Back to Library

Episode 98

THE VIBE CODER REVIEW: BASE44 VS. EMERGENT, LOVABLE & RIPLET

Keywords entrepreneurship, AI tools, Base 44, Emergent, Lovable, Riplet, Firebase GenKit, MVP, development tools, technology trends Summary In this episode, Dalton Anderson discusses various tools for…

Jan 6, 202600:45:05
Listen to the episode00:45:05
0:0045:05

Episode content

Episode Story

Replit Agent App Builder: Workflow, Git, and Deployment

Replit Agent plans, writes, debugs, and improves software inside Replit. This profile covers its checkpoints, Git, auth, publishing, testing, and limits.

Article · 1 min

Lovable AI App Builder: Cloud, GitHub, and Ownership

Lovable is an AI full-stack development platform with managed cloud, GitHub sync, security tools, and external hosting paths. This profile covers its tradeoffs.

Article · 1 min

Google Genkit: What It Is and What It Is Not

Genkit is Google's open-source framework for AI-powered and agentic applications. It is not the hosted Firebase Studio app builder tested in E098.

Article · 1 min

Firebase Studio App Prototyping Agent: Sunset and Scope

Firebase Studio's App Prototyping agent generated Next.js apps from prompts. New workspaces are disabled, and Firebase Studio is scheduled to shut down in 2027.

Article · 1 min

Emergent AI App Builder: Workflow, Credits, and Export

Emergent is an agentic full-stack app builder with previews, deployment, integrations, and GitHub workflows. This profile covers evidence, credits, and limits.

Article · 1 min

Base44 vs Emergent vs Lovable vs Replit: A Dated Test

A January 2026 hands-on test of Base44, Emergent, Lovable, Replit, and Firebase Studio using the same 122-page insurance app brief.

Article · 1 min

Base44 AI App Builder: Features, Export, and Limits

Base44 is a managed AI app builder for generating, testing, and publishing applications. This profile covers its workflow, backend, GitHub sync, and limits.

Article · 1 min

Research & Analysis

Vibe Coding Benchmark Method Research Note

A useful AI app-builder benchmark must answer a decision rather than manufacture a universal leaderboard. The decision might be which tool best supports a team's internal

Research Note · 1 min

One-Shot Evidence Ladder Research Note

For E098, one shot means one builder instruction before judging the initial result. It does not mean one sentence, no prior work, no clarification, or no setup. The share

Research Note · 1 min

E098 Vendor Current State Research Note

E098 was recorded on January 5, 2026. Product capabilities, plan gates, names, and documentation have changed since the run. The episode's results must remain dated.

Research Note · 1 min

E098 Lab Artifact Research Note

E098 is supported by the canonical transcript and two recovered screen-recording files in the original Dropbox production folder. The filenames are `screenshare_hd-2026-1

Research Note · 1 min

AI Prototype Production Gate Research Note

A preview becomes a production candidate only when the team can show evidence for what the system does, who owns it, how it fails, and how it is operated. The gate should

Research Note · 1 min

AI Generated App Architecture Research Note

An AI builder must either choose an architecture or leave important decisions unresolved. The generated interface can hide both cases.

Research Note · 1 min

AI App Builder Portability Research Note

Code access is one part of portability. A usable exit also requires the application data, schema, files, identity records, secrets inventory, scheduled work, integration

Research Note · 1 min

Field Notes

What One-Shot App Generation Actually Proves

A one-shot AI app build can prove initial instruction-following and visible interaction. It cannot prove security, correctness, scale, or demand.

Evergreen · 1 min

How to Evaluate an AI App Builder Before You Commit

Evaluate AI app builders by workflow, code, data, identity, testing, security, deployment, portability, cost, governance, and the exit path.

Evergreen · 1 min

How to Benchmark Vibe Coding Tools Fairly

A reproducible protocol for comparing AI app builders across requirements, function, security, accessibility, portability, time, cost, and variance.

Evergreen · 1 min

From AI Prototype to Production: A Release Gate

Move an AI-generated prototype toward production with evidence for requirements, architecture, security, testing, accessibility, operations, cost, and ownership.

Evergreen · 1 min

The Hidden Architecture of AI-Generated Apps

AI app builders make hidden choices about data, identity, permissions, state, integrations, deployment, operations, and exit. Review them before users arrive.

Evergreen · 1 min

Full episode

TranscriptSearch or read the full conversation.

E98 THE VIBE CODER REVIEW_ BASE44 VS. EMERGENT, LOVABLE & RIPLET

Transcript

Dalton Anderson (00:00.782) Welcome to Venture Step podcast where we discuss entrepreneurship, industry trends, and the occasional book review. Believe it or not, it's 2016. And that means New Year's resolutions, goals, starting the year off strong. And there's never been a better time than to start now on your side biz, your entrepreneurial endeavor, or your passion project. There's a couple of tools out there that would help you facilitate these goals.

