Ilan (00:00)
I'm mentioning a competitor on the podcast, that each

David (00:04)
Hey, welcome to Prompt and Circumstance. My name is David. And today we are going to do a deep dive on deep research.

Ilan (00:08)
and I'm Ilan.

Ilan (00:22)
Now is a great time to talk about our first sponsor for this week, Querio.ai

Querio is a powerful getting insights into your products. It allows you to ask natural language questions and then it will find the answers in your data. It saves hours a week for me, the product team I work with and our data team. And just for our listeners, they'll give you a two month free trial if you sign up today and let them know that David and Ilan sent

Ilan (00:52)
So David, before we get to the deep dive, did you hear about Lovable raising $200 million on a 1.8 billion valuation?

David (01:02)
That's huge. I think it was one of the biggest rounds in Europe. Yeah.

Ilan (01:08)
That's

what I've heard as well. Yeah. They are really killing it. You know, eight weeks ago, we reviewed their tool and I estimated their valuation, I think in the $300 million range. So pretty incredible in eight weeks, how far they've come.

David (01:31)
Yeah. Yeah. It really goes to show the game changing potential for these vibe coding tools.

Ilan (01:39)
Certainly though there are echoes of the dot-com bubble a little bit. There are lots of these companies getting crazy valuation. put AI in your name, vibe coding in your description and suddenly your evaluation just shoots up.

David (01:59)
Yeah, there is a lot of danger there. I really hope that companies learn from the past of what's happened with the dot com bubble. Yeah.

Ilan (02:11)
Certainly, though, you know, for those who are getting those valuations and taking advantage, you know, good for you.

David (02:21)
Yeah, exactly.

And there's similar news about Windsurf too, right?

Ilan (02:26)
Man, the Windsurf situation is really insane. I don't know how deeply you've read into this, David, but basically what happened is last Thursday, this feels like ages ago, it's literally been a week, ⁓ their exclusivity with OpenAI on a $3 billion acquisition ended. And the next day Google came in

Reverse acqui hired the founders of Windsurf and some of their top engineers to join Google and then created a perpetual license agreement for all of Windsurf's IP for 2.4 billion. this is something that is happening more often these days in the tech sector because of the increased scrutiny on mergers and acquisitions.

in the U.S. government. So basically you just hire the top people and then you license their product perpetually.

However, tables turn Monday, Cognition, who's the maker of Devin, which is an AI coding agent, acquired Windsurf. So they got everything. They got all the remaining engineers. It was done by the interim CEO who was appointed when the original CEO of left for Google.

So they got the entire IP of Windsurf and the remaining team all in one fell swoop because suddenly the valuation of Windsurf plummeted ⁓ since the top founders left.

David (04:14)
Yeah, you know some people on social media are saying that it was better to have you know acquired the company and got the people who on a daily basis work on the system rather than you know those who are who are higher up so we'll see we'll see how it goes the the drama definitely is there in the AI community

Ilan (04:37)
100 % Nothing like a good little bit of drama. And the last story I don't know if you saw this yesterday, but OpenAI released their ChatGPT agent

David (04:50)
Yeah, this is apparently going to be quite a game changer. ⁓ So it's going to make use of, I believe, Operator, what they've learned from that anyway, and a bunch of other capabilities in order to create this agent that a lot of the cheaper tiers can use. It's still paid, ⁓ but historically,

OpenAI has this $200 tier, right? And that has the exclusivity to some really cool stuff. But, you know, even if you're on, I think the plus tier, you would still get this agent.

Ilan (05:21)
Right.

Yeah, I think that it's coming very soon on the plus tier and you'll just be limited in how many agentic requests you can make per month.

David (05:42)
Yeah, do they have an agent that can do the dishes yet? Because I'm looking forward to that.

Ilan (05:48)
Yeah, you know, there was that post that was circling around about a year ago saying I want AI to do my dishes and laundry so I can focus on poetry and writing. I don't want it to do my poetry and writing so I can do more dishes and laundry. Well, let's get into it, David. Let's talk about deep research.

David (06:02)
Yeah, exactly.

