July 21, 2025 • 20 mins
Reflection AI Launches Asimov Code Comprehension Agent
Thinking Machines Lab Raises $2 Billion to Develop Safer AI Systems
Cognition Acquires AI Startup Windsurf, Boosting AI-Powered IDEs and Coding Agents
The Browser Wars Are Back - And AI is the Weapon
#startups, #AI, #browserwars, #codecomprehension, #AIsystems, #IDE, #technology
Mark as Played
Transcript
Episode Transcript
Available transcripts are automatically generated. Complete accuracy is not guaranteed.
(00:00):
Welcome to Innovation Pulse, your quick no-nonsense update covering the latest in startups and
(00:09):
entrepreneurship news.
First, we will cover the latest news.
Reflection AI's Asimov tool, Streamline's code comprehension, Thinking Machine's lab
raises billions pre-product and cognition acquires Windsurf for competitive edge.
After this, we'll dive deep into Perplexity's groundbreaking Comet browser and its potential
(00:31):
to revolutionize internet interactions.
Stay tuned.
Reflection AI has launched Asimov, a groundbreaking tool designed to enhance code comprehension
for engineering teams.
Asimov addresses a critical challenge.
While engineers spend most of their time understanding existing code, rather than writing new code,
(00:54):
there hasn't been a tool specifically focused on this until now.
Asimov helps engineers navigate complex code bases and captures the intricate business
logic within them.
What sets Asimov apart is its ability to act as a single source of truth.
For engineering knowledge, it ingests entire code bases, architectural documents, GitHub
(01:18):
threads and chat histories, effectively storing and recalling important information, much
like a seasoned team member with deep tribal knowledge.
Asimov's memory system allows team-wide knowledge sharing and can be updated by engineers, ensuring
that valuable insights aren't lost over time.
(01:39):
Utilizing a multi-agent architecture, Asimov excels in retrieving and synthesizing information
from large code bases, providing coherent answers to user queries.
This makes it a preferred choice over existing tools like CursorAsk and Claude Code.
Asimov's potential to evolve with reflection, AI's own models promises to enhance its performance
(02:04):
further, making it an indispensable tool for engineering teams.
Now we're about to explore their ambitious goals.
Thinking Machines Lab, an AI startup founded by Mira Murati, a former open AI executive,
has raised approximately $2 billion at a valuation of $12 billion.
(02:29):
The funding round was led by Andresen Horowitz, highlighting Murati's strong influence in
attracting investors, despite the company being launched only in February and having
neither revenue nor products yet.
The startup is set to release its first product soon, which will feature a significant open
source component beneficial for researchers and startups developing custom models.
(02:55):
Thinking Machines aims to build AI systems that are safer, more reliable, and applicable
to a broader range of applications than competitors.
Nearly two-thirds of its team consists of former open AI employees.
Murati's departure from open AI and subsequent launch of Thinking Machines mirrors a trend
(03:16):
where ex-open AI executives like Dario Amade and Ilya Sutskeva have also launched their
own AI ventures.
Investor interest in AI startups remains robust, contributing significantly to a surge in US
startup funding in 2025.
(03:36):
Cognition, the startup behind the AI coding agent Devon, has acquired AI coding startup
Windsurf.
This acquisition follows Google's high-profile reverse-aquihire that took Windsurf's CEO
and key leaders, leaving the rest of the team behind.
The deal between Cognition and Windsurf was rapidly assembled and finalized within a weekend.
(04:01):
Cognition is acquiring Windsurf's Intellectual Property and AI-powered IDE, along with all
employees not hired by Google.
Windsurf had reached $82 million in annualized recurring revenue with 350 enterprise customers
and hundreds of thousands of daily active users.
(04:22):
Cognition plans to integrate Windsurf's capabilities into its own offerings, which
already include Devon, an AI agent designed to automate coding tasks.
This strategic acquisition enhances Cognition's ability to compete with giants like Open AI
and Anthropic by offering both AI coding agents and an AI-powered IDE.
(04:45):
Windsurf's team will continue developing their IDE in the short term, and the company
will regain access to Anthropic's clawed AI models which were previously cut off.