In this episode, we're going to be doing an overview of some of those tools. Those be Riplet, Lovable, Base 44, Emergent. And I also tried out Genkit by Firebase, which is part of the Google Cloud offering from Google. But if you are unfamiliar with any of the stuff that we're talking about and you want additional information or you're curious, there's several topics that were covered in various episodes. Episode 37, Unlocking the Power of Cursor AI.

your AI powered coding companion. In this episode, I built a to-do list app using REST APIs and Go, Go Lang. Episode 77, the windsurf story.

from Billion Dollar Dreams to Founder Back Laughs, Windsurf, it talks about how basically how not to run your vibe coding company and what they did to their employees. And it's overall a very, how to say, it's not the right way to do things. And it's interesting, it's a lot of drama. And it was a brutal story of how not to treat your employees or your people or.

people that definitely work at your startup and to be a leader. Episode 58, The Rise of Grok, Rapid AI Development and the Future of Thinking. In this episode, I had used Grok to create a prompt and do some video games. And then I also used Riplet to make a video game and I got crushed by the AI. If I remember that all correctly. So those were the related episodes. In this episode, we're going to just be going through a refresher of

Dalton Anderson (02:13.902) these apps that are used to do these things and they have matured over time. It's been quite a while now. And this day and age, like six months is a couple of years in progress. So I was quite surprised how they have evolved. One company I did not try out was Bolt. Just I wasn't happy with the last one and I already have quite a list going for myself right now.

And so it just didn't make the cut. Maybe in the future, I'll try it out, see how does. And I have this large spreadsheet of the prompt that I gave each of these apps. So I would be able to test it with the controlled prompt. Okay, so the first thing, there's gonna be some terminology that I'm gonna be using. there's one is one shot. All of these are one shot. What does one shot mean?

One shot means you provide instructions to the system and you see how well it performs. So if you ask your kid, hey, I need you to wash the dishes and then put them in the dishwasher, run the dishwasher and then wait an hour and a half and then take the dishes out of this washer, dry them, put them away. You need to put the dishes in this area, the forks and knives and utensils over here.

and the cups go here. And depending on how your kid is, they might just straight up forget. I would be the one who'd forget. I'd just forget that there was stuff in the dishwasher and never do it. I'd put it in the dishwasher and have good intentions and I'd just forget and I'd probably be playing video games. But if your kid's more organized, then they might figure it all out and do it and they might get 80 % of it. So they might put the cups in the wrong spot or.

the plates they put in the right spot, but maybe the plate orientation, like they put the small plates where the big plates are, whatever. And so that's one shot. It's providing instruction and then seeing the results. So I'm not constantly, you know, hey, do you have dishes in the dishwasher? You know, is, are they dry yet? Is the dishwasher done? yeah, it is done. Thanks. Okay. Is

Dalton Anderson (04:36.942) Is the cups in the right area? Looks like the cups go over there. None of that. So it's just one one time instruction. And then there's other stuff like. Full stack versus orchestrator. Majority of these that I tested previously were not full stack, but now appear to be full stack. But the classic and leader of full stack would be base 44, which is an Israeli based company, and then their competitor is emergent.

What had prior been not a full stack, but more of a quick MVP demo thing would be lovable in Riplet, but previously they weren't and now they are. So they have their own database. They create the database. They do the hosting. So they've got it all going on. And it's quite interesting how quickly those two companies have matured versus Base 44 and Emergent. Emergent is new in my eyes, but I was always familiar with Base 44.

So the first comparison is gonna be the full stack generators. And as I said, Lovable and Riplet are full stack-ish and then Emergent and Base 44 market themselves as full stack generators. So Base 44, what I liked about them is I liked that it was very quick to get started and it was simple, it was straightforward and overall clean.

Very clean.

didn't feel like it gonna be a hassle to get started. it felt pretty seamless. And that's great when you're trying to get started. Like the whole point of using these programs is to try and get something quickly. And from those iterations, you can figure out whether or not the idea is worth pursuing in the first place.

Dalton Anderson (06:35.224) But if you have to log in, set up your APIs, then you have to figure out some permissions and manage all this stuff. I might as well just do it locally. Like why am I bothering with all these things? So it was really nice how quickly you were able to get started. And I'll share my screen.

screen share.

