrails, clear off-ramps, logging, visibility into decisions, and humans who understand what the system is doing.

The goal is not to eliminate AI because it sometimes makes mistakes.

The goal is to design systems that expect those mistakes.

For executives, this also creates an accountability question. Someone still needs to understand:

• Why the AI made a particular decision

• What information it used

• How mistakes are caught

• Who can override the system

• How business impact and ROI are measured

The more powerful AI systems become, the more important this layer of accountability becomes.

Final Thoughts

The AI hype cycle has created enormous expectations. Software is becoming easier to build, models are improving quickly, and companies can experiment with AI faster than ever.

But the lesson from Andrew Brooks is not that companies should slow down.

In fact, he argues that waiting too long may be a greater risk than experimenting.

The key is to approach AI with a clearer understanding of what happens after the demo.

Successful AI transformation requires real workflows, real data, human judgment, measurable outcomes, and systems designed for the complexity of production environments.

For founders, operators, and innovation leaders, the opportunity is not simply to use more AI.

It is to identify where AI can create a capability that makes the organization meaningfully better.

Have a question for a future guest? Email us at innovators@alchemistaccelerator.com to get in touch! 

Timestamps

00:00 🎙️ Meet Andrew Brooks
04:14 💸 Where companies misunderstand AI
07:32 🧪 Why an AI demo is not production ready
09:50 ⚠️ Where companies are wasting money on AI
14:42 ✅ Can AI actually solve these problems?
16:18 🧠 Human + AI beats replacement
19:19 📚 Why context changes what AI can do
23:40 🧩 Why AI orchestration matters
29:56 📊 Accountability, governance, and ROI
34:35 🏠 What AI can learn from the smart home era
38:59 👤 Why humans still want agency
40:52 💡 How to become an internal AI champion
44:24 🏃 Entrepreneurship is an ultramarathon
46:39 🔮 What Andrew got wrong about AI

Full Transcript

00:00:00:02 - 00:00:27:15
Andrew Brooks
The hype tax from our perspective is, oh, I see, I can do this easily and therefore I'm going to make the logical jump that something I can jam into GPT and have it spit out a response is a business system that is governed, that is secure, that is extensible, that is scalable. That confusion of, you know, what's easy to do in a one off is actually a scaled production system, creates this, this kind of confusion with, with executives.

00:00:27:15 - 00:00:36:20
Andrew Brooks
But like, why is it so hard? Why isn't it done? Why isn't it easier? And that creates frustration and and friction in actual successful rollouts.

00:00:36:21 - 00:00:40:15
Ian Bergman
Isn't that the point of AI that it magically solves all of these things for us?

00:00:40:16 - 00:01:02:07
Andrew Brooks
AI can solve real problems at scale in the enterprise. We process hundreds and hundreds of thousands of transactions. But if it's a complex task, 100% autonomy is probably a goal that we would say we're not there yet. You need to have the, you know, the guardrails and the off ramps for this thing to actually, you know, have it work.

00:01:02:07 - 00:01:05:06
Andrew Brooks
And if you don't build that in at the beginning, people are going to be frustrated.

00:01:05:07 - 00:01:23:00
Ian Bergman
The magic of AI is in actually human and AI being complementary, right? Not substitutes for each other. And that the tensions that are created there, not just the collaborative work, help the humans do a better job. So it does feel like there's something very real.

00:01:23:01 - 00:01:41:06
Andrew Brooks
AI can be confidently wrong. It loves to be confident, and it can often be confidently wrong. And so you have to as you connect it to systems, remember it's confidently wrong. So you know, trust but verify quite, quite a bit from that perspective.

00:01:41:08 - 00:01:42:07
Ian Bergman
Andrew welcome.

00:01:42:07 - 00:01:44:10
Ian Bergman
To Innovators Inside. It's good to meet you.

00:01:44:16 - 00:01:46:19
Andrew Brooks
Great to meet you. Thrilled to be here.

00:01:46:21 - 00:02:00:10
Ian Bergman
I'm excited. We were just talking before the recording. We're recording this on a sunny Friday in the summer. So, you know, if any of our listeners are thinking, wow, like, they seem distracted, it might be by sun and waves.

00:02:00:13 - 00:02:02:06
Andrew Brooks
There you go. Exactly.

00:02:02:08 - 00:02:30:19
Ian Bergman
But no, we'll we'll try and focus. We'll try and have a conversation. It's one that I'm actually very excited about, because every single conversation about innovation today inevitably leads to some notion. Something about AI, just like no way to avoid it, right? Absolutely. And we're here to talk about sort of an interesting turn of phrase that you've got, which is the AI hype tax.

00:02:30:21 - 00:02:55:13
Ian Bergman
And so, you know, I can't I can't wait for the conversation. So this whole episode I think, is going to be about the real cost of pretending that an AI demo is actually business transformation. I want to welcome the audience to innovators inside, I'm Ian Bergman. Every week I sit down with founders, investors, and leaders actually doing the work of innovation and not just talking about it.

00:02:55:15 - 00:03:17:28
Ian Bergman
Today's guest has earned the right to answer that opening question. Andrew Brooks is the founder and CEO of contextual IO and a serial entrepreneur who co-founded SmartThings, acquired by Samsung and now running in devices all over the world, including My Basement and SMB live, acquired by Rich local. These days, he has a name for something that should make every executive just a little bit uncomfortable the AI hype.

00:03:18:00 - 00:03:50:10
Ian Bergman
Tax the price companies pay when they switch on ChatGPT or Copilot, call it transformation and wonder why nothing has actually changed. So Andrew, in this conversation, I'm really hoping that we're going to dig into how to tell whether an AI use case is production ready or just a great demo. What it really takes to turn AI from a chat bot into a teammate, and when you should kind of rent AI tools, which is an interesting new phrase for using someone else's software versus building capabilities that you actually own.

00:03:50:13 - 00:04:02:24
Ian Bergman
And because we always want to talk about the human behind the ideas, Andrews, a Princeton chemistry grad turned serial entrepreneur who has finished. Is this right? Multiple Iron Man's and once ran seven marathons on seven continents in seven days.

00:04:03:01 - 00:04:05:07
Andrew Brooks
That is. That is correct. Yeah.

00:04:05:08 - 00:04:10:26
Ian Bergman
Well, I'm not going to compare myself to you, but we'll get into that later. Andrew, welcome to Innovators Inside.