Yeah, all right. So here we've got ⁓ three deep research contenders here and there are more but I these are the primary three. We've got Gemini on the left, we've got ⁓ GPT in the middle, and we've got Perplexity on the right. And what we're going to do today is we're going to just compare how they differ, how they behave ⁓ on a prompt that we're going to give them. And this is just one example that we're going to go through today of how

people can use deep research. There are many other ways that you can use it. So why don't we get started?

Ilan (06:45)
Yeah, they do

take a while to return. Deep research is a kind of prompt and walk away ⁓ activity. So we'll prompt here and then ⁓ we'll tell you a little bit more while these are thinking.

David (06:57)
That's right.

Yeah, so I'm gonna come here, turn on deep research for Gemini and over here for tools, I'm gonna turn on deep research for GPT and over here, instead of search, I'm going to choose research, which is deep research right there. All right, and up here, by the way, we've got a stopwatch. So let's see how long it takes. Let's go.

Ilan (07:28)
Now, one thing I'm interested in seeing, it looks like GPT has already done this, but they usually ask questions to clarify before they go off and start doing the research. Do you want to answer GPT there?

David (07:37)
Mm-hmm.

Ilan (07:54)
So David, why don't we talk a little bit about the use case for deep research that we are testing out here.

David (08:01)
Yep, and in a similar way, Gemini came up with a plan which we can review, but let's go ahead and get started.

Ilan (08:09)
Sounds good.

David (08:12)
Alright, off we go. Now, Perplexity is taking the bolt approach, I suppose, and diving right into it. Whereas we're seeing maybe some Replit-like behavior with Gemini, where it lays out a plan and says, Hey, I'm going to do this. Are we sure about it? Yeah.

Ilan (08:31)
That's right. All right, so David, why don't we talk a little bit about the scenario here that we are deep researching with these tools.

David (08:41)
Yeah, all right. So if we look at the prompt, we can see that what we're trying to do is understand the market for vibe coding platforms, right? So what we are asking it to do is basically a market analysis. Maybe this could be applied for product people who want to consider entering a market, not necessarily a vibe coding market, but any market. And so you want to understand how big is the market.

I want to understand who the contenders are. I want to understand how well they're performing where some of the weaknesses are. Maybe that's where we make our entrants, right? ⁓ And also what's new, right? What are some recent trends that are happening there? ⁓ I've also asked it to provide ⁓ Porter's Five Forces, which is a classic analysis of a market where you're looking at ⁓ things like ⁓

the ability to negotiate for suppliers, let's say, And so very curious to see how they fill that out.

Ilan (09:46)
Yeah, absolutely. So the, the power here of deep research is really diving deep into a topic. And it's something we mentioned in our last episode that this is something that a product manager should be doing more often. A really powerful tool in the discovery phase. You know, I don't know about you, but I have spent weeks before just

researching the market for a product that I'm considering. It's something that you do often in the background. You know, you find a 30 minutes here and hour there where you can kind of dive in and, try and understand what is the lay of the land. And these tools can really give you a broad

look who's out there, what do they do, as you said, what are the forces and analyze it for you as well.

David (10:43)
Yeah, and like many AI generated things, I would say that this is going to be a stellar first step in your analysis. I certainly hope that nobody takes the output here and then just copy pastes it into a presentation to to leadership.

Ilan (10:52)
Mm-hmm.

⁓ well, look, I see that Perplexity is giving us a chart that looks like something that's ready to copy paste.

David (11:08)
Sure, why not? So yeah, Perplexity seems to have wrapped up ⁓ and it's generating the report now. Let's see how deep it goes.

Ilan (11:11)
You

Yeah, once it's done there, we're going to evaluate it. So why don't we talk a little bit about how we're going to evaluate these reports.

David (11:30)
Yeah, so we're going to look at it across three categories. The first is going to be accuracy/completeness. The second is going to be the readability or the report quality and insightfulness. And then the third will be delighters. Like any good product people, we want to delight our customers. And so curious what the delighters are for these products.