Dalton Anderson (06:59.822) So this is Base 44, their website. I'm already logged in. And if you go over to this, I already have an app deployed. But I think if I go over to Workspace and then you click over here, you should be able to see the prompts and how it works. So on the left side, it's almost a 1 third panel on each of these apps. So 1 third of it is going to be this.

chat that you have with their agent or their chat, whatever they want to call it. But basically their workflow agent and their creator, the facilitator, however they market it. So in this example, I provided a prompt, a very long prompt. It's one hundred and twenty two pages of information and it just goes over a whole bunch of stuff. One thing I liked about base was it asks questions right away. So it says, oh, I see this is a comprehensive

document and these are the things that you're trying to solve for and what overall like this is lot of information. What is your priorities here? And then I specified my priorities and then it created the plan and then it wrote all the code after asked the questions. And that was overall pretty decent and the results are good. So let me go to preview and I'll share.

Dalton Anderson (08:30.894) I swear I had it up.

Dalton Anderson (08:41.797) I'll just take the...

Dalton Anderson (08:47.18) Here we go. It's a workaround. Okay, so this app that was created, I customized the UI where maybe the CSS, where I determined the colors I wanted. I wanted it to look a little different. One problem with these type of apps is they all look the same if you don't specify how it should look. What I mean by that is you might get a white background.

and some kind of dark blue and a turquoise kind of color. And they look pretty bland. Like it's quickly you quickly know that, hey, this is not a. Website that is unique, it just looks like everybody else, which you don't necessarily want. It depends what you're going for, but I didn't want that. I want to have some kind of different color and flavor to.

My dear.

But sorry, I wasn't sharing. was sharing the wrong screen. OK, so now I'm sharing the right screen. And let me zoom in a little bit because I know that I've got some commentary that, hey, it looks a little small when you're sharing. So I'm zoomed in. This is the dashboard. And the general sense of what I was trying to do is make it easier for people to manage their risk. There's a lot of information in different places, and it's difficult to figure out where it's at and when it's due and who to contact. And when you contact them, they need a lot of information. It's hard to gather all up.

So making the process overall easier. One thing that it did mess up on was these quick actions. You can't see them. It's just, it's white. And when you highlight it, then you can see it. But so it's new submission, add property, upload loss run. And then there are little panels, menus that you can go through. Just quite nice. It created a dashboard. All this stuff is

Dalton Anderson (10:48.332) demo data, but I think it did a pretty good job of describing what you might be looking for in this scenario. It's not what I'm looking for, but I overall thought that the demo and the MVP on a one shot was pretty decent where it was able to not only create a dashboard for me in these quick actions for the portfolio, you can enter a new submission and you can just type in, I think I have an address up here. If I go to

Dalton Anderson (11:21.365) square.

Dalton Anderson (11:26.55) You can't see what I'm doing, but I'm just looking up an address that I have. So this address and then the city is New York.

Dalton Anderson (11:38.029) to you.

Dalton Anderson (11:42.38) zip code.

Dalton Anderson (11:53.218) just moved here so this is the county

Dalton Anderson (12:01.134) I'll leave it blank, see what does.

Dalton Anderson (12:06.478) And so it enriches the property so so-called enriches the property doesn't necessarily enrich it.

Dalton Anderson (12:18.478) But then you can select the occupancy type. we'll say retail. Or not retail, sorry. This is an apartment. It might be mixed. I would say mixed used. I'll say this is a large, ABUVT and NYC and has.

Thank

Dalton Anderson (12:51.166) or parking lot.

Dalton Anderson (12:55.999) and a event space.

Dalton Anderson (13:03.862) It does have a burglar alarm. It does have a sprinkler. It does have a fire system type of sprinkler. Hmm. Preaction. All right. Now continue. And they could fill out all these different things. I mean, this is. This isn't the right way to have everything formatted, but on a one shot basis, this is pretty good. Like I'm pretty impressed with what it.

decided on its own to create. it has a different type of coverages. And has the general liability, has occurrence and general aggregate. I think it's really good, really good stuff. And then you go in and you generate the limits and you generate the report, which is pretty sweet. So let me just do some demo stuff real quick. Let's do this. Let me just do a fast expiration date. Should be.

Let's see, one, 15, 27. And if you're listening to the podcast and you're not on Spotify, this stuff is all being videotaped and I'm sharing my screen. But basically what I'm doing is I'm filling out, let's do 100 mil.

Dalton Anderson (14:29.391) 13 % so I'll do 13 million. Replacement costs AOP deductible 50k on a 118 million makes sense for occurrence. Hi. annual sales I don't know. 12?

Overall units 500, brain palerol 350 I have no idea.

Continue. So this is it creates. It creates the accord for you, the accord 125, 126, and the 140. These things are used to, oh, my legal name is missing. These things are created for, man, that's a bummer. These things are created for you to create a submission in the entrance space. These things are required to fill out. And so this,

creates a submission and the submission creates these documents that typically you'd fill out manually and they no longer have to do that. But overall really impressed with Base 44. I think in terms of demoing the capabilities and like what to do on a system basis, Base 44 stands out 100 % and absolutely crushes their competitor when it comes to the whole situation.

