YT YouTube2Text

YouTube2Text → Transcripts → The 8 types of software worth building in the AI era

The 8 types of software worth building in the AI era — Transcript

by Build Great Products · 7,127 words · 1,024 segments · language en · Watch on YouTube

Open in transcriber Summarize with AI Download SRT/VTT

Full transcript

0:00If you're still coding the same SaaS
0:01dashboards and CRM tools in 2026, then
0:04I'm sorry to tell you, but your product
0:06might not make it. And there's a reason
0:08that these types of softwares are no
0:10longer viable businesses, and I'm going
0:12to get into that in a little bit more in
0:13a second. But what I want to show you in
0:14this video is the eight types of
0:16software that you should be building in
0:18this AI era. The eight types of software
0:21that have the biggest opportunity right
0:23now, and are going to be the most likely
0:24to succeed in the next 5 to 10 years.
0:27So, in this video I want to break down
0:28why traditional SaaS will die because of
0:30this AI era, this agentic era that we're
0:32entering into. I want to show you what
0:34replaces it so that you can start
0:36building something that actually has a
0:37high chance of being successful, and
0:39that you can start building today. And
0:41I'm going to show you how to actually
0:42start building this as well. If you
0:45don't know me, my name is Chris, and for
0:46the last 15 years I've designed apps and
0:48advised startups on product and design,
0:50and I've helped hundreds of people to
0:51build real products and software with AI
0:53inside of my community, and helped them
0:55get their first paying customers as
0:57well. So, if you're interested in that,
0:58you can head over to the link in the
0:59description down below. Otherwise, let's
1:01get into the video and show you the
1:02eight types of software that you should
1:04be building today. So, in this video I
1:06want to show you the eight types of
1:08software that you should be building
1:09right now, and the eight places where
1:11there's a massive amount of opportunity
1:13in this kind of AI era that we're in at
1:15the moment. I think a lot of people have
1:17seen the conversations around software
1:19being dead, or agents replacing all
1:21software, but I don't think this is
1:23true. And I think actually products
1:25matters more than ever. And by product I
1:28mean the taste that goes into creating a
1:30great piece of software, the brand and
1:32the product identity that goes into
1:34something that's differentiated in the
1:35market, and actually thinking about how
1:37you deliver outcomes for people rather
1:40than just create a set of features that
1:42you might have seen in a, you know, a
1:43dashboard or a CRM tool from the last
1:46sort of 5 to 10 years in software. And
1:49the thing that has changed is that
1:51anyone can build software now with these
1:53AI agents like Claude Code and Codex,
1:56but very few actually build something
1:59great. And that is the entire reason why
2:00I created this channel. Call it's called
2:02build great products for a reason,
2:04because I think there is still a massive
2:06opportunity to build great products over
2:09the next 5 to 10 years. And product it
2:12doesn't necessarily have to be just an
2:14application. A product is a combination
2:16of all of these different types of
2:18software or just different versions and
2:20different types of software and digital
2:22kind of software products that you might
2:24not have thought of. And I want to break
2:25down these eight types of software that
2:27you should be building in this AI era
2:30that have a massive amount of leverage
2:32and that there is a massive amount of
2:33opportunity right now to build these
2:35because not many people are actually
2:37building them. I also want to talk a
2:39little bit about why this change is
2:41happening right now as well. And there
2:42is a reason that these type of
2:44traditional kind of SaaS products are no
2:46longer working anymore. These generic
2:48SaaS tools that rely on just list of
2:50features
2:51are going to be very difficult to be
2:53viable as businesses moving forward in
2:55the next 5 to 10 years. This is because
2:56of these three things. First, building
2:59got incredibly cheap. So generic
3:00software is now very, very easy to
3:03create. It's a weekend clone basically.
3:05You can build an entire very complex,
3:08detailed piece of software in just a
3:10weekend. The moat has moved to
3:12expertise, taste, data, and
3:14distribution. And what do I mean by a
3:16moat? Well, if you don't if you don't
3:17know what a moat is around you know, any
3:20sort of business that we talk about this
3:21more in software specifically is a moat
3:23is the thing that makes a business
3:25defensible against things like
3:26competitors or people copying you. And
3:28the moat for a traditional service-based
3:30business would be the expertise or
3:32knowledge of the individual person. The
3:34moat for a software business used to be
3:36the fact that it was very, very
3:37difficult to build software that it took
3:39a long time and a lot of people, a lot
3:41of developers to actually build any any
3:43piece of software. And so your moat is
3:45that it would actually take someone
3:47a very, very long amount of time to
3:49build a a of software that would be
3:52competitive with you. But, what this
3:53also meant is that generic kind of badly
3:56designed and built products were viable
3:59businesses. I just don't think that is
4:01going to be the case anymore, and you're
4:02going to see the bar be raised for what
4:05a good software product actually looks
4:07like. And so, the things that we need to
4:08focus on are expertise. So, how do we
4:11translate your specific expertise into a
4:14piece of software that people are going
4:15to use to get a specific outcome? Taste.
