Delivery Isn't Free

00:00:03: Hi, folks.

00:00:04: This is All

00:00:05: Things

00:00:06: Product with Petra Wille

00:00:07: and Theresa Corks

00:00:09: And we're so happy you are here!

00:00:17: Teresa I know You will love what i say next.

00:00:23: Now With AI Delivery Is Basically Free.

00:00:28: It's So Cheap.

00:00:30: it's basically free.

00:00:31: doesn't cost us a dime to develop software these days.

00:00:36: Shoot!

00:00:38: Its all we hear about in our calls right?

00:00:41: I mean, when i look at my AI bills it's certainly not free.

00:00:46: No!

00:00:47: Uh yeah let's unpack this so...I'm seeing this everywhere is delivery-free.

00:00:52: now that delivery is free complete the sentence.

00:00:56: uh..i don't think Delivery Is Free?

00:00:57: I Don't Think Delivery Will Ever Be

00:00:59: Free?!

00:01:00: I Think Delvery Is Getting Cheaper But Not All Delivery is getting cheaper.

00:01:06: So I think there's a couple ways I want to unpack This.

00:01:09: Yeah

00:01:11: The first thing is building a single thing, Is getting significantly cheaper.

00:01:19: I think that's true.

00:01:21: But if the way companies respond to building a Single Thing being Cheaper means they build lots more things Then delivery isn't going to stay cheaper.

00:01:35: Because let us just think this through We have a product.

00:01:38: It's brand new zero-to one product

00:01:42: M.V.P.,

00:01:43: so the minimum product that we think will get us a customer...

00:01:47: Deliver value, yeah!

00:01:48: It has one feature.

00:01:50: now deliveries free.

00:01:51: So he had five features in week One.

00:01:53: Now it has six Features.

00:01:55: Now on Week Two We add Five More Features and now Has Eleven Features.

00:01:58: On Week Three We Add Five more Features.

00:02:01: That Has Sixteen Features.

00:02:02: Here's What's Happening In Those three weeks?

00:02:05: We just did this very quickly.

00:02:07: what's happening in those three weeks is feature one was easy to build.

00:02:10: There's no codebase.

00:02:11: It was super easy to build, uh feature two is a little harder to build but it's still easy to Build by the time.

00:02:17: you're getting to Feature seven Feature ten Feature twelve Feature fifteen.

00:02:22: your data model looks like a Frankenstein strategy.

00:02:25: Uh You because you're using coding agents to build with almost No engineering insight.

00:02:30: cuz you expect engineering To be free have spaghetti code everywhere.

00:02:36: You've rewritten the same codes seventeen times which means when you want to change something, your changing it in seventeen places.

00:02:42: Performance already down?

00:02:44: Yeah!

00:02:44: Your performance of your product is probably terrible from a like speed and performance standpoint.

00:02:49: The maintainability of your code is terrible but sure we got fifteen features for free And I think this if you don't understand engineering It's really easy to naively thing.

00:03:02: Coding agents makes everything free.

00:03:05: Coding Agents make things faster, that is certainly true

00:03:10: and experimentation easier

00:03:12: And it allows us to create throw a code very easily but we can't mix those things up.

00:03:19: actually I really love the SVPG.

00:03:22: folks are starting use this phrase build to earn versus Build To Learn.

00:03:29: And there's this idea of like yes,

00:03:31: we love it.

00:03:31: build to learn is.

00:03:32: We're gonna write code To build an interactive prototype that were going put in front Of people and that is dirt cheap today.

00:03:39: and good

00:03:39: long as you throw It away.

00:03:42: Yes

00:03:43: right?

00:03:43: As long as we throw it away when we billed to earn.

00:03:47: now he know what to build.

00:03:49: we're building so that a paying customer buys it.

00:03:52: We still need to care about all those engineering principles, what you've always cared about which is maintainability of our code and non-functional

00:03:59: requirements as we work on them.

00:04:01: Reusability

00:04:02: of out code the cost to maintain that code?

00:04:06: The security of that code.

00:04:07: how well that codes scales.

00:04:09: And can AI help with us?

00:04:10: yes Is free.

00:04:13: no way like not even close.

00:04:17: And

00:04:18: so that you would still need a skilled engineering team,

00:04:23: observing

00:04:24: and overlooking what AI does.

00:04:26: Absolutely!

00:04:30: It takes time like... AI can generate a lot of code really quickly.

00:04:35: we get more sophisticated in our AI reviews but it probably still requires the human to look at.

00:04:43: is this architecture going to grow given our company's vision?

00:04:48: Mm-hmm,

00:04:49: right.

00:04:50: And like maybe here's the challenge with AI episodes Like maybe in six months I'm gonna regret saying this because a is going to get really good at understanding your product vision and creating The perfect code base that scales infinitely.

00:05:02: we never may

00:05:03: be.