I would say when it comes to...

Dalton Anderson (16:05.209) the actual product that it delivers in a one-shot basis versus their competitor, which I'm gonna start sharing in a second. So sharing to them instead. Emergent. Emergent has, I would say, the best UI. The UI was brilliant. You would type in the prompt and then it would have this whole highlight around the chat box.

chat box and do all these cool things and these animations. But as far as actual product goes and what was provided, it's quite limited. And I'll show you in a sec. I I know that I'm sharing both screens at the same time. But in a general sense, I provided this massive prompt, as you could see. And then from there, it asks questions.

Dalton Anderson (17:02.319) It asks questions after the prompt. And I was I was hoping that it would provide a really good output given the amount of questions it was asking compared to base 44. But overall was disappointed with. The actual output of. The product versus the amount of time I spent answering the questions, so I asked some questions like, like.

I'm not going to start doing any work until you answer my questions, you answer the questions, and then it has more questions, which I don't mind the questions, but if you're providing additional context, and that means you're detailed information that takes time to provide, and you've got to think about it, and they're thought provoking, you would appreciate it more if one, you didn't run out of credits.

Like I ran out of free credits and actually I'm at negative 12 credits negative 12 cents is what I'm what I'm at. I'm on the free tier and I've got negative negative point 12. So it's interesting and it didn't necessarily provide a great product and then asked me to update. So let me preview what I have. See if I can go open up a new tab share this tab instead. So.

This is just a landing page. didn't do anything that was interactive. It didn't really have any good demo information. It has this stuff where you can just test your enrichment and the enrichment doesn't, I I don't expect it to work, right? But then you go to create the submission and that's it. It's done. No longer, no longer are you able to move forward. So it has one button.

and then as a landing page and that's all it has. And so it's more of a landing page and a, hey, look at us, this is what we can do versus an actual demo of what the capabilities might be. And so overall, very disappointing given the amount of time I devoted to creating the app versus

Dalton Anderson (19:20.429) what was given, answering all the questions, waiting for the build, which took a long time. This is a massive request, and so it may have just cut it short, given that I ran out of credits, and then just carried on with a landing page. Maybe they built 10 % or 5 % of what they were supposed to build, and then carried on.

I didn't necessarily care to find out given that I'm out of my credits and I'd have to pay to try something out. And what I did try out, I didn't get anything out of it. Okay. So those were the two competitors. So there was Emergent and then there was Base 44. By a long shot, Base 44 provided a much more comprehensive output and great demo, great MVP.

Overall really impressed with what they provided with giving my prompt. mean, I have high expectations because my prompt was very detailed. I mean, it's hundred and twenty two pages worth of information. Talks about the JSON schema and the information should be implemented and which APIs to use and as keys like it has everything that you might want as an AI system to create something of value. And so I'm expecting value and base 44. I feel

that it provided something of value for the time that I put in. The time that put in was pretty minimal. mean, if you're talking about creating that prompt, it took me forever. That's something I've been working on for a while. It's not necessarily a prompt, it's really documentation. Documentation on ideas and next steps, sprints, the whole nine yards.

Dalton Anderson (21:15.193) But as I said, bass 44 crushes emergent in this test. Maybe in the future they'll be closer. And I say that because Lovable and Riplet, Riplet wasn't close really to Riplet last time I did these tests. Riplet was far above Bolt and Lovable and has really been my preference for testing, quick testing.

for this kind of space, whereas I don't need something heavy duty like base 44, I just need something to just test and get started real quickly. Riplet was my choice, but after testing both Riplet and Loveable, I didn't have any issues with Loveable and I thought that overall it was great. And I liked some of the availability of

choosing your UI style before the prompt start and how you wanted some of some of the branding to be. And they're leaning on more, I would say, the aspect of customization and uniqueness versus other websites like you just you just get you just get going. And I appreciate that because a lot of these tools, if you don't specify the UI, it's going to look the same like if I gave the same prompt to all these tools.

most likely the colors are going to be similar. The landing page is going to be similar. And they almost look the same thing because it's input output, input output. It's just math. It's just numbers, just LLMs. They're just doing what they do and they're doing what they're comfortable with and they're doing what they learn from. And so most likely they're just going to be using what's most common with what's the most common websites types. There you go. And that's what you get.

So you don't get uniqueness, which is something in what is becoming bland, people will appreciate. That's the way to stick out. Okay.

Dalton Anderson (23:24.162) lovable in Riplet. So let's start with Riplet, which I enjoyed quite a bit. So let's do that. Let's share my screen.