00:04:11:02 - 00:04:13:27
Andrew Brooks
Yeah. Great to be here. I'm looking forward to the conversation.

00:04:14:02 - 00:04:18:00
Ian Bergman
Well let's just jump right in. You know you're talking about the AI.

00:04:18:01 - 00:04:26:21
Ian Bergman
Hype tax. It's your headline. What is it. And where exactly are companies wasting the most money when they say they're doing AI?

00:04:26:24 - 00:04:44:21
Andrew Brooks
Yeah, I'm going to answer it in a couple of ways because there's there's layers to what we perceive. So number one is, you know, most companies who say, hey, we're doing AI, that might mean exactly what you just described. You know, we've got copilot, or maybe we've got GPT licenses, maybe everybody's got Cloud Coworker running on their desktop.

00:04:44:21 - 00:05:14:19
Andrew Brooks
And what we would consider that is, is personal assistant AI. And it's immensely valuable. I'm sure everybody who listens to this is using it every single day in their life. There's a tax that is paid with those tools, though, in that when employees are using those tools relentlessly, they're generating endless content. Because AI loves to spit out tokens, it loves to create content and artifacts and pretty web pages and long form essays.

00:05:14:20 - 00:05:34:10
Andrew Brooks
And so there's a tax that actually happens in efficiency in organizations when when everybody is just producing a lot of content with AI, who is consuming that content? Are people actually reading it, understanding it, distilling it, or are they just putting it back into another AI? To summarize, and you get a game of telephone, you know, happening from from our perspective.

00:05:34:14 - 00:05:54:10
Andrew Brooks
So so there's a tax that you pay for how people use assistance with it within themselves. And we're an AI forward company. And we kind of have a rule which is I don't want to see your AI writing stuff in our slack channels, right? We we don't. If you want to communicate something to your colleagues, let's think about how that communication happens.

00:05:54:12 - 00:06:34:25
Andrew Brooks
And so that's one layer. And we don't spend a tremendous amount of time on the assistance. But we see it happen again and again and again. And as a result you sometimes have companies who are like, you know, I'm not sure where the benefit is. The second layer of the hype tax from our perspective, is what you just described, which is, oh, I can do this easily, and therefore I'm going to make the logical jump that something I can jam into GPT and have it spit out a response is a business system that is governed, that is secure, that is extensible, that is scalable, and that that confusion of, you know, what's easy to

00:06:34:25 - 00:06:53:08
Andrew Brooks
do. And I don't even want to call it demo, but what's easy to do in a one off is actually a scaled production system, creates this, this kind of confusion with, with executives. But like, why is it so hard? Why isn't it done? Why isn't it easier? And that creates frustration and and friction in actual successful rollouts.

00:06:53:15 - 00:06:54:04
Ian Bergman
You know, I.

00:06:54:04 - 00:07:23:20
Ian Bergman
Use the word trivial a lot with my team, and I don't think there's a word that drives them more insane because it's sort of like what? Like theoretically this should be trivial, right? But but I get how that that mindset shows up. Yeah. And is there a, you know, is there something that a CEO, an executive can ask, can, you know, go to the team and say, hey, Monday morning, that demo was amazing.

00:07:23:27 - 00:07:31:27
Ian Bergman
What are the three questions that I, as a CEO can ask to understand? Is this thing actually production ready and how to get there?

00:07:32:00 - 00:07:59:03
Andrew Brooks
Well, first of all, if it came into existence over the over the weekend when somebody was vibe coding, it's not production ready. But that's a good point. I think the the couple of things that we would, we would focus on is a lot of the business, AI business solutions that we see companies starting with are automations. They are there's a workflow or a document processing flow or something like that that they're seeking to use AI to automate.

00:07:59:04 - 00:08:28:19
Andrew Brooks
And that makes a lot of sense to great place to start. There's clear value in AI can play a really powerful role. So number one would be, you know, what's the breadth of our true production data or the formats of our through production data that you quote unquote tested this on. And I'll give you a practical example. You know, we'll enter into conversations with clients and say, hey, we have this inbox and all of our work orders come into it or our invoices or, you know, our vendor compliance documents, whatever it is.

00:08:28:20 - 00:08:46:21
Andrew Brooks
And we have humans who are extracting these documents and keying the information in. And we want to use AI to do that. Okay. Great. Perfect use case. What are those invoices look like? Oh, it's a single page PDF with a work order attached to it all the time. Throw my hand up. No it's not. It's that inbox is a mess.

00:08:46:21 - 00:09:07:12
Andrew Brooks
It's 100 page documents and they're rotated wrong. And it's a Chili's menu. And you know, all of these this different mess. So a production system isn't just that I can take a document, throw it into GPT and say, can you extract the invoice number? Of course, anything can do that. A production system is how do you handle the thousands of different permutations and variations?

00:09:07:12 - 00:09:12:27
Ian Bergman
But isn't that the point? Isn't that the point of AI that it magically solves all of these things first?

00:09:13:01 - 00:09:37:02
Andrew Brooks
Well, I think there would be a second type tax, right? Is is the belief that, you know, a free form prompt can just cause outcome to happen? Is is just not it's not true. As an example, if you have documents that have grids with numbers in them, AI struggles to extract that information. It's not perfect. And let's say it's 90.

00:09:37:03 - 00:09:50:00
Andrew Brooks
Let's pretend it's 99% perfect. You've still got that 1% that has to be handled. What do you do with that? Where does that go? You know, that's the difference between a production system and a and and and not in that case.

00:09:50:01 - 00:09:59:12
Ian Bergman
So so where are companies wasting money in your experience. Right. Like where is it just not working when they say that they're doing AI.

00:09:59:14 - 00:10:28:14
Andrew Brooks
Gosh, where are they wasting money? I'm not positive that I would look at most of our clients or our type of client who are lower mid-market companies and say that they're wasting money in trying to CI as part of a transformation journey, because I think there's a tremendous amount of learning. I would actually say the bigger risk to companies right now is delayed efforts, because they're fearful, as an example, that their data is bad, their data is unclean.

00:10:28:20 - 00:10:54:29
Andrew Brooks
AI is a deploying AI in that model is actually a great way to clean the data, right? It slags it. It identifies it provides a recommendation as to how to to curate it. And so I would say I'll answer the specific question, but I think the counter is not acting is perhaps more costly to businesses than acting. I think where we see where I think people are wasting money, this is just a personal gripe.

