#86: Hot Takes in AI (Part VII)

TBPN, Project Glasswing, and Perplexity Computer

Another month, another Hot Takes in AI segment. Given the pace at which technology is developing, I could write a new piece every day. But for my own sanity, I’m not going to do that (once per week is a solid balance). Oh, and I said last edition that I wasn’t going to make any more Perplexity predictions. That promise lasted one week. I have a new take coming up.

Let’s get going.

Hot Take #1: OpenAI’s acquisition of TBPN was about buying engagement, not reach.

For those of you not laser-focused on tech news, OpenAI bought a podcast for a price tag rumored to be in the “low hundreds of millions”. Technology Business Programming Network (TBPN) is a daily, three-hour live show where two hosts discuss the latest happenings in technology and interview executives and other interesting figures from across the industry.

In my opinion, it’s an entertaining listen and their roster of guests rivals that of another top tech podcast, All In. But check out this: TBPN has only 85,000 YouTube subscribers and their live stream averages just 4,000 to 10,000 views daily.

Those statistics don’t scream “major media exit.” A few things to unpack.

Despite having relatively low subscriber numbers (for what it’s worth, I would give an arm and a leg to have these figures for Relentlessly Curious), where TBPN shines is in their clips performance. Today, live streaming is the “hot” form of media, and the red-hot form of media is “clips”. Think of small snippets of conversations or scenes from a longer-format video (i.e. live streaming). Their clips regularly reach hundreds of thousands of views, sometimes breaching millions of views on X.com. They also have advertising sponsors for their clips, which makes the claim that TBPN has a $30 million advertising run-rate business more realistic. So, from a pure media economics perspective, the suspected valuation is quite high, but not truly out of this world.

Taking a step back, it may seem that buying a podcast is contrary to OpenAI’s current directive on increasing their share of the enterprise and coding market. They recently shut down Sora and Instant Checkout in a heavily publicized move to reel in any side bets that were diluting company focus.

Buying TBPN wasn’t about expanding OpenAI’s reach into mass consumer adoption of AI. It was the opposite. They want to win back the power users of AI, those who work in tech or in tech-adjacent careers. TBPN consistently attracts the biggest names in tech and has incredible engagement and view counts on their X.com clips, the epicenter of tech discourse.

Despite claiming TBPN will maintain editorial independence, the brand association with TBPN may warm up AI power users to spend less time in Claude Code and more time in Codex (OpenAI’s agentic coding software). Also, the acquisition is an acquihire of people who truly understand the cultural zeitgeist of technology in 2026. If the TBPN crew can help OpenAI craft the right marketing messaging to capture enterprise clients from Anthropic or Google, the acquisition pays for itself and then some.

One last thing. TBPN has a vibey, but approachable brand. If OpenAI can leverage TBPN to repair their brand image post-Pentagon deal, they may win back retail investors ahead of a likely in the back half of 2026. Given their IPO likely could surpass $1 trillion, a few hundred million is a drop in the bucket for the sake of improving “vibes”.

Hot Take #2: Anthropic’s doomsday messaging around their new model Mythos is actually a good thing for society.

If you’re deep in Claude Code (like me), you understand just how great Anthropic’s top model Claude Opus 4.6 is. What I find most impressive about Opus is its ability to act as a strategic partner when designing a workflow or process. Agentic coding capabilities aside, its ability to synthesize the context I provide and quickly identify edge cases and holes in my logic helps build stronger, cleaner process automation.

Apparently, Anthropic’s new model, Mythos, has coding skills so capable that it has identified cybersecurity vulnerabilities across the web.

As a result, Anthropic quickly assembled a coalition of major enterprises named Project Glasswing because Mythos, “has already found thousands of high-severity vulnerabilities, including some in every major operating system and web browser”. Companies like JPMorgan, CrowdStrike, and Amazon are expected to work closely with Anthropic over the next 90 days to understand and fix any of the vulnerabilities surfaced by Mythos.

What’s notable is that many of these cybersecurity risks have been hiding in plain sight. As AI’s coding capabilities keep improving, it will continue to identify risks we weren’t able to comprehend with existing technology, which is particularly scary if this technology gets into the wrong hands such as a foreign adversary or a malicious domestic actor.

Riddle me this. Anthropic has built a differentiated brand in the AI space around “safety”. In fact, their legal entity is listed as a public benefit corporation, which effectively means they can prioritize stakeholders beyond shareholders.

With a brand focused on “safety”, Anthropic chooses to virtue signal frequently about the risks associated with their AI models and the steps they are taking to manage them. They release what is called a System Card with each model launch, that dives into model performance against benchmarks and potential failure modes (as well as how bad actors can exploit their technology).

Their fearmongering is a form of marketing. They focus on the downside of their models as their competitors often don’t highlight them. It makes Anthropic seem more responsible and trustworthy, particularly critical for enterprises considering using Claude models. Furthermore, it allows them to further embed themselves in enterprise and government infrastructure, which bodes well for long-term adoption of Claude. No wonder why Anthropic is crushing the B2B market.

But in the instance of Mythos, I believe caution is necessary and the 90 day pause is justified. If Anthropic is willing to risk their lead in the AI race for the sake of getting major enterprise players prepared for the risks associated with Mythos, this is a moment worth paying attention to.

Hot Take #3: Perplexity’s move towards agentic solutions represents the only thing AI companies should be building for: outcomes.

If you’re building an AI company, you should pay attention to this. Perplexity’s annual recurring revenue (ARR) jumped 50% from February to March, on the back of their Perplexity Computer release. Their monthly active user counts topped 100 million as reported by the Financial Times, and a new pricing model, including both subscription and usage charges, has catapulted Perplexity out of the characterization of being a glorified search engine.

Perplexity Computer, a series of AI agents that leverage 19 different models to tackle the project you give it, has provided a new runway for Perplexity to build on. It also signals to the market what users value: agentic solutions.

Anthropic has Claude Code, OpenAI has Codex, and now Perplexity has Perplexity Computer. If you can knock out a complex task for a person or a company, they are willing to pay for it. Better yet, if the software you offer allows users to write in plain English what they want, and your software creates a tech product without them writing a single line of code, you’re golden.

People are willing to pay for outcomes. If your AI can’t automate a customer workflow, freeing them up to focus on higher-value tasks or scaling their business without additional headcount, good luck. Software creates insights, AI creates outcomes