4:18So, how can you focus on the quality of
4:20the experience and the actual products
4:23that you're building, the brand, and the
4:26vision behind that product, and the
4:28actual quality in terms of what happens
4:31when people use the product, so how it
4:32feels, how it looks,
4:34um the how intuitive it is, all of the
4:37different kind of key user flows through
4:40that product. And also, raising the bar
4:42on on the creative potential of these
4:44products as well, because AI makes
4:46building so cheap, the actual ability to
4:49explore a vast amount more of creative
4:52options in software becomes immediately
4:55viable because we no longer have to
4:56spend months and months and thousands
4:59and thousands of dollars on things like
5:01a redesign. You can literally test out
5:03hundreds of different creative
5:05variations very, very quickly using
5:07these AI tools without having a designer
5:09sit and design all of those different
5:11variations. You can have Claude and
5:12Claude design, take inspiration from
5:14places and create experiences that you
5:16couldn't possibly imagine before. They
5:18are now viable to create because AI made
5:20it so much more cheaper to build. So,
5:21taste becomes a moat as well. Your
5:23specific taste on how you want to
5:25execute something and the vision that
5:26you see, the vision that you have for
5:28that thing in the world. Also, data is a
5:31moat because the amount of specific
5:33industry data that you have that other
5:34people don't have access to, that is not
5:36in AI's regular sort of standard set of
5:39distribution, is going to be a moat that
5:41you can protect and you can defend
5:43against competitors as well because you
5:45only you have that specific data. Or, if
5:47you can collect a specific set of data
5:49from a specific industry, this is also
5:51why data brokerage is seeing a massive
5:53amount of popularity at the moment as
5:54well, because people are looking to
5:56leverage domain-specific data in order
5:59to be able to resell that to different
6:01AI companies as well. So data, if you
6:04have any specific industry expertise or
6:06knowledge, is going to be a a massive
6:08moat as well, and that's going to be
6:09across a ton of different industries.
6:12And then also distribution is a moat, so
6:13access to an audience, access to a
6:15community, a community of people that
6:17you've built around something specific.
6:19That's part of the reason why I'm
6:20building my business as a community
6:21business as well through this channel,
6:23through my community.
6:24Um is that community and audience is
6:26massively um massively important now,
6:29because it gives you access to that
6:30distribution. And so these things are
6:32the new moats, because building got
6:34cheap. The moat is no longer the code.
6:36The moat is all of the things around the
6:38code that you can still spend time and
6:41care kind of crafting and building over
6:43time. The second is that agents became
6:46the new users of software, or they
6:48become an equal um user of software as
6:51compared to humans. And so software now
6:54has two customers instead of just a user
6:56clicking around the user interface.
6:57People actually continue to use the
6:59tools, and then agents use these tools
7:02um directly as well. So through things
7:04like MCP servers, APIs, CLIs, and I'll
7:07talk about these a little bit later as
7:09well. Agents can now access your
7:11software and use it for you without the
7:13need for a UI. And there are some
7:15instances where UI combined with an
7:18agent-native kind of product makes a lot
7:20of sense. And there are instances where
7:22you might want to do something more like
7:24a headless piece of software, which runs
7:26um which is just used by agents
7:28basically to call when it needs to
7:29achieve a specific outcome. And then the
7:31third thing that changed here is that
7:33pricing moved to work rather than
7:36features. So people are paying for
7:38outcomes rather than the number of
7:40logins to a specific software tool that
7:42they have. So, people want to pay in
7:44order to get a specific outcome. It's
7:45very, very similar to the world of
7:47service work where you pay somebody to
7:49help you achieve a specific outcome. The
7:51same is going to be true for software
7:53products.
7:54And so, what I want to present to you
7:55here is eight types of software products
7:57that you can build today
7:59that will work in this agentic era where
8:01building is cheap, where agents are the
8:03new users, and where pricing moves to
8:05outcomes versus the features that you're
8:08building. And so, let's jump into the
8:10first business opportunity that we have
8:12here in software products. So, the first
8:14one that I want to talk about here is
8:16software-enabled services. So, this is
8:19where you don't sell accountant
8:20software. You want to be the accounting
8:22firm that's 10 times faster. And the
8:24reason that this is so powerful in this
8:27AI era is because before you would have
8:29a 20-person firm delivering a specific
8:31service to a client, whereas now you can
8:34have two people plus AI deliver the same
8:37client outcome, basically. You have
8:40AI-native firms like Crosby and Pilot
8:43for things like accounting, and this is
8:44great for things for industries like
8:46accounting, legal, anything where
8:48there's a lot of complex kind of
8:50knowledge work around something that can
8:51be turned into a specific process or
8:54kind of productized where you can
8:56basically implement a bunch of AI
8:57systems kind of behind the scenes to do
8:59the generic and the boring and the more
9:01administrative stuff, which is
9:03especially true with things like legal
9:05and accounting, um in order to help you
9:07achieve the same client outcome. But I
9:08also think this is going to be massive
9:10in the creative space as well for things
9:12like design, copywriting, brand
9:15development, all of these kind of
9:16creative pieces of work for businesses.