00:10:55:00 - 00:11:10:24
Andrew Brooks
I think a lot of AI is AI outreach from a sales and marketing standpoint. We all get it in our inbox. I think it's actually insulting. It's very transparent. And so I wouldn't, you know, we don't do a lot of that work, but that's one of those areas where I'm like, I'm not sure that's a great use of of you.

00:11:10:24 - 00:11:34:14
Ian Bergman
Have you have a real example of that or something adjacent. You can anonymize them, of course, but like something real where you can kind of walk through how we, how customer got from, oh like possibility. This can be amazing and solve a real problem for me, kind of, you know, pushed it right into the hype cycle and then was like, oh, maybe this is something I need to learn from rather than continue with.

00:11:34:16 - 00:11:55:04
Andrew Brooks
Oh, that's a that's a good one. So I do have a good example. We work with a client who does vacation property rental management down on the Gulf Coast of Alabama. They manage to find 503,000 rental units. They actually build the units, they sell them to the owners, and then they operate the, the, the rentals on behalf of the owners.

00:11:55:06 - 00:12:15:06
Andrew Brooks
And as a result, they do work orders, tens of thousands of pork orders a month to go fix things in these units, ranging from replace the batteries in a remote, replace a light bulb, the toilets leaking, the fridges running, you know, whatever that might be. And they had because they're sending out thousands of of work orders a month.

00:12:15:06 - 00:12:35:27
Andrew Brooks
And these work orders have to get back to the owners. They wanted. They had humans reviewing these work orders and saying, is this appropriate to get back to the owner? Yeah. And because, by the way, we don't want to send pictures of my feet or hey, this this, this place is a pigsty or, you know, like, there's appropriate things to go back to the owner.

00:12:35:27 - 00:12:55:03
Andrew Brooks
So they had a human in the loop there that was actually managing that. Well, we built them an AI system that said, we the AI system will enforce all of the rules that you defined for what a good work order looks like. And in fact, it turned out that the rules were so poorly structured and defined that the AI was just flagging every single one.

00:12:55:03 - 00:13:17:19
Andrew Brooks
It was like fail, fail, fail, and causing, frankly, more work than than than it was solving for. And I'm the first to admit that. And so in that case, it was wait a minute, time out. AI is creative and it's thoughtful, but it also will, you know, if you say every item that you use in this work order must have a picture of the serial number.

00:13:17:21 - 00:13:35:10
Andrew Brooks
AI is going to say, where's the picture of the serial number of the battery that you installed in the remote? And is that really a helpful thing? Probably not. And so that caused them actually take a step back and say, what are the rules for what a successful work order looks like? How are we actually communicating to the people in the field that are doing this?

00:13:35:10 - 00:13:40:14
Andrew Brooks
Let's get that right first before we seek to layer AI on top of it.

00:13:40:16 - 00:14:07:08
Ian Bergman
Yeah, I mean, that's actually a really interesting story. And I think it makes it makes the the risks, right, of embracing the hype too quickly evident. Right. But what's interesting to me is you also described a learning loop and an iteration loop that is possible today. That would have been extraordinarily challenging under any kind of reasonable cost pressures, honestly, even five years ago.

00:14:07:10 - 00:14:33:09
Ian Bergman
And I don't know. And I find that a certain amount of the AI hype I don't want to say this. I'm trying to be articulate with such an articulate guest here. But like, no, I find that a certain amount of the AI hype, the true bits of the hype are about, wow, we actually can solve this. And maybe the thing that people forget is no, there actually do need to be a few iterations and lessons along the way.

00:14:33:09 - 00:14:42:25
Ian Bergman
We do need to remember that secure, scalable, you know, compliant infrastructure is a thing. But is it hype to say actually it's solvable?

00:14:42:27 - 00:15:01:07
Andrew Brooks
Yeah, it's it's certainly not hype to say it's solvable. And in fact, I'm going to put a little cherry on top of that example. One of the things that the AI was doing, I don't want to say unexpectedly, because we told it to review the full ticket detail, but there was a line items on the ticket that would be invoiced back to the owner.

00:15:01:08 - 00:15:18:03
Andrew Brooks
The AI was suddenly saying, hey, you didn't actually put the light bulb that you said you installed on this invoice. You didn't put the batteries that you say you installed on the invoice. And so they were missing out on actual billable back to the client opportunities. And so, you know, there's there's these tugs and pulls of, of of benefits.

00:15:18:09 - 00:15:52:22
Andrew Brooks
These are AI can solve real problems at scale in the enterprise. We process hundreds and hundreds of thousands of transactions, ranging from trade services to work orders for rental properties to field services, tickets to managing forklifts, you name it, we're processing it where? But I have yet to see a solution that is 100. If it's a complex task, 100% autonomy is probably a goal that we would say we're not there yet.

00:15:52:22 - 00:16:13:00
Andrew Brooks
You need to have the, you know, the guardrails and the off ramps for this thing to actually, you know, have it work. And if you don't build that in at the beginning, people are going to be frustrated. They're going to be upset, you know, where did this go? How did this get processed. And there's a little bit of there's elegance to observing.

00:16:13:00 - 00:16:18:15
Andrew Brooks
How does the AI go through its logic steps and its reasoning steps that you can then kind of iterate and learn from as well?

00:16:18:16 - 00:16:50:13
Ian Bergman
Yeah. Well, and that's actually really interesting. Like, you know, a friend of the podcast, Vivian Ming, would argue that the magic of AI is in actually human and AI being complementary, right? Not substitutes for each other, and that the tensions that are created there, not just the collaborative work, help the humans do a better job. So it does feel like there's something very real, and it feels like these sort of academic realizations are really spreading their way into the real world.

00:16:50:16 - 00:17:11:15
Andrew Brooks
I think that's right. And, you know, we have a phrase, it's on our website, which is give, give your teams superpowers, right. And that's what we see AI doing is giving the humans superpowers. I'll give you a because I think examples of the way this all lands for people, one of our clients is a fleet services management. They manage forklifts and pallet jacks.

00:17:11:15 - 00:17:45:24
Andrew Brooks
Basically anything that moves in a warehouse that isn't a human, they will provide managed maintenance for and it's maintenance, it's repair, it's it's upkeep, etc. and as part of their value to the warehouse operator, they obviously keep the stuff running. That's important. But on a monthly basis or a quarterly basis, they would scan through invoices from the vendors who are maintaining this equipment and say, we think there's some cost savings opportunities, like we saw this duplicate thing or, you know, maybe this labor rate looks a little high, but it was very manual.