Dalton Anderson (23:37.947) think I said, let's start with Lovable and then I said start with Riplet. We're gonna start with Lovable. So Lovable, as I said prior, has matured pretty quickly. Before Riplet was the leader in this kind of quick MVP space, Lovable was another company that was emerging. They got some funding and then they started competing and maturing rapidly.

And overall, Ripley was first to the space and then Lovable is a fast follow-ish. mean, the space is pretty new, so there's not necessarily some senior advisor in this space like this coveted, oh, I've been here first. But overall, liked the panel that Lovable had and I didn't really have any questions. It just kind of just got started.

just did its thing and didn't ask any questions and it created what would be a great MVP. Love the styling, it's unique and it has a nice landing page and then it also has this developer API and so it has a cord 140, a cord 126, cord 125 API.

And it explains what the API, which each API does, like the Cord 140 is a property API, building specs, construction class, protection class, and subjects of insurance. So I thought that was really cool. And overall, the style, it talks about how it traditionally is approached versus with Risk OS, which is what AI decided to call the app. And then you can then do a demo or a

look at the workflow, the workflows right here, and then you can scroll up and then you go to a demo, which we will try out.

Dalton Anderson (25:38.091) And then you just kind of type in your address. So I'm going to type in my address and it goes to enrichment.

And it so-called enriches the property. It has the same enrichment characteristics every time. So it's just dummy data that fills in, but it does look cool. And then it does quotes, which is not necessarily how commercial property would ever work. You'd never have an instant quote. It's got to be underwritten.

Underwritten, I said underwritten, but there needs to be an underwriter involved. So it would not work this way, but then you could sign in for your quote, sign for your quote, and then boom. But overall, I really like the approach of how the website looks. It's clean, it's usable. Great. Is it better than base 44? No.

make that clear, Base 44 is definitely a favorite for the information provided it and the implementation of just the overall demo. So, lovable versus Riplet. And so, let me share my screen.

Dalton Anderson (26:57.392) So Riplet, I was pretty unhappy with. You could see how many errors I was having. it was saying, I sent in an error. It said, okay, your website's ready to be made to test out. So I went to go test it and they had their own authenticator and it wasn't working and it gave me a 404.

gave it the error and asked it what's going on. And then it says, agent has encountered an error while running. We're investigating the issue. And I was, okay, I waited a little bit and nothing happened. And I was like, what were your findings? And then it goes to this whole thing. It says it's fixed it and then it broke again. And then I had to fix it again and it broke again. And so overall, I was less than thrilled with the process.

I really spent like 30 minutes trying to get the website to run, whereas Lovable was just one shop stop. I think overall processing time took. I don't know if it has the time on here.

Dalton Anderson (27:58.662) So at nine.

Dalton Anderson (28:04.422) Not sure.

But it did take a long time for a Ripplet to get going. But once it got going, it did create a nice demo. Do I like it? I think I like it more than lovable, but this is not a unique website. This is an everyday website that you would see if you use this app. that happens when I click analytics and then it breaks.

So I have go back to preview.

Dalton Anderson (28:44.037) Might be cooked here,

Dalton Anderson (28:50.417) share. Wow. Yeah, we're replicates running some issues here.

OK, so I'm back. So don't click on the analytics tab, but they've got the dashboard and I created a submission and you can see it. You can have your portfolio. You can see your little submission that you have. And then you have your dashboard. This is more for an agent, I would say, versus what I was trying to do. I think base 44 makes a lot more sense what they provided. And then I don't know. This.

definitely is way less of a comprehensive MVP than what I'd be expecting. mean, it's really, it looks nice, right? But it's not very functional.

It's not very functional, it's a apartment with event space.

Dalton Anderson (30:00.666) do like the AI classification results, and I think that's nice. It does change every time.

Dalton Anderson (30:11.065) And then it gives you this market selection, which is not how this would work. And then you close, generate a cord, and then it has all these cords to download. And then you can view it in your portfolio.

A decent. If I had to pick between the two, Lovable did a way better job. Briscoe, the Riplet preview looks, I would say more professional and enterprise-y, but I think that's more of the selection I asked Riplet to do. I asked it to be like brutalism and use these colors. And so this was something that was created on my preference.

whereas I didn't specify anything with Ripplet and this is what you would typically get. You get this white page with these green accents and blue secondary colors. It's quite common. Lovable definitely had a more comprehensive output when it comes to the API demo, the developer kit for the API, the interactive demo, the ability to bind, the ability to reset your demo.