Ilan (11:57)
Yeah, I'm looking forward to that. ⁓ and on the completeness, thankfully we have a podcast where we talk a lot about vibe coding tools. So, ⁓ we're looking for. Knowledge about a vibe coding that at least what we have, right. And hopefully extending beyond that. ⁓ so things like the lovable, fundraising round that we just mentioned.

or a few weeks ago we mentioned this vibe coding startup called Base 44, which got acquired by Wix. So these are kind of new events that have happened that we'd like to see in the report, as well as growth metrics, things like we've seen with Lovable or Bolt who've posted very publicly about how many users or what their ARR growth has been.

David (12:50)
Yeah, it's going to be great to dive into these reports and I really hope to learn a thing or two. So why don't we start with Perplexity? It's done now.

Ilan (13:00)
Great.

David (13:01)
So we have a fairly lengthy report, which is great. So global market analysis of Vibe coding platforms. All right, what are we looking at here? It's valued at $4.8 billion and projected to reach 30 billion by 2032. That's pretty good compounded annual growth, 27%.

So I do like how it shares its sources, how it provides the citations for, ⁓ you know, justifying its statements.

Okay, anything of note here? I do like the the executive summary. I appreciate that.

Ilan (13:41)
One thing that jumps out at me is the, it's saying that the current value of the market is 4.8 billion, but we just mentioned lovable raised 1.8 billion. That's just one tool. That's one of the smaller ones. We know that Replit and v0, well, Vercel who owns v0 are each valued at $3 billion. So just my, my quick math on three platforms is getting me to 7.8 billion.

David (14:04)
Mm-hmm.

Ilan (14:11)
So I do question a little bit the sources. I guess it is saying that that was in 2024, but with the speed that the market is moving, some information from 2024 seems pretty dated.

David (14:11)
Mm-hmm.

Yeah, things can be dated week over week. So here's a chart that is provided. Let's have a look at this.

So first off, I mean, thank you for a chart. I appreciate that. my initial reaction is that it'd be nice if text didn't overlap, but that's okay. So here it's counting GitHub copilot as part of the vibe coding market, which is fair. I mean, it's, it's arguable that it is. ⁓ It is missing.

Some of the other platforms though, like for example v0 by Vercel I don't see that on here. They are a sizable competitor to Replit and bolt and lovable

Ilan (15:17)
They're also using older information here. Cursors passed 100 million. ⁓ Lovable has passed 100 million as well, I believe, in ARR, at least 50 million in ARR.

Here they're showing cursor at 65 million, which is an older number and lovable at 7 million, which is way far off of their current ARR or even the information from just a few weeks ago.

David (15:44)
Yeah, this is quite dated, this information. However, I mean, even with v0, they were around in 2024, and I think it's a shame that they were not included in here. All right. What else do we want to have a look at here?

Ilan (16:04)
So this is interesting how it's broken down the market segmentation, ⁓ which is an important aspect here and explains a little bit of why it ⁓ included GitHub Copilot along with some of the other vibe coding tools that we were thinking about. So it's broken down no code, full stack builders. Those are things like lovable and bolt and Replit. ⁓

It has AI enhanced IDEs like GitHub Copilot and Cursor, ⁓ design first tools where I saw you just highlight v0, ⁓ as well as autonomous AI engineers.

David (16:42)
Mm-hmm.

Yep, right, like Devin, like what we just mentioned earlier. Yep.

Ilan (16:50)
That's

right. ⁓ So that's interesting. One thing that I see missing is some of the newer entrants into the pool of tools here. mean, vibe coding has very quickly become commoditized and we've seen platforms or applications like builder.io launch a vibe coding tool in the last few weeks and that's not mentioned anywhere.

David (17:20)
That's right. Yeah, it is missing that. Let's let's head down to the to the Porter's five forces. And then we can head over to Gemini because it looks like Gemini has finished. So let's see five forces. What is this chart?

Ilan (17:34)
Sounds good.

David (17:42)
Let me, ⁓ it scores it. It's a score of how each of the platforms does on the five forces. Interesting. Okay. ⁓ So let's have a look. Let's have a look at, ⁓ so competitive rivalry. It's high. That's fair. Yeah.

anything here of

Ilan (18:04)
I question medium high intensity for threat of new entrants after what I just mentioned about the commoditization of these tools.