00:17:45:26 - 00:18:04:12
Andrew Brooks
As soon as we started actually using AI to extract all of those invoices and work orders, we can now put every single line item through a cost savings analysis comparing to historic comparing other like vendors looking at the contract, hey, you're not allowed to bill me for towels because those are disposables and my contract says you can't do that.

00:18:04:13 - 00:18:28:25
Andrew Brooks
So suddenly now that human who is who is doing the quarterly business review with their client anyway is coming with with an entire litany of look at all these cost savings opportunities that we had on your behalf because we're running this we have this intelligence. That person is now we're not replacing that person. That person is now just equipped with far more capabilities to do an amazing job on behalf of their client.

00:18:28:26 - 00:18:59:03
Ian Bergman
Understood completely. And I think, I think, you know, that framing of delivering superpowers, amplifying expertise, I think it's, you know, human centric and true. But I do have one question. Before we sort of jump to the next thread here, is it possible to actually avoid the step that we all went through, which is playing with the chat interface, realizing, oh, wait, this thing can write poetry?

00:18:59:03 - 00:19:19:13
Ian Bergman
Realizing, oh wait, this thing can generate LinkedIn slot for me. Oh wait, this thing can analyze the document for me. Is it possible to skip that or do we like right now professionally, just kind of need that unlock in our brains to actually take the next step and think about what a what real thing AI might be able to solve for us?

00:19:19:15 - 00:19:40:18
Andrew Brooks
I think there is an I mean, our name is contextual for a reason, kind of based around the, you know, the context window that AI is working in, basically the corpus of information that it has to make decisions. I think there is an unlock that we all need to go through, which is not just call response AI. I ask ChatGPT to write me a poem.

00:19:40:18 - 00:20:04:04
Andrew Brooks
It writes me a poem. That's it. But our long running exchanges where it's knowledge and it's, you know, it's it's it's it's intelligence, as it were, seems to be building. So as an example, I'm sure you experienced this. I'm taking it out of a personal life. I'll ask a question in my, you know, long running chats with ChatGPT and it'll reflect back on, you know, my wife, whose name is Aaron.

00:20:04:05 - 00:20:25:12
Andrew Brooks
He'll say, well, I'm not sure Aaron would do X, Y, and Z, and you're like, whoa, you know, there's a bigger you know, there's a bigger knowledge set that's happening here. And so I think there's an unlocked around the amount of knowledge that an AI system can maintain. And that doesn't necessarily mean there's a long running chat. It does mean in the systems we build.

00:20:25:13 - 00:20:54:00
Andrew Brooks
How much do you put into that series of steps to make the most intelligent decision that you can? People say, oh, I'll give you another example. We have a client that's in the commercial refrigeration space. So they build ammonia systems and they have a guy, Greg. And Greg is their best estimator, right. Like he's amazing. And so they wanted to build digital Greg and Greg's like, there's just too much in my brain to get into this AI.

00:20:54:00 - 00:21:09:22
Andrew Brooks
And we're like, let's record a bunch of interviews with you, and then we'll transcribe them and then we'll get all that information out. And I've used this story before, but it's a real story. You know, Greg knows that you don't hang half inch cable on the ceiling because people hold on to it. You got to do three quarter inch, you know, cable.

00:21:09:24 - 00:21:24:10
Andrew Brooks
And so by seeing the corpus of knowledge that I can work on, people start to recognize like this is not just write me a an AI slop post, I can put a lot into it and as a result, get the most nuanced outcome out of it.

00:21:24:12 - 00:21:51:07
Ian Bergman
Yeah, I love it. We've all like, we've all got a Greg. Right. And and what's really interesting is, you know, it's so hard to understand in my mind before you, you know, without really just digging in with AI, how much context, how much intelligence sits not just at the human layer, but at the interstitials between humans in conversation, all of these things that never existed in our data systems before.

00:21:51:08 - 00:21:51:29
Andrew Brooks
Yeah, yeah.

00:21:52:01 - 00:22:15:02
Ian Bergman
And now we can. But okay. You know, I want to I want to push you just a little here. Sure. You're, you know, you run a orchestration and context platform, right? Fundamentally, your business depends on companies believing that the hype tax that they're paying right now is real and asking for help. So, yeah, do me a favor. Like just steal me on the other side for me a little bit.

00:22:15:03 - 00:22:20:16
Ian Bergman
Like when is just turn on copilot. Actually the right answer.

00:22:20:19 - 00:22:43:04
Andrew Brooks
I don't believe for any company that that that alone is the is the right answer. I do think that the personal assistance going to become where our clients explore. So rather than be going in and having to sit down with you and say, let's get on the whiteboard and figure out all the use cases and like, you know, right up the ROI and determine how we're going to measure it.

00:22:43:06 - 00:23:04:27
Andrew Brooks
I think what happens is those copilots, those coworkers, those those assistants start moving up the stack. You know, Susan, over in accounting has built her own agent that does these things. Great. But now, you know, Tom over in supply has a different agent. We get a little bit of agent sprawl. So you've got to let them do that innovation.

00:23:04:27 - 00:23:36:12
Andrew Brooks
But they're going to need that orchestration layer common data fabric ultimately to ensure governance, to prevent agent sprawl, to to get the data into a single place where intelligence can happen on top of it. So I don't think it's ever going to be enough, but I also don't think it stops at just help me write this email. I think increasingly we're going to see our clients have already built a pretty robust capability, and now they just want to scale it out across the organization.

00:23:36:12 - 00:23:40:01
Andrew Brooks
That's great for us, lets you know we want to engage with them in doing that.

00:23:40:01 - 00:24:01:15
Ian Bergman
So so how do we know that the the for lack of a better term and correct me if there's a better one, but the orchestration platforms, the scaffolding. Yeah. Aren't sort of that next layer of hype tax like. Oh right. Like, you know, all I have to do is think intelligently about the intelligence platform for my entire business and all my problems.

00:24:01:22 - 00:24:09:27
Andrew Brooks
And then it's yeah, magic. Magic happens. Right? It certainly feels a little, you know, 1999 middleware potential. Yeah, a little bit.