With this acquisition, Cognition strengthens its position in the rapidly evolving AI coding
landscape.
And now, pivot our discussion towards the main entrepreneurship topic.
(05:14):
Welcome back to Innovation Pulse.
I'm Dana, and today I'm joined by my brilliant co-host, Jacob Lasker, who just sent me the
wildest article about someone claiming people will pay $2,000 for a single AI prompt.
Jacob, please tell me you're not about to justify spending too grand on asking ChatGPT
a question.
Oh, this is so much bigger than ChatGPT, Dana.
(05:36):
We're talking about perplexity's CEO, basically declaring war on Google Chrome, while simultaneously
trying to reinvent what a web browser even is.
And yeah, that $2,000 prompt?
He's dead serious about it.
Wait, hold up, perplexity.
Aren't they that search engine that's supposed to compete with Google?
(05:56):
Now they're making browsers?
Exactly, they just launched something called Comet, and here's the wild part.
It's not just a browser with AI bolted on.
The CEO, Aravind Srinivas, thinks the browser is actually the secret weapon for building
AI agents that can do real work.
Like he's talking about agents that can recruit Stanford graduates who worked at Anthropic,
(06:20):
draft personalized emails to all of them, and manage the entire follow-up process.
Okay, that actually sounds terrifying and amazing at the same time, but I'm stuck on
something.
Why does it have to be a browser?
Can't AI agents just, you know, connect to apps directly?
This is where it gets fascinating.
So there's this whole technical battle happening behind the scenes.
(06:44):
You've got people working on something called MCP, Model Context Protocol, which is basically
trying to let AI talk directly to every app and service, but Srinivas thinks that's backwards.
How so?
Think about it this way.
When you want to book a flight, you don't call the Airlines API.
You go to their website, you're already logged in, you can see everything, and if something
(07:08):
goes wrong, you can just take over and fix it.
The browser is this environment we've spent decades perfecting for humans to interact
with digital services.
Oh, so instead of building some complex system where AI has to authenticate with hundreds
of different services, it just uses the browser the same way you and I do, with all our accounts
(07:29):
already logged in and accessible.
Exactly.
And here's the clever part.
When you switch from Chrome to Comet, it imports all your logins, extensions, everything with
one click.
So suddenly, the AI has access to your Gmail, your LinkedIn, your Amazon account, but it's
all happening on your computer, not on Perplexity's servers.
(07:53):
That's actually really smart from a privacy standpoint.
But wait, if I'm understanding this correctly, they've basically built a version of Chrome
with an AI Assistant sitting next to every web page?
That's the sidecar feature.
But here's where it gets wild.
Srinivas shared examples of people using it to unsubscribe from spam emails, create Facebook
(08:14):
ads, even control smart home devices.
One person used it to find the exact timestamp in a YouTube video where someone said something
specific instead of watching the whole thing.
Okay, that YouTube thing is actually brilliant.
But I have to ask, if AI is just extracting the key information from websites, what happens
(08:35):
to all the companies that spend millions making their websites beautiful and engaging?
Are we just turning the entire web into a database?
Srinivas actually addressed this, and his answer surprised me.
He thinks brands will become even more important, not less.
Like if you're the verge, people will still want to know what the verge thinks about something.
(08:56):
But the way they get that information might change completely.
Hmm, I'm not entirely convinced.
If an AI can just extract the key points from any article instantly, why would I ever actually
visit the website?
The business model of the web is built on people actually showing up and seeing ads.
And that's where things get really interesting.
Srinivas made this pointed comment about how Google uses AI to figure out the perfect moment
(09:22):
to show you ads on YouTube to maximize engagement time, while Perplexity is using AI to save
your time by getting you exactly what you need.
The incentives are completely different.
Oh snap, that's a fundamental philosophical difference.
Google makes money when you spend more time on the web, but Perplexity makes money when
(09:42):
they save you time.
That's actually kind of profound.
Right.
And this is where that $2,000 prompt starts to make sense.