9:19I feel like there's going to be a ton of
9:20super successful agencies who have a
9:22very small number of people who build
9:24their own software products and their
9:26own AI systems in order to help them
9:29leverage AI in order to serve their
9:31clients or, you know, five to 10 to 20
9:34times faster and it
9:35And this means that if you set up these
9:37AI systems and this AI software for your
9:39own business, and this is something that
9:41I'm doing for my business as well at the
9:43moment, is building a custom piece of
9:44software that runs my business entirely
9:46the way that I want to run it. Now that
9:48I'm starting to kind of add these
9:50systems into my business so that it runs
9:52a little bit more on autopilot, whilst I
9:54kind of focus on the more higher
9:56leverage stuff, you'll be able to do
9:57that in in these kind of agency types of
10:00businesses as well. And so what I would
10:02think about is that if you have run or
10:05you've worked in a service business
10:06where you have some sort of expertise,
10:09how can you turn that expertise that you
10:10already have and return an outcome that
10:13you've already gotten for a customer or
10:15client, put in an AI system on top of
10:18that process, kind of map out that
10:21process, figure out how you can automate
10:23or manage different parts of that with
10:25AI to make it, you know, five times
10:27quicker, two times quicker than it would
10:29be before, and how would you then
10:32kind of put that into your service to
10:33deliver the same outcome for someone. So
10:35rather than selling a software product
10:37to someone, you sell your productized
10:40service that leverages AI software that
10:42you have built and AI systems that
10:44you've built to run that service
10:46business. The second type of business
10:48that is going to have a massive that has
10:50a massive amount of opportunity at the
10:51moment is the vertical agent harness.
10:54Now this is something that a lot of
10:55other people have talked about as well.
10:56People if you watched any of Greg
10:58Eisenberg's stuff or a bunch of other
11:00videos on this subject, maybe some of
11:01the Y Combinator stuff talking about how
11:03their latest batch is like half full of
11:06these vertical agent harnesses.
11:08Basically, you want to think about
11:10there's a few things to break down here.
11:11So first, what is a harness? Well, a
11:13harness is everything around the model,
11:16basically. It's the control is the
11:18controlled environment that you put a
11:20model into. It's the rules, the tools,
11:23the knowledge, the workflows, and the
11:26memory that surround the model. So you
11:28think you can think of we already have
11:30coding harnesses at the moment, things
11:32like Claude Code and CodeX that has an
11:34environment around the model that gives
11:37it these tools, knowledge, workflows,
11:39memory, and rules in order to be operate
11:41and write code effectively.
11:44And so, you can think about, you know,
11:46this is going to happen in every single
11:49niche, I imagine.
11:51You have tools like Harvey, Abridge,
11:53Evenup. These are doing these kind of
11:55vertical agent harnesses, these products
11:58that basically allow you to interact
12:00with the AI and have very specific
12:01knowledge, workflows, memory, rules, and
12:03tools are associated with that specific
12:05industry, like accounting and legal are
12:07great examples where we're kind of
12:09seeing emerging companies build these
12:12agent harnesses. But, I think this is
12:13going to be massively prevalent in every
12:15industry, even really, really niche
12:16ones. And again, I would think about how
12:19do you take your expertise if you're an
12:20expert in a specific niche or subject or
12:22you have a really deep passion for
12:24something. So, I will give you an
12:25example here is I'm building an agent
12:27harness for Dungeons and Dragons, for
12:29tabletop gaming, called Lawwell, which I
12:32plan to launch very, very soon. And this
12:34is because that is one of my passions.
12:36That is something that I've done a lot
12:37before. I've been a Dungeon Master for
12:39Dungeons and Dragons campaigns and I
12:40know it very, very well, and I want to
12:42create an agent native agent kind of
12:45agent harness product for people who
12:48want to run their D&D on tabletop
12:50campaigns leveraging the power of AI to
12:52help them run and plan those sessions.
12:56And so, I think this is going to have
12:57like wide implications across a vast
12:59number of industries and businesses.