00:24:09:28 - 00:24:11:00
Ian Bergman
It does, doesn't it?

00:24:11:01 - 00:24:33:04
Andrew Brooks
And and so I acknowledge that completely. You know, it's it's we do an AP inquiry service for a company basically their customers or they're the people who they're supposed to pay are sending an email saying, am I going to get paid? And they don't respond. They only respond to the squeaky wheels. It's a pain in the butt. They don't have a portal or they didn't have a portal to actually respond.

00:24:33:08 - 00:25:03:07
Andrew Brooks
But the problem there is that that data existed in multiple systems when you were going to get paid wasn't just, you know, in the finance system, it was in Salesforce and Dynamic CRM, and it was spread in in different locations. I don't think you can solve that problem without a common orchestration layer. I just think that that, that when when companies have disparate systems, disparate data across different systems, processes being executed in different locations, you have to have that orchestration.

00:25:03:09 - 00:25:28:26
Andrew Brooks
You just have to have that orchestration layer. I also don't think that any of the individual ERP CRM systems are going to effectively solve that, largely because, number one, they may not have access to the full data set that is required. Number two, most of our solutions pull in federated data from outside the organization to enrich the decision making process.

00:25:28:28 - 00:25:53:20
Andrew Brooks
And those systems are designed to hold that. And number three, 80% of the time we're building a new human AI interface that is a different UI or a different tooling that the people are working with or in, in order to achieve the outcome, and that those existing systems aren't going to do that. So it's a self-serving answer. I recognize that, but I think those are the causes where this is not just the next layer of tax, it's a it's a real need.

00:25:53:21 - 00:25:54:13
Ian Bergman
Well, yeah.

00:25:54:20 - 00:25:56:10
Ian Bergman
Which is, which is, I think, really fair.

00:25:56:10 - 00:26:05:03
Ian Bergman
And, you know, a quote unquote self-serving answer on one side of the coin can also be the answer of a founder who has nailed the expanding market segment, right?

00:26:05:08 - 00:26:08:03
Andrew Brooks
Like, yeah, there you go. Right.

00:26:08:06 - 00:26:35:07
Ian Bergman
But I'm actually glad that you brought up the notion of the ERP, the systems of record these, you know, in some cases, enormous systems, in some cases smaller that are a little bit hard to displace. And here's why. Before actually before we started recording, we were talking about how you've listened to a previous episode with Dave Lambert and his sort of a thesis that says that software is getting radically cheaper to build, which, right is true, like true.

00:26:35:07 - 00:27:09:28
Ian Bergman
And if you follow that to its logical extreme, software becomes effectively free and would replace everything that exists that we're paying for and we would no longer rent. That's not what's happening in the market. Yeah. What I see is that we are adding intelligence layers. We're adding new capabilities, Susan. And accounting is building wild new tools and is connecting them to the systems that exist.

00:27:10:00 - 00:27:23:14
Ian Bergman
Yeah, I have some theories on why that is, but I just wanted like, is that what you're seeing and what's going on there? Why aren't we naturally just displacing all of these sometimes very expensive tools? Yeah.

00:27:23:18 - 00:27:42:18
Andrew Brooks
I mean, I love, love and kind of a way the oh, I can vibe code and replace my Salesforce system over the weekend. Right. Like, it's just, you know, it's not true. A now, it might be that you have a set, you're only using a subset of those capabilities and that you could replace those and replace those with something that's more purpose built.

00:27:42:20 - 00:28:05:28
Andrew Brooks
Our opinion is this to your point and to Dave's point, the cost of the nominal cost of building new software is going to zero. You can turn Fable Loose on our platform with a with an instruction, and it's going to spit out some really powerful software pretty fast. Building is not operating, maintaining and owning. Over time you are inheriting a cost when you purpose built a solution.

00:28:05:28 - 00:28:27:01
Andrew Brooks
Whether you think you are or not, you are inheriting a cost. Things are going to change. Models are going to change. Data structures are going to change. The system is going to need to be maintained. And so it's important for our customers to recognize when do you purpose build solutions versus when do you rent a feature. For us it's all about moat deepening.

00:28:27:01 - 00:28:53:09
Andrew Brooks
If you have unique data and you can build a moat around that data. Better insights, a better ability to look around the corner that makes you more competitive. Purpose build AI software. If you have a unique operational ability or capability or process purpose built some AI solutions that make that faster. If you have a unique set of relationships with your customers and how you interact with them, by all means.

00:28:53:12 - 00:29:13:02
Andrew Brooks
If if something is great examples customers come to us and say, hey, can you use AI to extract our invoices? If all you're interested in is getting the vendor and the invoice number and the line items and sticking them into your your financial system, there's a million tools that can do that. Go rent that capability because it is not moat deepening.

00:29:13:02 - 00:29:35:25
Andrew Brooks
It's not differentiated. If you want to pull that data out and do something else creative with it, like the cost savings example for forklift management, that's a that's a moat deepening capability. It makes it means that when that company goes to a client, they're no longer saying our value proposition is we keep your forklift on. Yes, but we pay for ourselves in the act of doing so.

00:29:35:26 - 00:29:56:20
Andrew Brooks
Right. Like suddenly that's a that's a moat deepening. And so I think the, the just because it is cheap to make software doesn't mean it's cheap to operate and maintain software. It does take investment over time. And the only reason you should do that is if you believe that it gives you a differentiated capability versus what your competitors have.

00:29:56:26 - 00:30:29:25
Ian Bergman
So how do you think about the accountability cost and the accountability layer in all of this? When I hear you talk about sort of the importance of orchestration, next generation middleware, etc., I certainly understand it from a technical and a governance perspective, but I also understand it from the perspective of ultimately someone's but is on the line. And that's a really interesting thing, I think in the world of AI, right.

00:30:29:26 - 00:30:47:28
Ian Bergman
Like, you know, who's actually accountable if we're building a bunch of, you know, AI teammates or managers or managing a bunch of, you know, agents. Yeah. If where does the accountability sit? Well, it sits with that human. So how do you think about that as your as you're talking to your clients.

00:30:48:04 - 00:30:53:13
Andrew Brooks
There's a few layers of that. So so number one.

00:30:53:15 - 00:31:14:12
Andrew Brooks
I think Reid Hoffman had a quote that, you know, successful AI lives at the workflow level. And what I read that to be is it is the individual whose job is being affected, impacted, etc., where the AI lives and exists. And that doesn't necessarily mean that they're accountable, but they certainly have to be very deeply knowledgeable about what's happening and why it's happening.