David (18:15)
So as in you think that ⁓ the threat of new entrants is actually low or that it's actually higher? It should be like maxed out, yeah, I agree. The bargaining powers of suppliers. So supplier in this context would be like AWS offering cloud compute.

Ilan (18:23)
I would say that it's higher.

That's right.

Right. ⁓ As well as the underlying models that they're using

In this case, okay, I accept what they say. mean, there are a lot of options out there for the suppliers.

David (18:54)
Mm-hmm. Yeah.

Threat of substitutes. So by substitutes, what this refers to is you're not substituting one vibe coding platform for another. You're using a non-vibe coding solution to solve the same

Ilan (19:13)
I'm curious to read more here about what it talked about, but, you know, I see it mentioning established no code or low code platforms, like maybe bubble or super blocks and hybrid approaches. I wonder why it didn't, I see the first one is traditional IDEs with manual coding.

To me that's actually the, you know, that's the incumbent here that you're trying to replace.

David (19:45)
⁓ So the market segmentation, was looking at the providers in terms of how they segmented and we got some pretty good charts. All right, now GPT has finished. It said that it finished it in seven minutes, which I disagree with. My timer showed 13 minutes. take that for what it's worth.

Ilan (20:02)
Ha

David (20:10)
Alright, let's go over here to Gemini first because it was second to finish.

Alright, so executive summary. That's nice. It has emerged in 2025. I don't know about that. Is that right?

Ilan (20:28)
It's emerged as a true player or true alternative in 2025. would say that up until this year, you know, the underlying model capabilities were lower, but I would agree to quibble there. I'm noticing as well that the numbers that it's using are similar, though it's claiming they're from a different year, you 4.3 billion in 2023.

David (20:44)
Mm-hmm.

Ilan (20:57)
Um, and growing to 24 billion by 2031, those numbers way off. If you're looking in 2025.

David (21:05)
Yep, and so the way that Gemini documents its sources, the way that it cites its sources is with these little learn more buttons. So you click on this to learn more, and then there'd be one or multiple cards down here where you can click on it to view its sources.

So that's interesting, different way to get to go about it.

Ilan (21:27)
Definitely do you have a preference?

David (21:29)
I like the cards actually for readability purposes because as I'm reading through it, I find that maybe for me, I get a little bit distracted with the superscript numbers. So I find it a little bit more clean. How about you?

Ilan (21:41)
Hmm.

I actually lean the other way. I really like the footnotes because I like to see ⁓ in context of a particular sentence or a particular statement ⁓ where that's coming from. I guess when I'm reading, I have no problem glossing over ⁓ those footnote

David (22:03)
Yeah, fair. All right, let's have a look at the rest of the report. So key players. Hey, they mentioned the unicorn valuation of lovable.

Ilan (22:15)
Amazing.

David (22:16)
I don't see mention of...

of Wix here.

Ilan (22:20)
No, that's, that's true. Though I'm curious as we go down, if it mentions it anywhere else in the report.

David (22:26)
Mm-hmm.

Yeah, let's keep going down. Okay.

Ilan (22:31)
I do like that it ⁓ broke into this section, ⁓ the vibe coding paradigm shift. You know, that's a lot of what we talk about. Even the reason that we launched this podcast was the paradigm shift AI coding platforms compared to the traditional ways of coding. And I don't think that Perplexity's report quite highlighted how the market is moving and how quickly that shift has happened.

in the same way as ⁓ Gemini has done here.

David (23:01)
Yeah, it certainly provides a better narrative for somebody who doesn't know what vibe coding is about. ⁓ Like even here, it talks about an explanation of the term vibe, which is quite nice.

Ilan (23:09)
Right.

Does it mention Andrej Karpathy, ⁓ who coined the term?

David (23:21)
Yes, it does.

There you go.

Ilan (23:24)
Cool.

David (23:25)
that. So it was it was in February 2025. So there you go. I guess it was it's fair to say that that this had emerged this year. Boy, has it only been that short amount of time five months? Like, that's just transformative in five months. Yeah, you know, there are companies out there like legit companies with real revenue, where 100 % of the code is AI generated.