13:02The models are very, very generally
13:04capable, and the gap is the harness you
13:07build on top of the model that combines
13:10these these rules, these tools, these
13:12knowledge, this work these workflows,
13:14and this memory all combined around the
13:17model. Um and that is a massive uh area
13:21of opportunity right now. And the other
13:23reason that these work really, really
13:25well is that
13:27they also get better as the models
13:28improve as well, which is one of the key
13:30factors throughout all of these
13:31different opportunities is that if the
13:34AI models get better, your business gets
13:36better. And that is why these work so
13:38well. So, for software-enabled services,
13:40as the models get better, your systems,
13:42your software improves running that
13:43process. As your the models get better,
13:46the model within your harness is going
13:48to help your products to get better
13:50overall as well.
13:51Um and it's going to get cheaper over
13:53time as well. So, these models are going
13:54to get more capable, cheaper to run. You
13:56can run cheaper models over time to be
13:58able to do specific jobs. You can
14:00really, really see how this is going to
14:02basically improve over time.
14:04The third type of opportunity that there
14:06is right now in software is headless AI
14:09software. So, what do I mean by
14:10headless? Well, this is What I mean by
14:13this is that it doesn't have a UI. It
14:15doesn't have a user interface that a
14:16human would interact with. It's purely
14:18agent-native. It's purely for agents to
14:21interact with the tool in order to get
14:23some information or use a specific tool
14:25and then bring it back into your
14:27projects that you're working on. So, an
14:29example for this is I built a tool
14:30called eyedropper, which is basically a
14:32CLI, which is If you don't know what a
14:34CLI is, a CLI is a command line
14:36interface. Command line interface is
14:39something that runs in your terminal on
14:40your computer, but this is a This is
14:42something that agents can use on your
14:43computer. They can use the command line
14:45interface to run things. So, eyedropper
14:48is a tool that uses a CLI to basically
14:50fetch your design system from a cloud
14:53server and then pull it back into your
14:55project. So, what this means is that if
14:57you have a design system across your
14:59brand and you want to use it in multiple
15:00projects with AI, all you have to use it
15:03do is to use eyedropper to go and get
15:04your design system, bring it back into
15:06your project, and that means you don't
15:07have to copy and paste your design
15:09system between all of these different
15:10projects
15:11with whatever AI that you're using.
15:13Another example of this is something
15:16called Posters, which is a social media
15:19scheduler a which basically allows you
15:21to schedule your social media posts with
15:23your AI agent. Your agent basically uses
15:26Post as through uh CLI or the MTP, and
15:29then it schedules your social media
15:31content for you. There is no UI unless
15:34And in some instances, like Post, there
15:36is a UI, but you have to the UI is the
15:38layer that you pay for. So, and Post is
15:41actually more fits into this open-source
15:43model, which I'm actually going to talk
15:44about a little bit later. Um but there
15:46is a massive amount of space for these
15:48but there is a massive amount of space
15:50for these purely headless AI software
15:52products that give agents tools to be
15:54able to complete a job where you don't
15:56need a user interface in order to see
15:58the output. It just is the agent
16:00communicates with the CLI to API or the
16:02MTP, and then gets the result and brings
16:04it back to the user. And this is going
16:06to be prevalent across, again, like a
16:08lot of different use cases for AI. And
16:11great if you're a builder with a data
16:14source or one hard or narrow skilled you
16:16want to turn into something that the AI
16:18that the AI can call in order to achieve
16:21that outcome. There is a trap here that
16:23you might build or Model Labs will
16:25bundle three in the next quarter, but I
16:27think there's a massive amount of
16:28opportunity here and it's generally
16:29following usage-based pricing for this
16:32as well.
16:33The fourth type of opportunity that I
16:35see in AI software right now moving
16:38forwards is agent-native systems. Now,
16:41this is this is a terminology that I
16:42have kind of coined for the purposes of
16:44demonstrating this to you here, but what
16:46I believe an agent-native system is is
16:48basically a plugin combined with some
16:51sort of a dashboard or UI layer where
16:54you run the plugin or the skills inside
16:57of your agent tool or with your agent,
16:59and then those skills or that plugin
17:01interacts with a a visual dashboard
17:04where you can see the progress of the
17:06plugin or where you can see different
17:09aspects of what the what the plugin has
17:11kind of brought back. And this is what
17:13I'm actually building with Product OS.
17:14So, Product OS is my plugin to help
17:17people build products with AI, and it
17:19takes you through all of these different
17:21steps in the journey from defining your
17:22idea to validating it to figuring out
17:24the product offer to designing it,
17:26creating the design system, building
17:28[snorts] the product, deploying it, and
17:30then getting your first customers. All
17:31of that stuff. But, what I want to do
17:33with a dashboard is basically allow
17:34people to see their progress throughout
17:36that plug-in because there's a lot of
17:38complexities and different skills within
17:40that plug-in. To be able to visualize
17:42your progress and see kind of where you
17:43are inside of that plug-in, I think it's
17:45going to be really, really important.