Overall comparison between different products the traditional way of doing it versus this product Great great success. I really appreciate it. It looks great. How does it compare to base 44 base 44 crushes everybody? Simple base 44 does a great job great great job. So the last one is really about Google and Google's gen kit

GinKit is different than any of these other apps. So I would describe GinKit as more of an orchestrator versus an actual generator of a full stack environment where Firebase already has the back end. When I say the back end, like the authentication, easy setup of API keys if they are already integrated into Firebase. And so there's less hurdles to build stuff from scratch. But the GinKit

Dalton Anderson (32:21.393) doesn't build things from scratch. It will build the UI for you, but it's not building the backend. as far as like live demos, like Base 44 or the stuff that Lovable put together doesn't do that. So let me share my screen. I couldn't one shot this, which sucks because...

I was running into issues when it was generating. It was you can see the security. This security check recommended. It says please review your applications dispendencies if you're running react or next JS applications immediately update to the latest stable versions react 19.2.1 or the latest version of next JS and it has a bunch of versions. I'm using the gen kit. The gen kit is giving me the old.

versions of Next.js and it's causing issues with my app. But that being said, I did generate three apps.

Dalton Anderson (33:28.111) And each of them have good aspects about them, but also each of them have drawbacks as well. No live demos. It gives you an idea of how the UI might work. There's nothing about the backend being generated, which is also kind of a good thing and also kind of a bad thing. It depends how you want to look at it.

How could that be a bad thing? Well, as these apps become more complex and they have users hitting the page and the website and the database, it has scalability issues when you are having the AI decide your architectural. Making architectural decisions is a big deal. I think an architectural decision

is a one door decision. There's not an easy way to reverse what you're doing once you get started because all your code's tied to it. The UI is tied to it. There's the functionality. And once you have all the data set up, then you've got database, database dependencies and these jobs for reports, financial reports, dashboards. Once you get integrated with this, this decision is really hard to change your decision.

And so the best way to do that is to minimize the changes that you need to make. And you should spend a lot of time on your architecture, like your database architecture and overall your architecture in general, but especially your database architecture. And that's why architects get paid a lot of money, because it's important to make the right decision the first time. Because if you don't, it takes a long time to fix the problem and it costs a lot of money. So you'd rather just pay for a great employee.

to help make those decisions. OK, but this is the first version of what Firebase Genkit created. And one thing I didn't like about Genkit was it doesn't ask you any questions. Even if you ask it to ask you questions, it just gives you the app blueprint, tells you what it's going to do, and then you say prototype this app or not. And that's it. You don't get any choices. It is what it is. You can change the UI.

Dalton Anderson (35:47.955) ish or the coloring or some of the features. as far as asking, having the system ask you questions about what's going on, what was your intent, what do you want to do? It doesn't do that, which is a bummer. I feel that's a pretty big gap. If you're trying to generate something, you should ask questions like lovable and replace should be asking more questions. But so if you look at this.

I'm not going to click on many things because if I do, it's going to error out. And when it errors, it kind of just dies. So if I click New Submission, it opens up the submission page, which I think is going to error.

Let's see, let's see, let's see, let's It's loading.

loading I'll give it five, four, three, two, one. All right, it didn't work. So that was that one. Let's try risk OS.

Dalton Anderson (36:46.706) Funny stuff, funny stuff for sure. Like I felt that.

Firebase should provide a pretty good...

a pretty good product here on their generation, but it just seems that they're so far behind. It's clear that it's not a focus because what it's providing is not where it needs to be for sure. It's all over the place. just doesn't have that interactability. You can't even navigate the pages correctly. And so,

Dalton Anderson (37:30.322) It's just disappointing the offering here. I thought that it'd be way better for the company that Google is and where they want to position themselves as the enterprise giant and AI specialist that operates at a scale that other companies can't. They don't have the compute, they don't have the talent, they don't have the environment, the infrastructure, the availability of data. So they've got all these things going for them.

And they're killing it in some regards, but a lot of times they're lacking in this instance, they're lacking and they got to do better. I don't know. I I spent a lot of time trying to get a good product come out of this, this whole gen kit. And this is the best thing I got. And it's just this weird hodgepodge of panels on this dashboard that aren't

are non-functional. So if I use this.

It doesn't do anything, causes an error. It has data populated, but you can't interact with it at all. mean, there's no backend, so you're not gonna be able to, but it doesn't really provide a good picture or make me wanna get started. Like I get excited about the Base 44. The Base 44 gold standard, it's great. Where this is not great. I tried three times to get something

at least mediocre, right? I'm best buddies with Google. I love Google. But what are you doing? What are you doing, Google?

Dalton Anderson (39:15.9) But overall, that's my review. My thoughts are if you're trying to get something quick, you can't really go wrong with Lovable or Booklet. If you're trying to do something legitimate and it's not just like a question of whether or not this would be interesting and you want to actually implement something and not necessarily scale with users, but just try something out.

and get closer to an MVP, I would suggest Base 44. Base 44 killed it. Their one shot was great. You had this interactability. You've got the demo. We've got some integrations. It just, it was great. Love it. And it had a pretty good UI implementation. You don't see it myself. I just looked up cool color schemes for businesses. That was one of them.