If an AI agent can do three hours of recruiting work in five minutes and do it better than
you could yourself, suddenly paying a premium for that kind of cognitive labor doesn't
(10:03):
seem crazy.
I mean, people already pay consultants hundreds of dollars an hour.
But I'm curious about the technical reality here.
Has anyone actually gotten these AI agents to work reliably for complex tasks?
This is where Srinivas gets refreshingly honest.
He admits Comet is still pretty brittle.
The interviewer, Alex Heath, tried to get it to list everyone who follows him on Twitter
(10:27):
that works at Metta, and it only found one person when there should have been many more.
But Srinivas' bet is that the next generation of reasoning models, GPT-5, Claude 4.5, will
get them over the reliability hump.
So we're still in the demo magic phase, but the vision is compelling enough that people
(10:47):
are willing to deal with the bugs?
Exactly.
And there's this interesting chicken and egg problem.
The product needs to be reliable enough that people will use it, but it also needs enough
users to improve through real-world testing.
Right now, Comet is invite only and limited to their 200 a month subscribers.
(11:08):
$200 a month for a browser?
I know, it sounds insane.
But think about it, if this thing can actually do the work of a virtual assistant, suddenly
$200 a month seems reasonable.
That's less than hiring someone for even a few hours a week.
Fair point.
But what about the competition?
I can't imagine Google is just going to let some startup eat Chrome's lunch.
(11:31):
Well, here's the plot twist.
There are rumors that OpenAI is building their own browser too.
And get this.
The rumors started circulating just two hours after Perplexity announced Comet.
Coincidence?
Probably not.
Oh, that's fascinating.
So we might be looking at a whole new browser war, but this time it's not about speed or
(11:52):
features.
It's about which AI can best automate your digital life.
And there's another wild card in play.
The Department of Justice might force Google to spin off Chrome.
If that happens, it could completely reset the browser market and give startups like
Perplexity a real shot at gaining market share.
Wait, that would be huge.
(12:14):
Chrome has something like 65% market share.
If Google suddenly can't bundle it with Android or pay billions to make it the default browser
everywhere.
Exactly.
Suddenly, there's room for innovation again.
But here's what I found most intriguing.
Srinivas thinks the browser is actually bigger than the chatbot race.
He argues that ChatGPT has basically won the conversational AI market, but the browser
(12:39):
is where the real value will be created.
That's a bold claim.
Why does he think that?
Because the browser is where you can build true end-to-end workflows.
A chatbot can help you write an email, but a browser-based AI can draft the email, send
it, track responses, schedule follow-up meetings, and manage the entire relationship over time.
(13:01):
It's the difference between a tool and an assistant.
I'm starting to see the vision, but I keep coming back to this learning curve issue.
When I first got an iPhone, it was intuitive.
I could figure out how to use it without reading a manual.
But with these AI agents, it sounds like you need to completely rethink how you interact
with your computer.
Srinivas acknowledged this.
(13:21):
He said even he has trouble remembering to use the AI features instead of doing things
the old way.
But he thinks it'll follow the same adoption curve as chatGPT, confusing at first, then
suddenly indispensable for early adopters and eventually mainstream.
Speaking of adoption, what about mobile?
Most people do more browsing on their phones than their computers these days.
(13:44):
They're planning mobile apps, but here's the challenge.
On iOS, every browser has to use Apple's WebKit engine.
So Comet on iPhone won't actually be able to use the full Chromium engine they've built
all their AI features on top of.
It's another example of how the platform owners control the innovation.
That's such a good point.
(14:04):
Apple and Google have these invisible choke points that most people never think about.
If Apple decides they don't like AI browsers, they can basically kill them on mobile.
And that brings us to the business strategy.
Perplexity has 30 to 40 million users for their search product, so they're hoping to
convert some of those people to try the browser.
(14:25):
But the real prize would be getting device manufacturers to make Comet the default browser,
which Srinivas admits is nearly impossible because of Google's existing deals.
So they're basically hoping to grow through word of mouth and hope that regulatory action
opens up more opportunities?
Pretty much.
Though there was this interesting tidbitters about Apple potentially being interested.