00:31:14:12 - 00:31:38:04
Andrew Brooks
It can't just be a black box that is that is, you know, opaque to them. And so I think from an accountability standards standpoint, number one is as system designers and builders, we are accountable to that user to provide visibility into what happened and why. For us, that translates to some architectural patterns that say break these things into constituent steps.

00:31:38:04 - 00:31:54:24
Andrew Brooks
Don't just jam it all in and say, good luck agent with 50 different tools that it can call and and hope for the best, right? Like you have to break it apart. You have to see where it is progressing through. You have to have good logging, good attribution. Why did I do this? Why did I make this choice?

00:31:54:25 - 00:32:20:00
Andrew Brooks
And making that visible to the the so so as a system designer, we're accountable to the user to make that visible and to give them the tools to interact with that and make it better. Right. That that creates a sense of of ownership. I think from a governance standpoint, there's some really interesting stuff coming out of the EU, which is basically like you've got to have every bit of every step of every decision that AI touched and why it touched in what data it touched.

00:32:20:02 - 00:32:20:25
Andrew Brooks
What like.

00:32:20:26 - 00:32:38:20
Ian Bergman
That's right. Which is wild when just a couple of years ago, you know, the model owners, creators and Frontier Labs were saying, I cannot audit the decision process inside this model. It's like actually impossible. So it really interesting. So for sure.

00:32:38:22 - 00:33:06:04
Andrew Brooks
Yeah. No. And so I think that's so number one system designers accountable to the users to provide visibility into what happened. System designers accountable to the users to let them interact and improve it. You know, it can't just be a you know, I can't I can't you know, I can't I can't affect the change. Again, coming back to like, giving human superpowers, one of those superpowers is the ability to affect change within, you know, within the systems that we've created.

00:33:06:06 - 00:33:34:22
Andrew Brooks
I think at the, at the CFO level, at the Ku level, accountable around ROI, not necessarily saying cost takeout, but how do I measure the impact and benefit of this solution? Right. It's not always cost takeout. If this data gives us the ability in six months to launch a new product or service that's, you know, that's the you know, that's how we're going to measure that ROI.

00:33:34:22 - 00:33:59:01
Andrew Brooks
And so I think accountability there, you know, the number one thing that that you got to avoid is, oh, CFO is stroking a check now for tokens. And they have no knowledge as to what the benefit of this system is, because that's going to break down in terms of belief and confidence and perceived value. And so accountability to kind of stretch across the organization that says this is why we're doing this.

00:33:59:01 - 00:34:09:26
Andrew Brooks
This is what it's going to cost. This is how we measure the impact. And here are the users who are going to attest to the fact that this is actually improving their lives or making them better or, you know, etc..

00:34:10:02 - 00:34:34:27
Ian Bergman
Yeah. No, I think it's it's a it's a really interesting and important framing. All right. Andrew, thank you for sharing some of your stories about what's happening today and how what the AI hype tax is, how companies are dealing with it. If it's okay, I want to do a little bit of a callback because you have lived through some very interesting and driven, some very interesting eras in innovation.

00:34:35:00 - 00:34:49:16
Ian Bergman
And I want to I want to talk a little bit about smart homes, okay? Because, you know, there's one way of framing this where smart homes promised us a teammates a decade ago in our home.

00:34:49:18 - 00:34:53:19
Andrew Brooks
Right. Your home would respond to you. The lights would be there, would.

00:34:53:19 - 00:34:55:00
Andrew Brooks
Be there.

00:34:55:02 - 00:35:22:00
Ian Bergman
You know, the magical color schemes and musical tones that apparently we all wanted, but didn't, you know? But, you know, what we mostly got was a bunch of chatty gadgets, interop challenges, etc. so like my framing of this and, you know, real meaningful advances in a bunch of ways. But it was interesting because it seems to me that you have lived the demo to production gap in IoT.

00:35:22:02 - 00:35:34:08
Ian Bergman
And so, you know, my question is like, what are the lessons here that enterprise AI people haven't learned yet? Right. Yeah. What what do we need to learn from this last decade?

00:35:34:10 - 00:35:54:27
Andrew Brooks
Gosh, it's such an interesting question. And drawing that corollary, my my brain is going in a lot of directions. I think, number one, when we were creating smart things, there were a lot of moving parts to make those systems work. You had a mobile app, you had whatever the connectivity of the mobile app was that the user was having.

00:35:54:27 - 00:36:18:14
Andrew Brooks
This is, you know, 2012, 2013. So a decade plus ago, you had, you know, obviously the cloud, the cloud talking down to a hub, a hub, talking to all of these sensors, if talking to these sensors on Zigbee, which is 2.4GHz, which happens to be the same as Wi-Fi. Back in the day and other devices, Z-Wave, which is 900 megs, which goes long distances but is pretty weak.

00:36:18:14 - 00:36:39:02
Andrew Brooks
And so you had all of these different pieces. And the reason I say that is like inside the AI in this non-deterministic world, there's these pieces that are hard to understand. It's difficult for a customer who's calling into smart things when we're like, well, we don't know. Is your hub close enough to the device? Is there a microwave in between them?

00:36:39:02 - 00:37:01:20
Andrew Brooks
Is there a bit like a wall that is made out of metal that we, you know, we don't know about? Like solving those problems was difficult because you didn't have full visibility to the system. Number one. And number two, customers didn't care. Right. It's not they're not interested in you telling them that they're, you know, hub that needs to be moved like up a up a floor, right.

00:37:01:21 - 00:37:02:10
Andrew Brooks
Like that's.

00:37:02:12 - 00:37:03:10
Andrew Brooks
That's not my.

00:37:03:10 - 00:37:04:15
Ian Bergman
TV to turn on, please.

00:37:04:16 - 00:37:05:07
Andrew Brooks
Yeah, exactly.

00:37:05:13 - 00:37:06:03
Andrew Brooks
That's exactly right.

00:37:06:04 - 00:37:07:00
Andrew Brooks
Like, I.

00:37:07:00 - 00:37:29:00
Andrew Brooks
Got home, it was dark and my lights didn't turn on. Is is their issue. And so I think the, the learning is when you're dealing with complex systems and in this case with AI the complexity is again the non determinism. You don't you don't know kind of breaking it down into how do we make it just dead simple to, to do each step of this process.