00:05:04: but even if all of That happens Even If All Of That Happens even if All These Harder Engineering Problems Just Get Solved by Coding Agents We Still Have This Core Fundamental problem on Any Product The more bloat you add to the product, the harder it is to build on that product.

00:05:25: Yeah plus may I add another layer?

00:05:28: I just recently saw a chart basically let's say like apps released versus apps used in the app store.

00:05:37: And so apps released is spiking, of course.

00:05:40: But apps that people are actively using over time... That's very stable!

00:05:47: So it's not only the code bases get worse if you use AI delivery for stuff to earn money with.

00:05:59: Consumers have more options?

00:06:02: But they're more likely to be really sensitive when it comes to the quality of tools and apps that are used, because they will get very good in sensing which products have been thoughtfully built or designed.

00:06:16: Which ones basically AI-appslop?

00:06:20: So I bet there's a skill humans can develop next six months!

00:06:26: And then it's very hard to compete and earn with a product that is not well designed, thought through tested ruggedized where all the non-functional requirements are not even taken into account which is slow.

00:06:40: Where there's lot of feature creeps.

00:06:42: so yeah... That another layer.

00:06:45: why its'nt good idea?

00:06:48: think building is free.

00:06:50: I think

00:06:50: this other category did explore AI.

00:06:55: coding agents are really good at writing deterministic code.

00:06:59: And so if we're going to live in a world from five years ago and we do great discovery, when you know exactly the right thing that build sure delivery got a lot cheaper.

00:07:09: We can build Deterministic Code features really easily with Coding Agents.

00:07:14: But we don't live in Five Years Ago.

00:07:16: We Live In Twenty-Twenty Six Where It Turns Out That We Can Do Amazing Things With AI.

00:07:20: And if we're building AI products, which is probably what's gonna be required to survive in the future at

00:07:25: some point.

00:07:26: Yeah

00:07:26: coding agents aren't good.

00:07:28: don't make Building AI features free Right.

00:07:33: so I know i've been spending months building it two ai features and It takes a lot of error analysis and Creating evals that measure The right thing and iterating on prompts an orchestration and like Yes, there are some people out there that are starting to automate.

00:07:51: Some of those pieces with AI but I can tell you it's not very good at It.

00:07:56: no and so

00:07:56: like we're stepping into this new world where the types Of products were building our changing.

00:08:02: And yeah delivery is definitely Not free.

00:08:06: No Yeah because every button because of the learning curves to some extent as well.

00:08:12: Right?

00:08:12: Because everybody has to experiment a lot, play a lot and learn a lot why they go.

00:08:17: because nobody has figured it out right now with all these AI products.

00:08:20: And yeah we get this first.

00:08:23: I always see like a river in sums an another water is kind off.

00:08:27: um The water is going down a bit and you see the stepping stones.

00:08:39: of context broad and context engineering.

00:08:41: And everybody starts to have terminology to talk about certain things, your evils topic.

00:08:47: but still there's so much influx they're so much water sill.

00:08:51: um and still we are maneuvering this kind of early beginnings of AI and how it works.

00:08:58: It is like command line has been invented.

00:09:01: now everyone is figuring out.

00:09:08: are just scratching the surface on this stuff?

00:09:10: I mean, i can give a really personal example.

00:09:13: We... our AI generated snapshots and opportunity solution trees with Vistillia going really well.

00:09:20: we're in closed beta where actually seeing our beta customers convert to paid which is amazing were getting ready.

00:09:25: go to GA offer it everybody.

00:09:29: It's really fun to sit down at customer call and watch your customer like look at it and go wow!

00:09:34: This better than what you doing on her own Looks really successful, right?

00:09:38: Awesome.

00:09:39: I look at it and i go Oh that opportunity is poorly framed Mm-hmm.

00:09:44: oh That opportunity is closely worded with the key moment And It looks like a we had

00:09:49: dead in another episode if you're an expert.

00:09:52: Yeah So I see I See quality issues and so what do I Do catch?

00:09:57: I capture all those quality Issues.

00:09:59: I'm gonna Go do my error analysis.

00:10:01: how often did they show up In My product?

00:10:03: I'm Gonna try to create An eval To measure it But it's not this simple like we can all read about.

00:10:09: how do we do evals?

00:10:10: Okay.

00:10:10: Well, I gotta do.

00:10:11: a LM is judged to detect It but if in that single part of the tree there's actually four competing errors i Can't measure one without solving for the others.

00:10:23: This is like

00:10:24: A very

00:10:25: complicated data science problem.

00:10:28: Yes and i'm using fable To help me.

00:10:30: so We're talking at The time That were recording the latest Frontier Model.

00:10:35: that apparently was so great, that it was a national security risk.

00:10:40: to help me.

00:10:40: But its back in Europe!

00:10:41: Just if you're wondering, currently it's back in

00:10:45: the US as well.