(14:48):
Eddy Q apparently said nice things about perplexity during the Google trial, and Srinivas
would love to work with Apple on Safari or Siri integration.
That would be a game changer.
Apple's been struggling with AI compared to Google and OpenAI.
A partnership with perplexity could give them a real edge, especially on the search side.
(15:09):
And speaking of partnerships, did you catch the part about Mark Zuckerberg trying to recruit
Srinivas to Metta?
Apparently, Zook is paying nine-figure compensation packages to top AI researchers.
Nine figures?
That's like a hundred million dollars plus.
No wonder there's this talent war happening.
But here's what I don't understand.
(15:31):
If you're perplexity and you're trying to build a sustainable business, how do you compete
with those kinds of offers?
Srinivas made an interesting observation.
He noticed that anthropic researchers aren't getting poached as much as others.
His theory is that money alone isn't enough.
You need mission alignment and the feeling that you're working on something meaningful
(15:53):
that's actually succeeding.
That makes sense.
If you're working on cutting-edge AI research and your company is growing fast and your stock
options are appreciating, why leave for even more money at a place where you might not
have the same impact?
Exactly.
And perplexity seems to be hitting their growth targets.
Srinivas mentioned they want to IPO around 2028 or 2029 after hitting a billion dollars
(16:18):
in revenue and achieving profitability.
A billion in revenue?
That seems aggressive for a company that's essentially giving away most of their product
for free right now.
But this is where the browser strategy could really pay off.
If they can get millions of people using Comet Daily, and if those AI agents actually become
reliable enough for complex tasks, then usage-based pricing starts to make sense.
(16:42):
Maybe you don't pay $200 a month, but you pay $20 for that one recruiting task that
would have taken you three hours.
You know what?
That actually sounds reasonable.
I've definitely spent $20 on less valuable things than saving three hours of work, but
it assumes the reliability gets there.
Which brings us back to the fundamental bet perplexity is making, that the next generation
(17:06):
of AI models will be dramatically better at multi-step reasoning and task completion.
If they're right, we might be looking at a completely different internet experience
in just a few years.
So what should our listeners be watching for?
What are the early signs that this vision is actually happening?
I'd say watch for three things.
First, how quickly the reliability improves.
(17:30):
Can these AI agents actually complete complex tasks without human intervention?
Second, how other companies respond, especially Google and Apple?
And third, whether people actually change their behavior.
Do they start using AI assistance for tasks they currently do manually?
(17:50):
And maybe also watch your own behavior.
Are you finding yourself asking chat GPT or other AI tools to help with things you used
to just Google?
Because that's probably the first step toward the kind of AI-powered browsing perplexity
is betting on.
Great point.
The shift is already happening in small ways.
And if Srinivas is right, those small changes are just the beginning of a much bigger transformation
(18:13):
in how we interact with information and get things done online.
Well, I know I'm definitely curious to try Comet when it becomes more widely available.
Though I have to admit, the idea of an AI watching everything I do online is still a
little creepy, even if it's supposed to be private.
Fair enough.
But remember, these AI assistants are only seeing what you explicitly ask them to help
(18:36):
with.
It's not like they're constantly monitoring everything.
They're more like having a really smart assistant who only pays attention when you tap them
on the shoulder.
That's a reassuring way to think about it.
Alright, so next time you're frustrated by how long it takes to research something online
or book a complicated trip, remember that there might be AI agents just around the corner
(18:59):
that could do all that work for you.
The question is whether you'll be willing to trust them and pay for them.
And whether the company's building them can solve the reliability puzzle before the competition
heats up too much.
This is definitely a space worth watching closely.
Thanks for diving into this with me, Yaakov.
Until next time, keep your pulse on innovation.
(19:20):
See you next week, everyone.
(19:45):
Stay tuned for more updates.
Welcome to Innovation Pulse, your quick no-nonsense update covering the latest in startups and
(00:09):
entrepreneurship news.
First, we will cover the latest news.
Reflection AI's Asimov tool, Streamline's code comprehension, Thinking Machine's lab
raises billions pre-product and cognition acquires Windsurf for competitive edge.