00:37:29:01 - 00:37:46:07
Andrew Brooks
Right. You know, how do we how do we give you visibility to why this particular device was offline and when it went offline. And what to do to fix it is the corollaries are kind of similar is you have to like break these systems down into what is actually happening. Why did I make this choice? What data got?

00:37:46:09 - 00:38:00:26
Andrew Brooks
Data got written. But I think the second piece is, you know, we like candidly back in the day, we would get packages returned with like, you know, mean letters, right? Like it doesn't work.

00:38:01:01 - 00:38:03:09
Andrew Brooks
Right. And I.

00:38:03:13 - 00:38:06:18
Ian Bergman
Bet they got a little meaner than that sometimes.

00:38:06:20 - 00:38:23:15
Andrew Brooks
For sure. And I think the the lesson to take forward for, for enterprise AI is your users, your the people in your, on your team who you are about to put this system in front of are the same humans who you know are like my, my lights aren't. You know, my lights didn't turn on, my door didn't open.

00:38:23:16 - 00:38:41:22
Andrew Brooks
Right? Like they are frustrated, they are confused, and they don't have good tools and visibility to fix it. So it's your job to be a strong steward, to provide good, strong support, to again decompose these into the most understandable pieces. I think that's the there's a through line there that I had never really thought, but is is really true.

00:38:41:24 - 00:38:59:24
Ian Bergman
No, it really is. And I think I think we it's so we always talk about users and consumers and buyers and it but so easy to forget whether it's AI or not. Right. That like human behavioral patterns are a thing and they are the thing in a lot of ways. Yeah, yeah.

00:38:59:25 - 00:39:00:08
Andrew Brooks
We.

00:39:00:09 - 00:39:24:07
Andrew Brooks
We, we do some work where the AI extracts and categorizes stuff and then a human observes it and, you know, 30% of the time the human changes it. And I would look at that and say why did they change that? Is did they add value in changing that? Is this tweak that they made actually change the outcome. And and often the answer is no.

00:39:24:07 - 00:39:45:06
Andrew Brooks
But I think there's a, there's a secret little behavior here, which is I want I want agency, you know, not agent AI agent, I want human agency, and I want to have a role, and I want my brain to be felt and heard in this process. And so I'm making these changes so that I still I'm not just, you know, a click bot that pushes these things through, knowing that humans want that agency.

00:39:45:06 - 00:39:50:06
Andrew Brooks
They want to be intelligent creatures. Acting, you know, in thoughtful ways is part.

00:39:50:06 - 00:39:50:18
Andrew Brooks
Of it.

00:39:50:18 - 00:39:52:02
Ian Bergman
There's a whole other conversation.

00:39:52:02 - 00:39:55:15
Ian Bergman
We could have about the importance of self-esteem in human decision making.

00:39:55:16 - 00:39:57:07
Andrew Brooks
And there you go. But.

00:39:57:09 - 00:40:39:20
Ian Bergman
You know, we'll save that for the next one. But you gave you snuck in there. And you're actually a really interesting piece of advice for sort of the the managers, the implementers, the executives, the people that are helping guide their organization through, you know, through the AI hype cycle, through an AI transformation. Do you have a piece of advice for maybe, you know, the the next level down or maybe the internal, the agitator, the individual contributor who knows that change is possible, that something better is possible that's going to solve a meaningful issue in their organization, but they haven't yet developed the air cover.

00:40:39:20 - 00:40:52:10
Ian Bergman
They don't have. You know, they're in a risk averse org. They don't have the executive champion, you know, they're pushing up. What do they do to sort of like, understand and push through the AI hype cycle.

00:40:52:13 - 00:40:56:02
Andrew Brooks
Yeah. So.

00:40:56:04 - 00:41:26:18
Andrew Brooks
Broader comment within any organization that we work with, the individuals who become the cheerleaders of an AI solution we have seen get moved into bigger and bigger roles, driving more and more AI transformation. So to the extent I would say one bit of advice, to the extent that something that you're passionate about, you believe in and you're excited about being that cheerleader, especially in these earliest days, can can suddenly find you doing entirely different things within an organization than you might have expected.

00:41:26:18 - 00:41:53:26
Andrew Brooks
And so, you know, I if you're a put your hat in the ring kind of person, like that's a that's a great thing to do. I think to, to push up it is probably not there. We certainly see some people that are like, oh, you know, I did this over the weekend. Here's my vibe coded answer. I don't know that that is going to crush through, you know, a leadership team that might be fearful about data, fearful about governance, fearful about compliance.

00:41:53:27 - 00:42:30:08
Andrew Brooks
Certainly don't connect something to a system and say, check it out. Right. Like that's that's just going to aggravate the the fears and anxieties of of the of the organization. What I do think is, is a powerful mechanism and tool is be it the voice of, of quieting all of the AI noise that are out there. If your executives are on LinkedIn and they've they've clicked any AI thing now they're inundated with, you know, endless AI, new tools, new mic, new products, new models, new, you know, all this stuff.

00:42:30:09 - 00:43:08:00
Andrew Brooks
If you can be the voice of narrowing it down and, you know, to your to our the beginning of our conversation, how can we as an organization effectively use Copilot? How can we as an organization effectively use coworking? What's a step above that in terms of a long running agent that could do something? Hey, I'm going to show you that I can go into my co-working or my clod and put together a schedule action, and it reviews my emails and it, you know, reminds me to do things leading by that kind of very quieted example, I think is a way to, to, to create space for, for the bigger conversation.

00:43:08:01 - 00:43:27:15
Ian Bergman
That's that's actually wonderful advice. It's, it's it's practical. It leans into the fact that, you know, innovation doesn't have to be grand and huge. And I love this framework. I'm going to think about that for a while probably. Well, I'm doing something I've done multiple times, which is connecting to a.

00:43:27:21 - 00:43:29:00
Andrew Brooks
System of record.

00:43:29:01 - 00:43:31:27
Ian Bergman
Doing something and sending to my team. Hey, look what I just did.

00:43:31:28 - 00:43:34:13
Andrew Brooks
So yeah, it was a bit of a.

00:43:34:15 - 00:44:00:06
Andrew Brooks
You know, we and we have a phrase AI can be confidently wrong. It loves to be confident and it can often be confidently wrong. Right. You know we connected to the MCP of of stripe for something. And we just asked a simple question. How many, how many, how many customers does this, does this system maintain? And the answer was 1000 even 1000.