Ilan (23:34)
new entry to the market in 2025. Yeah.

haha

That is 100 % correct.

David (23:56)
Nice.

Ilan (23:58)
So interesting differentiators. Do you want to scroll down into the ⁓ market sizing growth projections?

David (24:09)
You know, I like how Gemini incorporates its citations into the actual prose. So like here it says a June 2025 report from Xpert Digital, for example, right? Whereas with Perplexity, it kind of just states the fact authoritatively with a citation. Yeah, I appreciate that.

Ilan (24:20)
Hmm.

Right.

One thing that I like here, which is also something that you see recommended a lot for product managers when we're writing is bolding the key numbers.

So your eyes get drawn right away to what's the key information. This is something that every, it's a tool that every PM can do in their writing to make it much easier to ⁓ go through and understand what are you trying to ⁓ please your attention to.

David (25:03)
All right. ⁓ So it talks about OpenAI's acquisition or attempted acquisition of Windsurf.

Ilan (25:11)
Though interestingly, it misses that ⁓ Google reverse Acqui- hire that we just mentioned.

David (25:11)
Good narrative.

You know, I wonder whether because Gemini is made by Google, whether there was some influence there.

Okay, let's have a look at.

Ilan (25:26)
Though you'd think that it would

have positively spun that news story, right? It would have just maybe mentioned the story of Google hiring the founders of Windsurf and getting a perpetual license for Windsurf's IP.

David (25:43)
I like that there is a table here in the output. you know, based off of the market segments, we can see how it's growing. I mean, the stark contrast is looking at software development overall versus any of these other categories.

Ilan (26:01)
That's right. So we're seeing software development at a 10 % versus generative AI overall at a what's that 43 % compounded annual growth rate ⁓ or AI code generators, know, by coding tools at 24%. So ⁓ very, very much a different growth scale here.

David (26:23)
Yeah, yeah, there's certainly some investors frothing ⁓ about the potential here. All right, so let's skip the segments. I'm curious about the Porter's five forces analysis. This is nice, it's quite thorough. ⁓ think it, okay, so it got lovable cursor, Replit, bolt, V0. And it didn't get anybody else.

Just these players here.

Ilan (26:54)
got it. I did include GitHub Copilot in its table as well.

David (26:59)
Although I think that our analysis was much better.

Ilan (27:03)
I tend to agree.

David (27:03)
All right.

Vibe insecurity, I like that. That's very important. ⁓ Some of the code that gets generated is not ⁓ very secure. So people who just straight up deploy to production might encounter some security challenges. One of the things that you had mentioned on your lessons deploying to production.

Ilan (27:10)
Mm-hmm.

David (27:27)
Okay, wow, this is lengthy. Okay, so five forces. Very good.

Yeah.

bargaining powers of supplies is very high they're saying

Ilan (27:36)
Hmm, interesting.

David (27:39)
it's thinking about, okay, it's really indexing on the ability of the LLM providers to set prices.

Ilan (27:49)
Got it.

David (27:50)
I mean, sure. I don't know if I agree with that because there are models out there, open source models that score very high on SWE-Bench, for example. And so, you know, it could take a little bit of work, but it is possible to replace a supplier with something open source.

All right. So getting down here, again, another table, which I appreciate because it's a nice summary of the five forces.

Ilan (28:16)
I also appreciate here that it has scored basically everything is high or very high. That was our analysis as well, where Perplexity was a little bit more muted.

in the moderate threat level.

David (28:33)
Yeah, this is certainly a ⁓ very volatile market to say the least. Okay, let's move on to GPT, see how that did.

Ilan (28:39)
100%.

Let's do it.

David (28:46)
Alright, so let's have a look at the report. ⁓ Let's full screen this. So it just kind of dove into it. There's sort of a summary up here. I suppose this is the executive summary. But an interesting difference between GPT and the other two.

Search interest jumped 6,700 % in a three month span. You're welcome from our podcast driving that traffic, of course.

Ilan (29:19)
That's right. I'm wondering if if it also used our podcast as a source for roughly 30 % of companies had a growth of 95%. Sorry, AI generate.

David (29:30)
Clearly a driver. All

So let's look at the market size. Well, that's not right.