After this, we'll dive deep into Perplexity's groundbreaking Comet browser and its potential
(00:31):
to revolutionize internet interactions.
Stay tuned.
Reflection AI has launched Asimov, a groundbreaking tool designed to enhance code comprehension
for engineering teams.
Asimov addresses a critical challenge.
While engineers spend most of their time understanding existing code, rather than writing new code,
(00:54):
there hasn't been a tool specifically focused on this until now.
Asimov helps engineers navigate complex code bases and captures the intricate business
logic within them.
What sets Asimov apart is its ability to act as a single source of truth.
For engineering knowledge, it ingests entire code bases, architectural documents, GitHub
(01:18):
threads and chat histories, effectively storing and recalling important information, much
like a seasoned team member with deep tribal knowledge.
Asimov's memory system allows team-wide knowledge sharing and can be updated by engineers, ensuring
that valuable insights aren't lost over time.
(01:39):
Utilizing a multi-agent architecture, Asimov excels in retrieving and synthesizing information
from large code bases, providing coherent answers to user queries.
This makes it a preferred choice over existing tools like CursorAsk and Claude Code.
Asimov's potential to evolve with reflection, AI's own models promises to enhance its performance
(02:04):
further, making it an indispensable tool for engineering teams.
Now we're about to explore their ambitious goals.
Thinking Machines Lab, an AI startup founded by Mira Murati, a former open AI executive,
has raised approximately $2 billion at a valuation of $12 billion.
(02:29):
The funding round was led by Andresen Horowitz, highlighting Murati's strong influence in
attracting investors, despite the company being launched only in February and having
neither revenue nor products yet.
The startup is set to release its first product soon, which will feature a significant open
source component beneficial for researchers and startups developing custom models.
(02:55):
Thinking Machines aims to build AI systems that are safer, more reliable, and applicable
to a broader range of applications than competitors.
Nearly two-thirds of its team consists of former open AI employees.
Murati's departure from open AI and subsequent launch of Thinking Machines mirrors a trend
(03:16):
where ex-open AI executives like Dario Amade and Ilya Sutskeva have also launched their
own AI ventures.
Investor interest in AI startups remains robust, contributing significantly to a surge in US
startup funding in 2025.
(03:36):
Cognition, the startup behind the AI coding agent Devon, has acquired AI coding startup
Windsurf.
This acquisition follows Google's high-profile reverse-aquihire that took Windsurf's CEO
and key leaders, leaving the rest of the team behind.
The deal between Cognition and Windsurf was rapidly assembled and finalized within a weekend.
(04:01):
Cognition is acquiring Windsurf's Intellectual Property and AI-powered IDE, along with all
employees not hired by Google.
Windsurf had reached $82 million in annualized recurring revenue with 350 enterprise customers
and hundreds of thousands of daily active users.
(04:22):
Cognition plans to integrate Windsurf's capabilities into its own offerings, which
already include Devon, an AI agent designed to automate coding tasks.
This strategic acquisition enhances Cognition's ability to compete with giants like Open AI
and Anthropic by offering both AI coding agents and an AI-powered IDE.
(04:45):
Windsurf's team will continue developing their IDE in the short term, and the company
will regain access to Anthropic's clawed AI models which were previously cut off.
With this acquisition, Cognition strengthens its position in the rapidly evolving AI coding
landscape.
And now, pivot our discussion towards the main entrepreneurship topic.
(05:14):
Welcome back to Innovation Pulse.
I'm Dana, and today I'm joined by my brilliant co-host, Jacob Lasker, who just sent me the
wildest article about someone claiming people will pay $2,000 for a single AI prompt.
Jacob, please tell me you're not about to justify spending too grand on asking ChatGPT
a question.
Oh, this is so much bigger than ChatGPT, Dana.
(05:36):
We're talking about perplexity's CEO, basically declaring war on Google Chrome, while simultaneously
trying to reinvent what a web browser even is.
And yeah, that $2,000 prompt?
He's dead serious about it.
Wait, hold up, perplexity.
Aren't they that search engine that's supposed to compete with Google?
