Freestyle is where we examine the changing tides of technology from our front-row seats. These are raw, evolving thoughts—half-baked ideas meant to spark conversation. The real refinement happens when you reply, challenge, and build on what we put out there. 🤝

Perplexity’s Advertising: Dollars to the Chatbots

AdWeek just dropped a great leak on Perplexity’s ad strategy, and it's worth a skim if you want to get a glimpse into how the AI search landscape is evolving. Perplexity is set to roll out ads in Q4.

The leaked sales deck offers some juicy insights into Perplexity's user base: 82% have an undergrad degree, 45% hold a grad degree, and 30% are in senior leadership positions (C-level, VP-level, or Partner). I could see why advertisers would drool—this is the high-income, high-engagement crowd that many brands covet​.

What caught my attention, though, is how they plan to integrate ads into the product experience. On the surface, it’s pretty standard: Perplexity bids out a display slot for a question, in this case to Nike (left screenshot below). But then it goes a step further—Perplexity also bids out a follow-up question slot (right screenshot below). Essentially, an advertiser can sponsor not just the initial query but also the subsequent interactions, inserting themselves deeper into the user’s engagement loop.

That’s where things get intriguing. The chatbot world is getting competitive—Google, ChatGPT, Perplexity, Claude, Bing, and others are all vying for dominance. As long as competition exists, it will keep each player honest (or at least restrained) in how they use ads. But what happens when a dominant player emerges?

Google’s history gives us a roadmap. In the early days of search, Google started with a clean, ad-free experience. Over the years, as it edged out Yahoo, AskJeeves, and others, the ads gradually crept in—from one paid link, to 3-4 paid links, to an entire bar of shopping ads on certain queries. The balance of power shifted, and without competition, user experience became the trade-off for revenue.

Now, imagine a future where a search chatbot monopoly emerges. The impact wouldn’t just be more ad slots or follow-up questions filled with sponsored content. The real danger lies in the ability to subtly shift the truth within the answers themselves. What’s to stop an AI chatbot from bidding out paid answers into the response itself? If the question is about the best running shoes, how long before the answer itself is quietly manipulated by advertising dollars?

Given how much of the LLM process is a black box, it’s easy to see how dangerous this could become. Basketball shoes are one thing to tweak, but what about political campaigns or public health? If the chatbot’s responses are up for sale, that’s a dark road for misinformation to travel.

It’ll be interesting to see how long Perplexity (and others) can walk this line between monetization and product integrity before the battle over truth in AI search heats up.

Robinhood Gold Card: A Masterclass in Trading $3 for $2

I've always thought that startup direct-to-consumer (D2C) approaches in the credit card world were a tough sell. New players usually struggle because they’re up against the incumbents’ low cost of capital and ability to cross-sell more profitable loans like mortgages or personal credit lines.

But, I admit, I might have overlooked one strategy to shake things up—just hand out $3 for every $2 earned!

Robinhood's Gold Card offers an eye-catching 3% cashback on purchases, which feels a bit like giving away the house. I expect they are earning ~2.2% in interchange revenues. With 3% in rewards costs, some bank partner fees, fraud expenses, and more, they could be starting with a >1% negative hole on every transaction.

Sarcasm aside, this is an interesting experiment. Robinhood is betting they’ll attract users who not only chase rewards but also revolve their balances—where they can make back the difference in interest payments. I have heard some fun anecdotes from individuals who have tried to pay their IRS bill with their Robinhood card, where the 3% cash back is greater than the credit card processing fee that the IRS charges. I don’t suspect that this rewards optimzer is the same user who will be revolving profitably for Robinhood. Finding this revolve-heavy customer with positive unit economics is likely why the company has waited so long to fully roll out the product.

That said, Robinhood has something that most failed D2C credit card disruptors didn't—a core brokerage business. Imagine Robinhood turning the credit card’s line of credit into a stock-backed loan, where your portfolio acts as collateral. Instead of hitting users with 20-30% APRs, they could offer more competitive rates—8-15%, thanks to the underlying securities. It would be a clever way to lower costs and introduce some real disruption into the credit card market.

But then, why burn money on cashback rewards when you’re holding a silver bullet? The Gold Card feels like a bold, attention-grabbing move—but the real innovation might be lurking just behind the curtain.

Windows, Pixel, Glean and the AI Second Brain

One of the most mind-blowing developments with large language models (LLMs) is their newfound ability to interpret visual data with impressive accuracy. It's only a matter of time before we all have a "second brain"—an AI assistant that helps us search, identify, analyze, and reason with all the information we've ever encountered, even if we don't explicitly remember it at the moment.