Ilan (29:40)
Oh wow, it's off by two orders of magnitude from the tens of millions billions.

David (29:51)
that's that's fascinating. So it's saying, by 2030, the size of the market will reach upwards of 150 million only as opposed to the hundreds of billions.

Ilan (30:02)
Mm-hmm.

The other ones were in the tens of billions. So at least off by two orders of magnitude. I think you and I would argue that ⁓ all of them are off by at least an order of magnitude. yeah, is a ChatGPT. It's interesting. It got the same range of compounded annual growth rate, ⁓ but completely off in the numbers.

David (30:09)
Tens of billions. Yeah.

Yeah, I wonder whether its sources were ⁓ themselves misguided. the final comparison of how these things do citations.

So Perplexity uses end notes. Gemini uses a little accordion. And here in GPT, we have, I want to say pills. If I'm using the UX term right, the UI term, we have these little pills that ⁓ tell us the domain of the source. So in this sentence here, where it talks about compounded annual growth,

⁓ we have two pills that follow saying, ⁓ grandviewresearch.com and medium.com. So that's interesting, different, different design.

So what's interesting is that later in the report here, they're talking about how low code was already 13 billion in 2023. So maybe there's a distinction to be made here between AI coding assistance versus low code, no code development. And that's how it's splitting things up.

Ilan (31:18)
right.

David (31:41)
I mean, that's interesting. ⁓ So again, you know, this is one of those reasons why you take this as a starting point and not necessarily as something authoritative.

Ilan (31:52)
One thing that's interesting here in market segmentation is when we saw this in some of the other tools, ⁓ they were more prescriptive in their segmentation. here, ChatGPT more describes how you might consider segmenting the market. So it says you could segment by user type, by product tool category, by functionality use case, or by geography. ⁓

where the other tools actually broke down the segmentation. They chose one of these that they felt was ⁓ the most applicable. And then they broke down the market by that segmentation.

David (32:33)
Yeah. And in fact, this is the kind of segmentation that I was looking for. Because if you put yourself into a situation where you want to enter a market, you want to know, ⁓ is there a particular user type who I want to focus on? And ⁓ is there a user type in a geography that I want to focus on? You know, like all of the, you know, technically savvy UX people in Sweden.

Ilan (32:49)
Hey.

David (33:01)
I have been taken care of by loveable let's say, right? So, okay, I'm gonna, you know, skip that market.

Ilan (33:02)
Right.

David (33:07)
All right. Let's head down and see who ⁓ GPT picked up. We got lovable. We got Replit. These are lengthy descriptions. like that. and then we have other. ⁓ So cursor, GitHub Copilot, Windsurf, bolt.

And that's it. Boy, where's, where's Vercel? Where's V0?

Ilan (33:32)
Interesting it had Bolt and tempo.new which is interesting. So it did get one of the kind of smaller new entrants into the space and this is the first one of these that ⁓ mentioned Windsurf by name ⁓ in the comparison. They all picked up on cursor and some of them talked about Windsurf in their analysis of the market but not as ⁓ a player or competitor.

David (34:02)
Yeah, in the summary, it mentions V0 down here, but it wasn't in great detail. I would have liked to see a bit of a paragraph on each of them.

Okay, so OpenAI reportedly in talks to acquire Windsurf. So no update on that, even though that news is, how old is that news now? Two days? Come on, it's ancient history.

Ilan (34:24)
It's

actually almost a week old at this point. ⁓ I also do see it talking about some newer entrants though into the market. ⁓ In the previous paragraph, it talks about Magic Dev and Poolside, some stealthy AI code startup that have raised hundreds of millions of dollars. So maybe a little bit more thorough ⁓ in its trend analysis.

David (34:30)
Is it? Yeah, look at that.

I want to point out that, ⁓ you know, with, with all of these, they, they have citations from Reddit as well. So various different social media platforms, when it comes to looking at what the market thinks about these tools, we're not going to dive into that, but I think that's, it's worthwhile to call out when you're doing your market analysis. All right. So with the five forces, we've got competitive rivalry, very high or high, I should say.