(05:56):
Now they're making browsers?
Exactly, they just launched something called Comet, and here's the wild part.
It's not just a browser with AI bolted on.
The CEO, Aravind Srinivas, thinks the browser is actually the secret weapon for building
AI agents that can do real work.
Like he's talking about agents that can recruit Stanford graduates who worked at Anthropic,
(06:20):
draft personalized emails to all of them, and manage the entire follow-up process.
Okay, that actually sounds terrifying and amazing at the same time, but I'm stuck on
something.
Why does it have to be a browser?
Can't AI agents just, you know, connect to apps directly?
This is where it gets fascinating.
So there's this whole technical battle happening behind the scenes.
(06:44):
You've got people working on something called MCP, Model Context Protocol, which is basically
trying to let AI talk directly to every app and service, but Srinivas thinks that's backwards.
How so?
Think about it this way.
When you want to book a flight, you don't call the Airlines API.
You go to their website, you're already logged in, you can see everything, and if something
(07:08):
goes wrong, you can just take over and fix it.
The browser is this environment we've spent decades perfecting for humans to interact
with digital services.
Oh, so instead of building some complex system where AI has to authenticate with hundreds
of different services, it just uses the browser the same way you and I do, with all our accounts
(07:29):
already logged in and accessible.
Exactly.
And here's the clever part.
When you switch from Chrome to Comet, it imports all your logins, extensions, everything with
one click.
So suddenly, the AI has access to your Gmail, your LinkedIn, your Amazon account, but it's
all happening on your computer, not on Perplexity's servers.
(07:53):
That's actually really smart from a privacy standpoint.
But wait, if I'm understanding this correctly, they've basically built a version of Chrome
with an AI Assistant sitting next to every web page?
That's the sidecar feature.
But here's where it gets wild.
Srinivas shared examples of people using it to unsubscribe from spam emails, create Facebook
(08:14):
ads, even control smart home devices.
One person used it to find the exact timestamp in a YouTube video where someone said something
specific instead of watching the whole thing.
Okay, that YouTube thing is actually brilliant.
But I have to ask, if AI is just extracting the key information from websites, what happens
(08:35):
to all the companies that spend millions making their websites beautiful and engaging?
Are we just turning the entire web into a database?
Srinivas actually addressed this, and his answer surprised me.
He thinks brands will become even more important, not less.
Like if you're the verge, people will still want to know what the verge thinks about something.
(08:56):
But the way they get that information might change completely.
Hmm, I'm not entirely convinced.
If an AI can just extract the key points from any article instantly, why would I ever actually
visit the website?
The business model of the web is built on people actually showing up and seeing ads.
And that's where things get really interesting.
Srinivas made this pointed comment about how Google uses AI to figure out the perfect moment
(09:22):
to show you ads on YouTube to maximize engagement time, while Perplexity is using AI to save
your time by getting you exactly what you need.
The incentives are completely different.
Oh snap, that's a fundamental philosophical difference.
Google makes money when you spend more time on the web, but Perplexity makes money when
(09:42):
they save you time.
That's actually kind of profound.
Right.
And this is where that $2,000 prompt starts to make sense.
If an AI agent can do three hours of recruiting work in five minutes and do it better than
you could yourself, suddenly paying a premium for that kind of cognitive labor doesn't
(10:03):
seem crazy.
I mean, people already pay consultants hundreds of dollars an hour.
But I'm curious about the technical reality here.
Has anyone actually gotten these AI agents to work reliably for complex tasks?
This is where Srinivas gets refreshingly honest.
He admits Comet is still pretty brittle.
The interviewer, Alex Heath, tried to get it to list everyone who follows him on Twitter
(10:27):
that works at Metta, and it only found one person when there should have been many more.
But Srinivas' bet is that the next generation of reasoning models, GPT-5, Claude 4.5, will
get them over the reliability hump.
So we're still in the demo magic phase, but the vision is compelling enough that people
(10:47):
are willing to deal with the bugs?
Exactly.
And there's this interesting chicken and egg problem.
The product needs to be reliable enough that people will use it, but it also needs enough
users to improve through real-world testing.