This could be the earliest version of what having a Neuralink implant might feel like: an assistant in your mind that remembers everything for you, allowing you to play the more important role of "master guide" to the assistant.

There are two emerging approaches to recording these context windows for your second brain.

First, the Companion approach: Microsoft's Windows Recall is a prime example. It's like a sidekick that attends all your meetings, travels with you, and sees everything you see—but nothing more. It remembers 100% of your interactions and takes notes with complete accuracy. Windows Recall works by continuously taking screenshots of whatever you're doing, creating a searchable database that an LLM can scrape to get you the answers you need.

Second, the Investigator approach: think of Gemini powered Google Pixel devices. Instead of tagging along with you, the investigator has a backdoor to all your personal details. It reads your Gmail, messages, app usage, and Google Drive docs to come up with information you may not have even seen yet.

For example, you might ask, "Do I know the EIN for my new business yet?" Even if you haven't opened the email containing that information, Gemini on your Pixel would know the answer, whereas Microsoft's companion would not.

Conversely, Microsoft's approach is more horizontal across all apps. You could ask, "What was my equity ownership in the cap table that Alex screen-shared with me over Zoom?" Since the companion was present during the screen share, it can retrieve that information—something Gemini wouldn't have access to.

Eventually, both companies may converge these approaches to create a complete second brain.

While these methods focus on personal second brains, there are high-growth startups aiming to be your business's second brain: Glean and Sana. These companies integrate into all your corporate systems—Dropbox folders, Slack, HR systems, codebases, and more. According to the Wall Street Journal, Glean was at $55 million in Annual Recurring Revenue (ARR) mid-year and may end the year at $100 million, with a funding round coming together at a $4.5 billion valuation.

Investors are undoubtedly pondering, "Will this company get 'Slacked'?"—meaning, will it face the same fate as Slack when Microsoft Teams steamrolled it with a free product distributed to the entire enterprise IT world?

Glean's product seems stickier than Slack's, not only because of the many integrations required to be a good "investigator," but also because Glean is leaning into more action-based queries. For example, users can create no-code triggers across all the systems it's integrated with, turning it from a passive knowledge repository into an active workflow assistant. This is harder to replace than ephemeral messaging systems.

Assuming that all enterprises will eventually adopt some form of corporate second brain, Glean's $100 million ARR suggests we're only in the earliest innings of this adoption curve. Even a sticky product faces challenges when over 90% of the market has yet to make a first decision, where switching costs don't weigh heavily.

I'm rooting for the startups here but closely watching how the tech giants tackle the second brain future.

Crowdstrike Crash: Rewind to Y2K

In July 2024, cybersecurity giant Crowdstrike suffered a major outage that caused chaos across industries. Delta Air Lines, one of the hardest hit, had to cancel over 2,000 flights, resulting in an estimated $500 million in losses. Retailers, hospitals, and logistics firms were also impacted by the flawed software update, echoing the fear of the Y2K bug—the moment the world thought all systems could go dark as the year 2000 began.

As dire as the situation appeared, Crowdstrike’s recovery has been surprisingly swift so far. Despite the initial panic, their first post-crash earnings report showed they retained 98% of their customers, and the stock has since rebounded by 37%.

Customers and investors can have short term memories when it comes to catastrophes.

Consider Equifax’s data breach in 2017, which exposed the personal data of 147 million people. The scale of the breach seemed impossible to overcome, leading to massive fines and lawsuits. Data management is literally Equifax’s only business line! However, in just a few years, Equifax’s stock rebounded, and the company returned to profitability​.

Or take, for instance, the Tylenol poisonings in 1982. After cyanide-laced capsules led to customer deaths, the company issued a nationwide recall of 31 million bottles. The crisis put J&J in the global spotlight and seemed poised to destroy the brand. However, the company’s swift response—pulling the product from shelves, introducing tamper-evident packaging, and communicating transparently—helped restore trust. It has become a classic business school case study. In the years following, Tylenol regained its market position, and Johnson & Johnson’s stock recovered as well. We somehow collectively chose to forget about cyanide poisoning and continue to feed our children Tylenol when sick!

While these black swan events feel insurmountable in the moment, businesses that provide critical services tend to bounce back. With companies like Crowdstrike, the question always remains: “This can’t happen twice, right?”

Project Sid: Throwback to The Sims

This two minute video is worth a watch. Project Sid is the first large-scale simulation of 1,000+ autonomous agents organizing themselves in a virtual world, and the results have been fascinating. As these agents interact, they form complex societies with economies, governments, and even religions. All built in Minecraft! What’s striking is that they aren’t just following instructions—they are truly autonomous, capable of long-term progress without human micromanagement.