00:44:00:07 - 00:44:20:09
Andrew Brooks
Right. And anybody with any kind of like thought would say that's unlikely to be the actual answer. But it was it was like, no, that's that's the number. Right. And so you have to you know, I think there's as you connect it to systems, remember, it's confidently wrong. So, you know, trust but verify quite, quite a bit from that perspective.

00:44:20:10 - 00:44:23:21
Ian Bergman
No wise wise wise words.

00:44:23:24 - 00:44:24:06
Ian Bergman
Well this.

00:44:24:06 - 00:44:43:08
Ian Bergman
Has been a really fun conversation. I want to I want to end on just a little bit of a human note, if that's okay. Like about, you know, we think about a third of our listeners are either founders or aspirational founders. And you've been in that seat. You've been in that seat a number of times. Yep. You're also a runner.

00:44:43:08 - 00:44:59:08
Ian Bergman
Marathons and Iron Man's. So I'm going to draw the messiest connection in the world. And I'm going to ask you, is building a company more like the Ironman or the Marathon Challenge, seven marathon, seven continents, seven days or get it all out?

00:44:59:09 - 00:45:02:09
Andrew Brooks
Like answer that.

00:45:02:12 - 00:45:22:12
Andrew Brooks
Yeah, I I'm going to answer it an alternative way. I've done 100 mile race in my life and it was brutal and it was long. And there were dark, dark periods in the middle of the night, and it was cold at times and hot at times because it was in the desert. It's it's it's more it's more like that.

00:45:22:12 - 00:45:41:24
Andrew Brooks
You have to recognize that there's going to be moments where you're, you know, your energy is at the highest level. And then, you know, a day later your energy's at the at the lower level. So I think it's been blessed in that a couple of the companies I've built have transacted relatively quickly. We you know, we found in Smart Things in 2012.

00:45:41:24 - 00:45:58:26
Andrew Brooks
We sold it to Samsung in 2014. You know, that was that was we were at the right place at the right time. You know luck meets preparedness opportunity. That's how these things work. But you have to recognize it takes five years to build anything great. Even in a world of AI, I still think it takes five years to build anything great.

00:45:59:00 - 00:46:22:22
Andrew Brooks
It just might be built, you know, bigger. And or as Dave was saying in the previous podcast, it might take less, you know, cost to do it, but it still takes time. And so, you know, I think the, the, the answer is a really long race is, is what is what entrepreneurship is. But there's a great book about ultrarunning and it's the title is Relentless Forward Progress.

00:46:22:24 - 00:46:31:21
Andrew Brooks
And what that means is just put your next foot in front of you. That's how you finish an altar race, and that's how you finish a, you know, you create and build something great from an entrepreneurial standpoint.

00:46:31:24 - 00:46:32:02
Andrew Brooks
No.

00:46:32:02 - 00:46:38:28
Ian Bergman
It's it's wonderful advice. All right. So my closing question for you, Andrew, back to AI.

00:46:39:01 - 00:46:40:04
Ian Bergman
Is there.

00:46:40:07 - 00:46:49:24
Ian Bergman
Something that you believed about AI two years ago, maybe three years ago, that you now think is wrong?

00:46:49:26 - 00:47:13:16
Andrew Brooks
Something I believe that I now think is wrong? I think 2 to 3 years ago I would have said, we're at such a logarithmic pace that you would just be able to throw any challenge and it would get a perfect response from it. I don't think that that is true. While we might be at a logarithmic pace, there are still challenges where humans play a role.

00:47:13:19 - 00:47:36:01
Andrew Brooks
Here's an example. Trade services company, which means import or export. Or they file stuff with the US customs in order to bring stuff into the country. There's something called a import or security filing ISF. It has ten data points that really matter. Hey AI, here's an extract this information so that we can file this with with US customs.

00:47:36:03 - 00:47:56:28
Andrew Brooks
We get a document that is labeled. This is an ISO but it's clear as a human if you looked at this this is an air shipment. Es are for water ocean shipments. This is an air shipment. It doesn't need that. It still tried to do it because it was told that, you know, this is what you do. You you file es and it still tried to do it.

00:47:56:28 - 00:48:18:12
Andrew Brooks
And here we are now, two years later, you would think if we're at the cusp of the singularity of, of like superintelligence, that just like you as a human or I as a human in two seconds looking at that, say, oh, this is an air shipment. I can tell because, you know, if that's an airline, I know that to be an airline, I'm going to do something different with it.

00:48:18:13 - 00:48:40:27
Andrew Brooks
We're still not seeing in our practical uses that level of kind of intelligence that still comes. It comes so naturally to humans. And so I think as a, as a, as an aspirational technologist, two years ago, I would be like, you know, in the same way I said to my kids who are now 22 and 19, you're never going to have a driver's license because self-driving cars, whoops.

00:48:40:28 - 00:48:46:01
Andrew Brooks
You know, like that was, you know, it's slow and then fast, but these things take time.

00:48:46:02 - 00:48:47:09
Andrew Brooks
Yeah, they really do.

00:48:47:10 - 00:48:59:20
Ian Bergman
Well, awesome. So for anyone in our audience who wants to kind of join you on the journey, the one foot in front of the other journey that you're on here, where should they follow you? Are you on LinkedIn? Do they head to your website? What's the best way for them to keep.

00:48:59:20 - 00:49:00:25
Andrew Brooks
In touch? Yeah,

00:49:00:27 - 00:49:26:15
Andrew Brooks
Yeah, certainly LinkedIn. You can find me, Andrew Carol Brooks on LinkedIn. I'm I'm relatively off the other socials. So that's the that's the place where if you're going to if you're going to connect with me socially, please do so. You can of course, go to contextualize and reach out to us there. We're always happy to talk just openly about what's happening, why it's happening, you know, and share our best thoughts around how I can impact people.

00:49:26:21 - 00:49:39:13
Ian Bergman
Amazing. Well, thank you so much, Andrew, for coming on innovators Inside. This was a super fun conversation. And I just want to say I hope you get a chance to decompress, relax and enjoy the end of a summer week.

00:49:39:15 - 00:49:40:15
Andrew Brooks
Absolutely. Thanks.

00:49:40:15 - 00:49:41:20
Andrew Brooks
Ian Chow.