Right now, Comet is invite only and limited to their 200 a month subscribers.
(11:08):
$200 a month for a browser?
I know, it sounds insane.
But think about it, if this thing can actually do the work of a virtual assistant, suddenly
$200 a month seems reasonable.
That's less than hiring someone for even a few hours a week.
Fair point.
But what about the competition?
I can't imagine Google is just going to let some startup eat Chrome's lunch.
(11:31):
Well, here's the plot twist.
There are rumors that OpenAI is building their own browser too.
And get this.
The rumors started circulating just two hours after Perplexity announced Comet.
Coincidence?
Probably not.
Oh, that's fascinating.
So we might be looking at a whole new browser war, but this time it's not about speed or
(11:52):
features.
It's about which AI can best automate your digital life.
And there's another wild card in play.
The Department of Justice might force Google to spin off Chrome.
If that happens, it could completely reset the browser market and give startups like
Perplexity a real shot at gaining market share.
Wait, that would be huge.
(12:14):
Chrome has something like 65% market share.
If Google suddenly can't bundle it with Android or pay billions to make it the default browser
everywhere.
Exactly.
Suddenly, there's room for innovation again.
But here's what I found most intriguing.
Srinivas thinks the browser is actually bigger than the chatbot race.
He argues that ChatGPT has basically won the conversational AI market, but the browser
(12:39):
is where the real value will be created.
That's a bold claim.
Why does he think that?
Because the browser is where you can build true end-to-end workflows.
A chatbot can help you write an email, but a browser-based AI can draft the email, send
it, track responses, schedule follow-up meetings, and manage the entire relationship over time.
(13:01):
It's the difference between a tool and an assistant.
I'm starting to see the vision, but I keep coming back to this learning curve issue.
When I first got an iPhone, it was intuitive.
I could figure out how to use it without reading a manual.
But with these AI agents, it sounds like you need to completely rethink how you interact
with your computer.
Srinivas acknowledged this.
(13:21):
He said even he has trouble remembering to use the AI features instead of doing things
the old way.
But he thinks it'll follow the same adoption curve as chatGPT, confusing at first, then
suddenly indispensable for early adopters and eventually mainstream.
Speaking of adoption, what about mobile?
Most people do more browsing on their phones than their computers these days.
(13:44):
They're planning mobile apps, but here's the challenge.
On iOS, every browser has to use Apple's WebKit engine.
So Comet on iPhone won't actually be able to use the full Chromium engine they've built
all their AI features on top of.
It's another example of how the platform owners control the innovation.
That's such a good point.
(14:04):
Apple and Google have these invisible choke points that most people never think about.
If Apple decides they don't like AI browsers, they can basically kill them on mobile.
And that brings us to the business strategy.
Perplexity has 30 to 40 million users for their search product, so they're hoping to
convert some of those people to try the browser.
(14:25):
But the real prize would be getting device manufacturers to make Comet the default browser,
which Srinivas admits is nearly impossible because of Google's existing deals.
So they're basically hoping to grow through word of mouth and hope that regulatory action
opens up more opportunities?
Pretty much.
Though there was this interesting tidbitters about Apple potentially being interested.
(14:48):
Eddy Q apparently said nice things about perplexity during the Google trial, and Srinivas
would love to work with Apple on Safari or Siri integration.
That would be a game changer.
Apple's been struggling with AI compared to Google and OpenAI.
A partnership with perplexity could give them a real edge, especially on the search side.
(15:09):
And speaking of partnerships, did you catch the part about Mark Zuckerberg trying to recruit
Srinivas to Metta?
Apparently, Zook is paying nine-figure compensation packages to top AI researchers.
Nine figures?
That's like a hundred million dollars plus.
No wonder there's this talent war happening.
But here's what I don't understand.
(15:31):
If you're perplexity and you're trying to build a sustainable business, how do you compete
with those kinds of offers?
Srinivas made an interesting observation.
He noticed that anthropic researchers aren't getting poached as much as others.
His theory is that money alone isn't enough.