17:46And I think there's going to be a ton of
17:48other products in this space as well,
17:50which are basically a combination of
17:51plug-ins and or skills and some sort of
17:53a visual dashboard as well. And this is
17:55a bit like a combination of a kind of
17:59traditional UI with something that's
18:02agent native or agent first as well,
18:04which is the plug-in or the skill that
18:05kind of runs the workflows. And what I
18:07would say is that if you're an expert
18:10and you have a repeatable method that
18:12you can basically turn into a process to
18:14run with your clients. So, one thing
18:16that this this works really well for is
18:18coaches, consultants, or educators
18:20because you have a process to get
18:22someone a result in some specific field.
18:25You turn that into a plug-in. You also
18:26combine that with a dashboard so that
18:28people can run the plug-in inside of
18:30whatever AI agent tool they want to use,
18:32whether that's Claude, whether that's
18:33ChatGPT, whether that's Codex, whether
18:35that's Cursor, whatever that might be.
18:37They can install the plug-in in whatever
18:39tool they're using and then give them a
18:40dashboard for that data to to sync to,
18:43basically. And also the dashboard layer
18:45can be an area where you can collect
18:47data on your students as well and see
18:49how they kind of perform over time in
18:51order to improve your own product over
18:53time as well. And you basically earn
18:55through paid plug-in access or
18:57membership or dashboard subscription.
18:59So, like a membership or like a access
19:01to a coaching program is how you would
19:03earn for this potentially as well. And
19:05this is because people already live
19:06inside Claude and ChatGPT.
19:09They're already working with these
19:10tools. So, in order to kind of get them
19:13into your process or like get a tool
19:15that kind of works for them is to
19:16basically turn a workflow or a process
19:18into something they can run directly in
19:20the tools that they are already using,
19:22which is Claude and ChatGPT.
19:24And I think this ecosystem for plugins
19:27and skills and the kind of marketplace
19:28for that um on these platforms and
19:30you're already seeing early signs of
19:32this where you can kind of submit your
19:34plugin to be listed on the plugin
19:36marketplace for ChatGPT and Claude. Um
19:39we're starting to see early signs of
19:41this kind of happen anyway and like when
19:42that it becomes a little bit more
19:44visible to people, that kind of
19:47marketplace is going to be a huge
19:48distribution opportunity for people as
19:50well. The fifth area of opportunity and
19:52this is one of my absolute favorite
19:55favorite areas of opportunity because I
19:56believe in this so much is taste-led
19:59niche products. And what do I mean by
20:01taste-led niche products? Well,
20:04I mean that as AI has made building so
20:07so much easier to actually build any
20:10type of software, there is going to be
20:12these niche products in all of these
20:14different industries that become very
20:16successful off the back of having not
20:19loads and loads of customers,
20:21but by having a very specific niche, a
20:23very specific community or group of
20:25people they serve very very well and
20:28also making them taste-led where you're
20:30kind of a leading with the creative
20:32approach on this where you have you
20:34know, so for someone who has an audience
20:36or a community and I would say anyone
20:38with a passion who's building a
20:39community or has some sort of an
20:41audience, this is going to be great for
20:43us. You can think about existing
20:44creators, people who already post
20:46content online. If you have any sort of
20:48size audience, you could turn that into
20:51an app or a piece of software that you
20:52can sell to your audience and it doesn't
20:55even have to be, you know, it could be
20:57in a space that's very very competitive
20:58with a lot of competitors in because
21:00what you're building for is the 1,000
21:02true fans or the however many true fans
21:05that actually directly trust you. You're
21:07not building for a wider market. You're
21:09building as part of a broader business
21:11strategy that fits around content and
21:13your audience or community, rather than
21:15just building to try and, you know, get
21:17a million-dollar exit, basically.
21:21Um now, an example of this is on one of
21:23my design videos that was very popular.
21:25I built a roller coaster tracking app
21:28that was basically kind of came out of
21:30an idea that I had for like my visit to
21:33LA when, you know, as a kid, I was super
21:36into roller coasters and theme parks and
21:37stuff. You know, I'm not so much
21:39anymore, but as a kid I was, and I would
21:41have loved to be able to create that
21:43sort of product, to be able to track
21:45those kind of rides and stuff as I went
21:47to those theme parks, like on an app.
21:50And if you had a community or audience
21:51based around that, you could easily see
21:52creating a niche product like that,
21:54which is very heavily kind of creatively
21:57inspired or like taste-led, rather than
21:59being, you know, optimized for
22:01conversion rate or all these other
22:03things. So, you could easily sell to
22:05your audience. And the trap is trying to
22:07build something for everyone, where you
22:08want to build for the 1,000 people who
22:12would tell their friends about it.