And so just took that and gave it to AI and asked it to add some extra colors and some swag and aura. And that's what I got. So love, love, love base 44. The gen kit has potential needs work. Maybe it does better when you have a back end built already. And that's really probably what it's used for. But I don't know. I mean, if you're going to have

the product available to prototype without the backend, then ask the user, hey, are you okay with GenKit generating the backend for you? It's nothing permanent. It's just quick generation to prototype this. What do you think? Yes, no. Or if you don't want to do that with GenKit, have the user generate the backend because what they provided, not good enough.

Not good, not even close, not even close to being good enough. So they need a lot of work there. But base 44, as I said, gold standard, best demo, best interactivity, best UI, overall great vibes, easy to use. Speed was the fastest. I don't know, there's really nothing to say bad about base 44. was great. It was seamless. Riplet had issues getting started with the authenticator, the

Dalton Anderson (41:39.155) Hort was turned off. couldn't accept information. Couldn't authenticate the user. It took me 30 minutes to do. Lovable didn't ask any questions, but overall liked the UI approach. I stopped the recording on accident. I would have been devastated.

Leveled in asking questions, but overall liked the UI approach.

had a really cool website, had the API docs, it created API docs for the accords, it had a nice demo.

Emergent?

It's marketed as a robust tool that has multiple agents. Like one is a prompting agent that prompts your prompt, creates a more robust prompt. And then they have an agent that codes and then they have a QA agent that tests for the bugs. And that's how they market it. It's like a three-part agent. And maybe it's good, but it seems like it's really expensive for what it does because you can't even get a prototype out without.

Dalton Anderson (42:49.779) without running out of credits and they only generated me a landing page. No interactivity, no web pages, just a landing page. That's it. So for the price, not worth it. Base 44 is great. And another thing I forgot to touch on, that's great with emergent or base 44, especially the base 44, you can download the code and or put it in your GitHub. So you're not locked in to

their system or their software, you can build what you want to build and then leave and then start that in like an enterprise environment like AWS or Google Cloud or Microsoft Azure and move on with your life. And then you could prototype little things in base 44 and then bring them over into the enterprise side of things later on.

Dalton Anderson (43:47.133) But...

Dalton Anderson (43:55.124) I enjoyed. The process of trying these out, what really surprised me was lovable and the amount of progress they've made and the maturity that the company is at now with their offering. One thing I wasn't surprised about was base 44. It's they're hyped up and they're hyped for a reason. They live up to the hype. Emergent was supposed to be a competitor, not even close, but really surprised about.

lovable and their progress. Riplet will let me down. I still like Riplet a lot, so I would still say eight one and two doesn't really matter. Maybe try both out and see which one works for you. And hmm. Overall, hope you enjoy this episode and you think it was useful. And if so, like, subscribe, check in next week. We will be having a guest on the show. And I'm really excited about it. Been trying to been.

planning this for weeks, especially difficult during the holiday season. And wherever you are in this world, good afternoon, good evening, and good morning. Thanks for listening and listen in next week. Goodbye.

SourcesFollow the evidence trail.

E098 Sources

Preserved episode evidence

[[E98 - Transcript]] is the canonical raw lab transcript. It supports Dalton's one-shot test design, the long shared requirements document, the generated insurance-submission prototypes, observed interface behavior, errors, time and credit friction, his recording-date rankings, and his warning about architecture and database decisions.

The transcript is evidence of one dated run under one account, prompt, plan, model, product version, and task. It is not a permanent ranking, a security audit, a production-readiness assessment, or proof of typical performance.

Two original screen-recording files were recovered from the episode's Dropbox raw-assets folder:

ArtifactSizeDurationEvidence use
screenshare_hd-2026-1-5__21-52-57-CFR.mp4103,571,958 bytesAbout 20 minutesBase44 and Emergent interfaces and previews
screenshare_hd-2026-1-5__22-9-41-CFR.mp4124,089,391 bytesAbout 42 minutesLovable, Replit, and Firebase Studio interfaces and previews

The filenames establish a January 5, 2026 capture date. Five representative stills are preserved in Assets. Their visible claims and demo data remain unverified until independently tested.

The original 122-page requirements file, exact plans, model selections, receipts, and machine-readable run logs were not recovered from the episode folder or immediate Dropbox raw-assets folder. The comparison can publish as a dated artifact-backed lab report, but not as a reproducible benchmark.