You need mission alignment and the feeling that you're working on something meaningful
(15:53):
that's actually succeeding.
That makes sense.
If you're working on cutting-edge AI research and your company is growing fast and your stock
options are appreciating, why leave for even more money at a place where you might not
have the same impact?
Exactly.
And perplexity seems to be hitting their growth targets.
Srinivas mentioned they want to IPO around 2028 or 2029 after hitting a billion dollars
(16:18):
in revenue and achieving profitability.
A billion in revenue?
That seems aggressive for a company that's essentially giving away most of their product
for free right now.
But this is where the browser strategy could really pay off.
If they can get millions of people using Comet Daily, and if those AI agents actually become
reliable enough for complex tasks, then usage-based pricing starts to make sense.
(16:42):
Maybe you don't pay $200 a month, but you pay $20 for that one recruiting task that
would have taken you three hours.
You know what?
That actually sounds reasonable.
I've definitely spent $20 on less valuable things than saving three hours of work, but
it assumes the reliability gets there.
Which brings us back to the fundamental bet perplexity is making, that the next generation
(17:06):
of AI models will be dramatically better at multi-step reasoning and task completion.
If they're right, we might be looking at a completely different internet experience
in just a few years.
So what should our listeners be watching for?
What are the early signs that this vision is actually happening?
I'd say watch for three things.
First, how quickly the reliability improves.
(17:30):
Can these AI agents actually complete complex tasks without human intervention?
Second, how other companies respond, especially Google and Apple?
And third, whether people actually change their behavior.
Do they start using AI assistance for tasks they currently do manually?
(17:50):
And maybe also watch your own behavior.
Are you finding yourself asking chat GPT or other AI tools to help with things you used
to just Google?
Because that's probably the first step toward the kind of AI-powered browsing perplexity
is betting on.
Great point.
The shift is already happening in small ways.
And if Srinivas is right, those small changes are just the beginning of a much bigger transformation
(18:13):
in how we interact with information and get things done online.
Well, I know I'm definitely curious to try Comet when it becomes more widely available.
Though I have to admit, the idea of an AI watching everything I do online is still a
little creepy, even if it's supposed to be private.
Fair enough.
But remember, these AI assistants are only seeing what you explicitly ask them to help
(18:36):
with.
It's not like they're constantly monitoring everything.
They're more like having a really smart assistant who only pays attention when you tap them
on the shoulder.
That's a reassuring way to think about it.
Alright, so next time you're frustrated by how long it takes to research something online
or book a complicated trip, remember that there might be AI agents just around the corner
(18:59):
that could do all that work for you.
The question is whether you'll be willing to trust them and pay for them.
And whether the company's building them can solve the reliability puzzle before the competition
heats up too much.
This is definitely a space worth watching closely.
Thanks for diving into this with me, Yaakov.
Until next time, keep your pulse on innovation.
(19:20):
See you next week, everyone.
(19:45):
Stay tuned for more updates.
Popular Podcasts
Dateline NBC
Current and classic episodes, featuring compelling true-crime mysteries, powerful documentaries and in-depth investigations. Special Summer Offer: Exclusively on Apple Podcasts, try our Dateline Premium subscription completely free for one month! With Dateline Premium, you get every episode ad-free plus exclusive bonus content.
The Breakfast Club
The World's Most Dangerous Morning Show, The Breakfast Club, With DJ Envy, Jess Hilarious, And Charlamagne Tha God!
Crime Junkie
Does hearing about a true crime case always leave you scouring the internet for the truth behind the story? Dive into your next mystery with Crime Junkie. Every Monday, join your host Ashley Flowers as she unravels all the details of infamous and underreported true crime cases with her best friend Brit Prawat. From cold cases to missing persons and heroes in our community who seek justice, Crime Junkie is your destination for theories and stories you won’t hear anywhere else. Whether you're a seasoned true crime enthusiast or new to the genre, you'll find yourself on the edge of your seat awaiting a new episode every Monday. If you can never get enough true crime... Congratulations, you’ve found your people. Follow to join a community of Crime Junkies! Crime Junkie is presented by audiochuck Media Company.