22:13Really lean into going super, super
22:16specific on your um on your ideal
22:19customer. Especially if you already have
22:21an audience or a community
22:22built around that. And one thing that I
22:25want to call out here as well is that
22:26caring more than anyone else is the
22:28scarce thing. That's the kind of moat
22:30that you have here is that you care so
22:32much about this specific niche or space.
22:34That you want to build the perfect
22:36product for those very, very specific,
22:38hyper-focused kind of people on that
22:41thing. The sixth type of massive
22:44opportunity that we have in AI right now
22:46is ambient consumer agents. And you're
22:48already seeing stuff like this uh
22:51basically happen with Instinct and with
22:54Muse and with ChatGPT dots and with
22:56Grokbot is these always-on ambient
22:59consumer agents. And this is the best
23:01new app might be a member of your group
23:03chat, and that is because the real key
23:06point to these products is that they
23:08live inside the tools that everybody
23:10already is. WhatsApp, iMessage, Meta,
23:14basically like Meta AI lives inside of
23:16WhatsApp. That is now Muse. So like you
23:19have your agents kind of inside your
23:21group chats or like where you're already
23:22interacting with people. So I can, you
23:25know, I'm messaging my friends in
23:26iMessage. I can also message my agent in
23:28iMessage. We're not adding in a new
23:30touchpoint or a new app. And the reason
23:33that consumer is having a massive moment
23:35right now is because we're seeing, you
23:37know, this prevalence and rollout of AI
23:39agents across businesses and enterprise
23:42and like work and productivity. But what
23:44we've not really seen is the right
23:46application of AI inside a consumer
23:49product. So this is a massive area of
23:51opportunity. You don't want to build
23:53anything that's too much novelty. If you
23:55want to take a big risk on something, if
23:57you want to take a big shot on
23:58something, and you know how to build for
24:00consumer, you understand families,
24:02clubs, teams, those sorts of things,
24:05then building an ambient consumer agent
24:07is a really good project to work on.
24:09It's a hyper-competitive space, but I
24:11think there's going to be a ton of
24:12products in this space that do really,
24:14really well. Agents can take action for
24:16people. The key, I think, is in
24:18abstracting the agent
24:21well enough so that people don't have to
24:23understand the concept of an agent
24:25because agents can now plan, book, and
24:28remember things. You don't have to
24:30download an app, so no app download
24:32needed. Basically, how do you translate
24:35the user experience of an agent into a
24:37conversation that, you know, your auntie
24:40or whoever can immediately understand
24:42the value of that. I think that's the
24:44real challenge with consumer agents is
24:47fitting the user experience of an agent
24:50chat into the familiarity that all of,
24:54you know, all of the billions of people
24:56that are not like massively focused on
24:59AI and know exactly how it works, that
25:01all of them would feel comfortable using
25:03and would get some sort of utility out
25:05of as well. It's to get some sort of
25:07usage out of that. That makes sense for
25:09that. The seventh type of products where
25:11there is a massive opportunity right now
25:13which is proprietary data products. And
25:15this is because anyone can copy your app
25:18but nobody can copy your data.
25:21And what I mean by that is, you know, we
25:23want to think about this concept of out
25:25of distribution data. So in an AI model
25:28there is in distribution which means
25:30that your the data that data is inside
25:32of the distribution of the um
25:36in the training set for that model.
25:38Whereas out of distribution means that
25:39it's very specific type of data that's
25:42not um in the training data for a
25:44specific type of model. The more unique
25:46data that you have and the more unique
25:48insight that you have off the back of
25:49that data, the means that the more
25:52likely that product is going to be
25:53valuable. Models and code are
25:55commodities and niche data is scarce. Um
25:59so an example here is like Strava heat
26:00maps, Waze traffic, Glassdoor salaries.
26:03So all of this information that is
26:05behind like in either inside of these
26:07companies or in people's heads and these
26:10are kind of just like big examples where
26:12it's like there is a ton of industries
26:14where where you can think imagine that
26:16there is a ton of data existing inside
26:19of those businesses and inside of those
26:20industries that that aren't accessible
26:23by most people's AI agents. And so how
26:26do you turn that data into a product
26:28that AI agents can use to help other
26:30people get a very clear outcome.
26:33So if you have a business or a community
26:36with data that nobody captures or isn't
26:38in distribution, you can build a product
26:40around that.
26:42And one thing to also think about here
26:43is that if you are building a products
26:45that is going to have a benefit for a
26:47certain type of customer, how do you
26:49actually build in this unique data or
26:51you build in some sort of data
26:53collecting mechanism so that you can
26:54build up this proprietary data over time
26:57in order to make that valuable. And so,
27:00if you have a business, if you have a
27:01community, and you have access to
27:02specific data that isn't in
27:03distribution, build a product around it,
27:05make it agent native so that agents can
27:07query that, charge for that access, and
27:10then you can easily kind of turn that
27:12into a product and monetize it as well.