Current vendor sources

docs.base44.com/Getting-Started/Quick-start-guide

Base44's documentation supports its current positioning as an AI-powered app builder with design, database, authentication, user-permission, hosting, code-view, ZIP export, and GitHub options. Plan requirements and export behavior are dynamic.

docs.base44.com/developers/app-code/local-development/github

Base44 currently documents two-way GitHub synchronization on eligible plans, along with important ownership, branch, permanence, and version-history conditions. The transcript's broad claim that a user can simply leave the platform needs these qualifications.

help.emergent.sh/articles/272715-features-and-tools

Emergent's current documentation describes natural-language full-stack web and mobile development, previews, testing, deployment, custom agents, and GitHub integration. These are first-party product claims, not independent evidence of production readiness.

help.emergent.sh/plans-and-credits

This controls current credit and plan details. Prices and allowances require a publication-date check and should not be treated as evergreen.

docs.lovable.dev/introduction/getting-started

Lovable's current documentation supports its chat workflow, Agent and Plan modes, version history, Lovable Cloud or Supabase backends, GitHub connection, and publishing.

docs.lovable.dev/integrations/github

Lovable documents two-way GitHub sync, local work, collaboration, and alternative deployment. Repository ownership and connection behavior should be described exactly as current docs state.

docs.replit.com/learn/build-with-agent

Replit's current guidance describes Agent as a builder that can plan, write, explain, debug, and improve applications. It explicitly emphasizes specificity, planning, context, review, testing, and checkpoints.

docs.replit.com/category/replit-apps

Replit's application documentation supports current cloud project, collaboration, version-control, publishing, visibility, deployment, and storage concepts.

Google and product-name corrections

firebase.google.com/docs/studio/get-started-ai

The transcript calls Google's prototype experience "Genkit," but the interface described is the Firebase Studio App Prototyping agent, which used Genkit flows for AI features. Genkit itself is a developer framework, not the no-code builder Dalton was comparing.

As of the current review, Google says new Firebase Studio workspace creation and new user signup were disabled on June 22, 2026. Existing workspaces can continue until the planned March 22, 2027 shutdown. Any public page must treat Dalton's result as historical product evidence, not a current buying option.

firebase.google.com/docs/studio/migrating-project

Google's sunset and migration guide supports the shutdown schedule, deletion warning, and migration routes to Google AI Studio, Antigravity, GitHub, ZIP, Firebase CLI, or other environments. It also says normal ZIP export does not include agent chat history and explains how shared workspaces can retain pointers to the original owner's backend.

firebase.google.com/docs/genkit/overview

Google describes Genkit as an open-source framework for full-stack AI-powered and agentic applications. It should not be ranked as if it were the same product category as the hosted prompt-to-app builders.

Software assurance and evaluation sources

csrc.nist.gov/pubs/sp/800/218/final

NIST SP 800-218 supplies a secure software-development vocabulary covering organizational preparation, software protection, well-secured software, and vulnerability response. It supports the production and benchmark guides, not a claim that any E098 prototype is secure.

owasp.org/www-project-application-security-verification-standard

OWASP ASVS supplies testable web-application security requirements. A project should select and version the requirements appropriate for its risk.

w3.org/TR/WCAG22

WCAG 2.2 supplies technology-neutral accessibility success criteria and supports the accessibility gate in the benchmark and production guides.

web.dev/articles/vitals

Core Web Vitals supplies user-centered measures for loading, responsiveness, and visual stability. Performance evidence still requires representative environments and workloads.

Test interpretation boundaries

The transcript repeatedly says "Riplet" or "Ripplet." Context and the described product indicate Replit. Public work should use Replit while noting the transcript's transcription error in provenance records.

The shared input was approximately 122 pages of product and technical material. Calling that a single prompt can obscure the substantial prior work. A one-shot label describes the number of builder interactions before the initial result, not the total human effort.

The generated insurance examples contained demo data, incomplete behavior, and incorrect commercial-insurance workflows. A visually coherent dashboard, generated forms, or a preview URL does not prove correct underwriting logic, compliant forms, secure authentication, durable storage, or a production architecture.

The transcript's speed, credit, failure, and quality comparisons cannot be generalized without recording plans, models, settings, start and end times, retries, manual edits, generated scope, and evaluation criteria. Vendor products change quickly.

Draft-time checks

For any current comparison, rerun every tool in fresh accounts with the same requirements, time and budget caps, success criteria, and captured artifacts. Check pricing, credits, model selection, code ownership, export behavior, hosting, database, authentication, secrets, logs, tests, rollback, security controls, deletion, and support on publication day.

Publication remains unauthorized. The recovered raw videos require a separate media review before any public upload. The five stills are usable as internal first-hand evidence and proposed public illustrations after final editorial review.