00:10:47: It does really shallow analysis...it mixes up the failure modes like I'm not gonna let just solve this problem for me?

00:10:55: Its' not there yet.

00:10:56: So i can tell from personal experience my delivery is absolutely NOT free.

00:11:01: I am still spending days looking at data trying to like distinguish between this error and this error, trying to run experiments to figure out how to make those errors go away.

00:11:12: And this is for a product that works reasonably well for customers.

00:11:17: but I don't want to build a product which works reasonably good

00:11:22: for the customer.

00:11:23: so it's really helpful with... So yesterday i was sitting on a demo where basically It is an entire new marketing claim, new logo.

00:11:37: New design systems and designed for basically print and leaflets and digital and social media.

00:11:45: so it's a entire new design system right?

00:11:48: And the designer was so creative in demoing because he really created with the help of AI obviously super interesting, exciting that spikes curiosity tool.

00:12:03: That helps the people to explore The extremes of the design system for example.

00:12:10: so and is this billed to earn?

00:12:12: Not but it's built to learn And It's not even good to learn.

00:12:15: maybe its just like build two convince the internal stakeholders over or something like that.

00:12:23: But still, because it being cheaper to build such a thing he was able to build and put in hands of people who can now get curious about new design system and explore then they have way more informed discussions afterwards The extremes and guardrails that the design system maybe should have, they would never be able to have these in-depth conversations without ten or fifteen people playing with it.

00:12:53: And see what he can do right?

00:12:54: Without a tool... ...the designed person will not have substantial conversation.

00:12:59: so for this kind of use cases I think its amazing.

00:13:04: we can spun up such tools at short amount time.

00:13:08: I completely agree, but i would put all of that in the build to learn category.

00:13:12: Yes Of course Put All That In The Discovery Bucket Right?

00:13:15: Yeah!

00:13:15: And then when we talk about delivery is

00:13:18: free enough for dead bucket?

00:13:21: yes We're Talking About Production Quality Scalable Maintainable Code.

00:13:25: yeah and I don't even think were there with deterministic code I can see on the horizon We might get There With Deterministic Codes But I Don'T See On The Short-Term Horizon That Were Gonna Get Them AI Products.

00:13:37: And I think even

00:13:39: with AI products, it seems like we're there because I can create a prototype of an AI generated OST and is going to look pretty darn good.

00:13:48: You are gonna go wow this amazing product!

00:13:50: In fact i hear from product teams every day that they doing these themselves.

00:13:54: They don't understand why would buy the product for them Because they could get.

00:13:58: Claude generate an opportunity solution tree.

00:14:02: Okay great but you're at the prototype level, your building to learn.

00:14:08: You are not a production quality product that you can trust and rely on.

00:14:12: And I'm not saying this just because of what i am building.

00:14:14: This is true in every now possible episode That I record The products doing really well.

00:14:21: What's real clear Is that team has immense domain expertise about problem they solving.

00:14:28: They put it into hours of hard work to train the AI, to apply that domain expertise expertly.

00:14:38: And... The first sixty-to seventy percent?

00:14:41: Easy!

00:14:41: It's a prototype.

00:14:42: it looks reasonably good.

00:14:44: Closing that last thirty percent is months two years of work and its not free.

00:14:49: Its

00:14:50: like words missing in the blog post at ten per cent.

00:14:55: so they really land well

00:14:59: I think.

00:14:59: on our previous episode we talked about how many apps are being added to the app store, but they're not being used.

00:15:07: This is the same thing.

00:15:08: that's sixty-to-seventy percent

00:15:09: prototype."

00:15:10: That's great!

00:15:11: You can ship something really quickly... But I think customers are going to require that last getting to ninety five per cent before it will be useful.

00:15:18: Yeah

00:15:18: so the important things we don't subscribe to delivery as free narrative and we do want to reemphasize.

00:15:27: because

00:15:30: Yeah, and I'll say i have every incentive to tell you that delivery is free.

00:15:34: because when people Tell You Deliverers Is Free what's the back half of That Sentence?

00:15:38: They Say Delivery Is Free.

00:15:39: they Either Say one Of Two Things Delivery is Free Taste as all That Matters Or They Say delivery is Free Discovery As All That MatterS.

00:15:47: yeah okay tasteis baloney.

00:15:48: We've Already Talked About That in The Past yep uh?

00:15:51: I Would Love To Tell You That discovery Is all That matters.

00:15:54: it's not true.

00:15:56: Delivery is not gone to zero like they both matter.

00:15:58: They're both gonna continue the matter forever.

00:16:00: Yes, discovery might be coming more important and I love

00:16:04: software

00:16:04: to tell you that.

00:16:06: But delivery is not free.

00:16:07: yeah it's not for people It's not free.

00:16:10: thanks Teresa

00:16:13: Thanks Petra.