27:14And then the last type of massive
27:17opportunity in software products right
27:19now is open-source core products. Now,
27:22what do I mean by open-source core?
27:24Well, I mean that the core of the
27:25product, the main functionality, is
27:27open-source. And by that, I mean it is
27:30free. Because the code is free, it's
27:32easy to create. What is not easy to
27:34create other things around the code that
27:36actually make that product product easy
27:38and usable and friendly for the majority
27:41of people. So, the money is in things
27:43like hosting, paid tiers, support, and
27:45community. An example of this is
27:48Postgress, as I was mentioning before.
27:50So, you kind of build the software
27:51product, you make the code very easy to
27:53access, you make it open-source so that
27:55people can set it up and install it on
27:57their own own machine. But where the
27:59money is is in the hosting of that
28:02software. So, if you want to use that
28:04software and be able to access it from
28:05any computer, then you have to use the
28:07hosted version in the cloud. If you want
28:09to have access to support, then you have
28:12to use the paid version. If you want to
28:14have access to other features which
28:16maybe leverage the data set that exists
28:18within that product in order to provide
28:20better outcomes or better insight, you
28:22have to use the paid version. And also
28:24in this instance, the published version
28:26on GitHub, the open-source version,
28:29basically becomes the marketing. That
28:31that is your marketing, that is your
28:33distribution. So, the code is the
28:35distribution. And this is if you're a
28:38builder and you're happy to work in
28:39public and you want to grow a community
28:41around the product you're building, and
28:42you also want to you're happy giving
28:44away the software for free, and there is
28:46a reason to charge for something else
28:47where you think people would actually
28:49want to pay for that on top of the open
28:51source code, then you know, things like
28:53hosted tiers, hosted cloud, paid tiers,
28:56support, and sponsorship, then you can
28:58build a product in this space as well.
29:00Things like
29:02Superbase, Posthog, cal.com, these are
29:04open source code with a paid cloud
29:08products. And this also gets around the
29:10issue of people copying your products
29:11because anybody can copy your product,
29:13but people can't copy because anyone can
29:16clone your products by copying the code,
29:18but what people can't copy is everything
29:20around that. So, your distribution, your
29:22community, your cloud service,
29:25everything that connects to that free
29:27kind of software. There is a trap here
29:29that you give it all away for free, but
29:31you want to make sure you have a clear
29:33path to monetizing things that people
29:35will actually pay for beyond the
29:37software code itself. So, how do you
29:39actually choose which which one of these
29:41pieces of software that you should be
29:43building? Well, I've created this little
29:45breakdown here, so you can screenshot
29:47this or pause it here and kind of figure
29:49out like which one you actually want to
29:51build here. And but basically to kind of
29:53go through these very Basically to kind
29:55of go through these very very quickly,
29:57software enabled services are great if
29:58you have service experience or if you
30:00already have clients or you already have
30:02some sort of agency or freelance
30:04business that you want to leverage AI in
30:06order to kind of five to 10 times the
30:09profitability or the efficiency of that
30:12business. And this can work both ways as
30:13well because you can also you can make
30:15it work more profitably or or you can
30:17kind of like scale your revenue or you
30:20can just change that so you're actually
30:21working like um half the amount of time
30:24to get the same amount of um outcome,
30:26and you're just making your job or your
30:28business way way more friendly to your
30:30lifestyle.
30:31The second for vertical agent harnesses,
30:33if you have deep niche expertise where
30:36you can really serve a specific type of
30:38customer incredibly incredibly well and
30:40you want to build an agent product
30:41around that. So, really great tool. It's
30:44really great software product to build.
30:45Headless AI software like CLIs, APIs,
30:48MCP servers that don't have a UI. It's
30:50really, really good if you have one
30:52specific data source or very hard narrow
30:55skill. Make it usage-based. You can also
30:58do this open source as well.
31:00This is a little bit more difficult to
31:01build generally because it's a bit more
31:03technical, but you can absolutely build
31:05this still as a non-technical person
31:07with uh AI coding agents like Code Code
31:09and Code X. If you want to build agent
31:11native systems, then you need to have a
31:13repeatable method that you've done and
31:15there's proof that works. So, coaches,
31:17consultants, those sorts of people. If
31:19you have a repeatable method, a
31:20repeatable process, or IP that you have
31:22that you do specifically with clients,
31:24then turning that into an agent native
31:25system is really, really it can be
31:27really, really powerful. Um and it can
31:30mean that you can serve more people and
31:31you can scale your coaching business
31:33without kind of changing any of the
31:34system that it kind of runs on.