So that's fair. New entrants, moderate to high. I mean, does require significant capital. I mean, yeah, fair.

Ilan (35:36)
I actually disagree with that. think that the rise of all these vibe coding platforms has significantly reduced the amount of capital that you need to be a new entrant into the market. ⁓ I feel like I'm practically ready to announce a Prompt and circumstance vibe coding tool that we could launch.

David (35:56)
It would be quite the ⁓ irony if ⁓ lovable made a competitor to lovable that then outsold lovable.

Ilan (36:06)
Exactly.

David (36:07)
Moderate threat of substitutes. Okay. So sounds like ⁓ it's saying status quo is the threat, which is

I like this call out here to Zapier as a way of simplifying some automation there. So that's interesting.

Ilan (36:25)
True, there are lots of incumbents in the automation market, low code, no code market. We mentioned bubble superblocks before. if your company already has purchased a license for one of those tools, then it's going to be pretty hard to dislodge that player. You're going to have to be going against all of the existing automation flows that have already been built.

David (36:54)
Yeah. I, I really want to call this out. ⁓ we, didn't, have a look at this deeply with any of the other reports, but I just want to call this out as like a significant force in, the vibe coding market. you and I talked about how, ⁓ if it doesn't work in one tool, you can just pick up your code and go to another tool. And there is so little.

Ilan (37:15)
Mm-hmm.

David (37:18)
⁓ effort and risk involved with that. ⁓ And so to me, the bargaining power of buyers, ⁓ you know, imagine if you just paid for one month on Lovable and it cost them however much to acquire you, right? The cost of customer acquisition ⁓ and sure, 20 bucks USD ⁓ and then you're gone because you weren't able to do what you wanted to.

Ilan (37:37)
Mm-hmm.

Great.

David (37:47)
⁓ so this is, think a major force in the web coding industry.

Ilan (37:53)
I did see an estimate from, I believe Aakash Gupta, the,

I'm mentioning a competitor on the podcast, that each

request in Bolt cost them three cents. So that's pretty high. Imagine if in whatever product you make, it costs your company three cents for each login or each action that you took within the application. So not only is the cost of acquisition ⁓ high or part of the pricing, but ⁓ it's a particularly costly model in terms of ⁓

how users are interacting with the tool.

David (38:41)
Yeah, that's a really good point, Ilan. you know, with SaaS product, the operating margin is historically very big. It's nice. It's nice and fatty. Um, and with the vibe coding tools, that is actually quite a bit leaner because of the cost to, run the models. And I actually wonder whether that ends up resulting in

Ilan (38:52)
Mm-hmm.

David (39:10)
Maybe some ⁓ vibe coding platforms using open source ⁓ models and self hosting it. So that's, know, they, they, they, they cut out the middle man if it were right. So, so they directly pay for the compute. They don't need to pay the premium of saying, yeah, this is, you know, something that we use from, from anthropic, let's say.

Ilan (39:33)
Yeah, definitely interested to see that, especially with a few weeks ago, Mistral announcing their coding tool. ⁓ You know, there are some of these other models that are available now. That being said, Claude reigns supreme for code and Gemini as well is known as one of the frontier models for ⁓ its capability to do really complex coding tasks.

Ilan (40:02)
David and I are setting up our tiers, so it's a great time for me to tell you about our other sponsor for this week, Co.lab.

Colab is a product that will help and a you could run as a side This is great for product managers or any professionals who are looking for a way to set up their side business but don't really have enough time to do the initial market testing and

They'll hand you a working product and they'll even find your first paying customers.

click on the link below and let them know that David and Ilan sent you for $250 off Validate and Build.

David (40:38)
Alright, cool. So with that, ⁓ let's go through the three things that we wanted to look at and see how our contenders performed. So on the accuracy/completeness, in my mind, I thought that Gemini did the best job.

Ilan (40:56)
Yeah, I also thought that Gemini did the best job. It had recent news like, ⁓ lovable raising their $200 million series a, it had a wider breadth of tools in its analysis. though that being said, while it's the winner for me, ChatGPT did a good job of at least mentioning.

some of those other entrants or newer entrants in its overall report.

David (41:26)
That's right. So GPT for second place.