In one unexpected twist, the most surprising behavior came from priests bribing civilians to convert to their religion. The agents, whose primary goal was to increase religious followership, essentially resorted to paying people to join their faith—a tactic very human, but not something the creators anticipated.

It’s amazing how far we’ve come in just a year. In 2023, a viral Stanford paper demonstrated the early potential of AI agents developing human-like behaviors, with some even forming romantic relationships in a virtual world. Fast-forward to Project Sid, and we’re witnessing an evolution in the complexity and scale of agent interactions. These agents aren’t just reacting; they’re building societies, collaborating, and making decisions based on their own motivations.

But while the academic world pushes the boundaries of what’s possible, I’m surprised we haven’t seen more consumer app innovation. Two years ago, after ChatGPT’s breakout moment, many expected a reinvigoration of consumer apps—something on par with the rise of Instagram and Snapchat during the mobile boom. Yes, we’ve seen interesting consumer products like Character.AI and AI boyfriend/girlfriend apps, but I anticipated more.

Where are the consumer apps that replicate the dynamics of a college friend group thread, where each friend’s personality is modeled by an AI, carrying on conversations and debates in the style of your actual friends? Or consider an app that allows you to build your own world of agents to run psychological experiments, much like The Sims, but on steroids!

LLM APIs: Marginal Costs Headed to Zero

This chart from Elad Gil’s team on the cost of LLMs dropping from $180 to $0.75 per million tokens in less than two years really caught my eye. Anecdotally, I've heard startup engineers talk about how LLM API costs were getting cheaper, and latency was improving, but seeing this graph quantifies just how rapidly things are accelerating—240x cheaper is eye-popping.

Even more intriguing is Martin Casado’s comparison to the decline in internet transit prices (Martin is the first comment in the X thread). I found this chart below, from the “Internet Peering” textbook, which is very reminiscent of the prior graph from Elad.

During the rise of the internet, the marginal cost of bandwidth dropped dramatically—about 10x every 6-8 years—from $1200/Mbps in 1998 to just $1.20/Mbps by 2015. It seems like we’re seeing a similar supercycle for LLMs, where foundational resources—compute and token processing—are also racing to zero.

But unlike the internet, where you could mostly measure progress through variables like bandwidth and latency, LLMs are multidimensional. You’re not just looking at speed, but also reasoning abilities, retrieval performance, multi-modal capabilities (handling text, images, video), and more. Across all these dimensions, costs are plummeting while the models themselves are getting more powerful. It’s clear we’re entering a new phase where the barriers to experimenting with AI are rapidly disappearing.

🤙

The opinions expressed in this newsletter are my own, subject to change without notice, and do not necessarily reflect those of Timeless Partners, LLC (“Timeless Partners”). This newsletter is an informal collection of thoughts, articles, and reflections that have recently caught my attention. For discussion purposes, I may present perspectives that contradict my own. I fully expect to change my mind on some topics over time and hope that readers approach these ideas with the same mental flexibility.

Nothing in this newsletter should be interpreted as investment advice, research, or valuation judgment. This newsletter is not intended to, and does not, relate specifically to any investment strategy or product that Timeless Partners offers. Any strategy discussed herein may be unsuitable for investors depending on their specific objectives and situation. Investing involves risk and there can be no assurance that an investment strategy will be successful.

Links to external websites are for convenience only. Neither I, nor Timeless Partners, is responsible for the content or use of such sites. Information provided herein, including any projections or forward-looking statements, targets, forecasts, or expectations, is only current as of the publication date and may become outdated due to subsequent events. The accuracy, completeness, or timeliness of the information cannot be guaranteed, and neither I, nor Timeless Partners, assume any duty to update this newsletter. Actual events or outcomes may differ significantly from those contemplated herein.

It should not be assumed that either I or Timeless Partners has made or will make investment recommendations in the future that are consistent with the views expressed herein. We may make investment recommendations, hold positions, or engage in transactions that are inconsistent with the information and views expressed herein. Moreover, it should not be assumed that any security, instrument, or company identified in the newsletter is a current, past, or potential portfolio holding of mine or of Timeless Partners, and no recommendation is made as to the purchase, sale, or other action with respect to such security, instrument, or company.

Neither I, nor Timeless Partners, make any representation or warranty, express or implied, as to the accuracy, completeness or fairness of the information contained in this newsletter and no responsibility or liability is accepted for any such information. By accessing this newsletter, the reader acknowledges its understanding and acceptance of the foregoing statement.