31:36Taste-led niche products. If you have
31:38passion or if you have a community or if
31:40you have some sort of an audience, these
31:41are great products to build. Ambient
31:43consumer agents. If you have ambition to
31:47making the next billion-dollar company
31:48and you want to really, really shoot for
31:50the moon and you really understand
31:52consumer products, then this is a great
31:55big bet that you can take, but it's
31:56really going to be a really, really
31:57competitive and difficult space to kind
31:59of win in.
32:00And proprietary data products are great
32:02if you have access to data that nobody
32:04captures or is outside of distribution.
32:05So, if you run a company in a specific
32:07industry where you've collected a ton of
32:08data and you want to run that you want
32:10to turn that data into a product, you
32:12can absolutely do that with AI now. And
32:14for open-source core products, if you're
32:16comfortable building in public or you
32:18think the thing that you're building um
32:20you what you could you're able to give
32:21away the code for free and then monetize
32:23it through some sort of back-end
32:25infrastructure which are revolves around
32:27hosting or support or services on top of
32:30that actual product, then this is a
32:32great way to go as well. And these are
32:34the eight types of software products
32:36that I believe have a huge amount of
32:38opportunity and I'm going to continue to
32:40be very have very incredibly high amount
32:43of leverage in this AI era and that you
32:46should be absolutely building instead of
32:48just generic SAS dashboards and CRM and
32:51habit trackers and all of these things
32:52that people are just building over and
32:54over and over again with AI tools. And I
32:56think this is really the line that I
32:57want to end on here is that software
32:59isn't dead, just generic software is and
33:01there is a huge amount of room for great
33:04AI native agent native products still.
33:07Um I have a I have a system called
33:10product OS which you can use to build
33:11any of this stuff. Um
33:13I teach inside my community or you can
33:15get it for free on my website as well.
33:17So you can just go over to
33:18buildgreatproducts.com {slash}
33:19product-os. So I'll product OS product
33:22OS for free.
33:24Um
33:24I also help people build this stuff and
33:27I'm actually going to be doing this
33:28pretty soon showing you how I built my
33:29D&D agent harness and go into how to
33:32build agent harnesses in a lot more
33:33detail. So feel free to subscribe to the
33:36channel or keep an eye out for that
33:37video if you want to understand how to
33:39build agent harnesses. What I would love
33:41to know is if you want me to do a deep
33:43dive into any of these eight types of
33:45software products, please leave a
33:47comment down below. If you think there
33:49is anything that I have missed here,
33:51please leave a comment down below. I
33:52love to see what people are building and
33:54uh let's get into my final thoughts off
33:56the back of this. So hopefully there you
33:58can see the eight different types of
34:00software products you should still be
34:02building now that have a massive amount
34:03of opportunity. And I think software
34:06isn't dead, it's just changing. You
34:07can't build generic feature-led software
34:10anymore. You can't build generic SAS
34:11products anymore. What you have to build
34:13is software products that are agent
34:15native that are going to work in this AI
34:18era where they get better and they
34:19improve when the AI models get better
34:22and which actually turn existing
34:24businesses and kind of amplify them
34:26instead of replacing people or replacing
34:28specific businesses or industries. And
34:31if you are building anything with AI,
34:33I've got a free list AI like I
34:34mentioned, the link is in the
34:35description down below. It's over at
34:36school.com/ai apps. If you have any
34:38questions, leave me a comment down
34:40below. I love helping people build this
34:42stuff, and I love seeing what everyone
34:43is building here. So, I'm just happy to
34:45look at anything that anyone is
34:47building, because I'm having an absolute
34:50blast building stuff with these AI
34:51coding tools. Thank you so much for
34:53watching. If you enjoyed it, don't
34:54forget to drop a like and subscribe, and
34:56I will see you next time.

About this transcript

This page contains the full transcript of The 8 types of software worth building in the AI era by Build Great Products, generated from the public captions YouTube serves with the video. The transcript has 7,127 words across 1,024 segments, with the original timestamps preserved so you can click any line to jump to that moment in the embedded player.

What you can do with it

Use the transcript to take notes, quote the speaker, build a study guide, generate a summary with ChatGPT or Claude via the YouTube Summary tool, or export it as a timed subtitle file with YouTube to SRT. You can also re-open it in the transcriber to translate the transcript into 100+ languages.

Free YouTube transcript tool

YouTube2Text is a free YouTube transcript generator — no signup, no daily limit. Paste any YouTube link and get the full transcript instantly, with timestamps, click-to-jump, translation to 100+ languages, AI prompts for ChatGPT, Claude, and Gemini, and exports to TXT, SRT, VTT, or Markdown.