Ilan (41:28)
That's right. One thing I'd also mention is one of our evaluation criteria was any of these mentioning the acquisition of base 44 by Wix, which we mentioned on the podcast a few weeks ago. And none of them had that in there. And in my opinion, that's a significant event demonstrating the growth and value of these tools.

David (41:56)
Yeah. So what it means is that our audience should continue to listen to us and not rely on these AI tools. Yeah.

Ilan (42:02)
Hehehehe

Exactly.

How about on the report quality, readability and insightfulness?

David (42:13)
I appreciated the ⁓ tables that Gemini created, you know, like the the five forces summary table. It, ⁓ it also had a good ⁓ narrative in terms of the sections throughout its report. And you know, I contrast that with what GPT output, which had neither tables nor charts. It was pure text. So at least Perplexity, you know, gets

Ilan (42:36)
That's right.

David (42:41)
some points there for having something visual.

Ilan (42:44)
Yeah, I was going to say the same thing. I also appreciated Gemini's narrative structure and the tables that are included at the end of each major section with the comparisons. ⁓ But I like the visual nature of what Perplexity put together. The only thing I would say is that some of those bar charts didn't necessarily make sense like on the five forces.

David (43:10)
Yeah, I've never seen Five Forces expressed with a bar chart.

Ilan (43:14)
Wonder what Tufte would think of that. right, and finally we have Delighters. Who wins for you in terms of Delighters?

David (43:17)
Yeah, right, exactly.

Well, there were two delighters for me that stood out. The first was the sort of planning.

that Gemini made where it said, look, here's what I'm going to do. Anything that I should change before I dive into that. And ⁓ GPT did a version of that where it asked a few questions. So kudos to them for that. ⁓ I do appreciate the speed of Perplexity. So maybe if you're not doing such deep research, you're doing, I don't know, like mildly deep research. You just have a ⁓ thought in your mind.

Ilan (43:57)
You

David (44:00)
⁓ that is more sophisticated than just a simple search would do, maybe Perplexity is the one for you for that.

Ilan (44:09)
Those are good ones, David. Another delighter for me was the visual aspect of Perplexity, having some charts. Y'know you said earlier on, hope you don't just copy paste something that you see in here into a presentation. But if the information is accurate and they're getting good sources, then that is the kind of thing that you could copy paste into a slide deck.

if you want to, if you're using some of this information to help make a case for a product. You know, not to say that you should just rely on this report. I think we described in pretty thorough detail why this is just a starting point, but that doesn't mean that you can't use those, those visuals.

David (44:56)
Yeah, absolutely. so, hey, congrats to Gemini on overall being the vibe winner of our comparison here. So again, this is just one example of what people can do with deep research. Even though it's not absolutely perfect, ⁓ it is an excellent ⁓ start, right? With the... ⁓

Seven minutes or 13 minutes depending on how you count GPT with that amount of time spent Where you can you know go grab a coffee? ⁓ You have a very thorough initial analysis of what you need So whether it is entering a new market whether it is investigating a problem that you want to solve whether it is ⁓ Looking at what what news applies to a company that maybe you want to apply for

Ilan (45:25)
hehe

Mm-hmm.

David (45:50)
⁓ in a job application. Getting up to speed very rapidly on a topic is in your fingertips now.

Ilan (45:51)
That's right.

David (46:00)
Go and use this for your research purposes.

Ilan (46:05)
Amen. And the last thing you mentioned this last week, David, in our three tips for AI, but I just want to rehash it, which is that if you are trying to do some of this initial analysis, you want to digest this information in a different way. We strongly, strongly recommend importing this information into NotebookLM and then generating a podcast. You know, that'll give you the 20 minute breakdown.

so that you can get the summary, kind of keep that in your mind. And then you can always read the report and check the sources for more information.

David (46:42)
That's right, absolutely.

Ilan (46:44)
All right, with that, that's the end of our deep dive on deep research.

Thank you so much for listening or watching. You can follow us @pandcpodcast on all of the socials and like and subscribe, leave us a comment, leave us a review, let us know what you'd like us to talk about next and we'll see you next week.

David (47:05)
See you next week.