We’re entering the context era of AI.
Not just smarter models—but AI that knows what tabs you have open, what docs you wrote last year, and what stories you’ve forgotten. From AI-native browsers like Dia and Comet, to birthday letters stitched from forgotten Google Docs, the next leap isn’t new capabilities—it’s deeper context.
📺 AI Browsers are Coming
The AI Browsers are coming and I think this will lead to a huge shift in how regular people interact with AI in the next 6-12 months.
Over the last few weeks there’s been several big releases and leaks:
The Browser Company, maker of the Arc browser, released Dia, a separate product built with AI from the ground up. Currently in beta and available to Arc Members.
Perplexity launched Comet, a new browser deeply integrated with Perplexity AI. Available only to Perplexity Max customers ($200/mo)
ChatGPT is rumored to be weeks away from releasing its own browser.
Gemini in Chrome was previewed at I/O.
Personally I haven’t used any of these browsers, but I’ve seen a few YouTube videos that showcase what makes them feel different from just bolting on a chat copilot.
Riley Brown shows off how Dia’s built-in AI can leverage your open tabs and bookmarks as context to enable real workflows.
You can create “skills” that are triggered by slash commands—essentially reusable prompts. He demos one that summarizes YouTube videos.
You can also reference bookmarks with @mentions, storing more complex prompts or reusable context in Google docs or Notion. That could be super useful for writing emails, notes, business comms, and even this newsletter.
Perplexity’s Comet has its own angle: it can browse the web on your behalf, with your credentials. That opens up use cases no public agent can reach.
Matthew Berman demos Comet by asking it to hunt down a Nintendo Switch 2 in stock. At first, it returns a generic list of retailers. When prompted to check stock itself, it opens each site, checks availability, and reports back.
ChatGPT’s rumored browser could be even more impactful. They already have the best general-purpose models, massive consumer adoption, and good UX.
The browser is the battleground. It’s where we spend most of our day. It’s where do we do most of our work. And it’s barely evolved in 10+ years.
If AI-first browsers can integrate chat and agents directly into core navigation and workflows, they could become the primary way people experience AI, leading to serious lock-in.
🌐 The AI Birthday Letter That Blew Me Away
This piece from the Atlantic about Gemini writing a shockingly personal birthday letter really resonated with me.
A few months back, I asked ChatGPT to write a biography of my life. It didn’t have access to my Google Drive or Gmail, but it still surprised me how well it inferred my life from web search data alone.
I also asked it to take some creative liberties and fill gaps in my narrative with plausible events and experiences based on place and time. It invented childhood details, like riding bikes in a suburban cul-de-sac that were factually incorrect, but directionally accurate.
The author used Gemini, which did have access to their personal data, and the result was a letter that felt deeply specific—like something only a close friend could write.
Inspired, I tried again across Gemini, Claude, and ChatGPT with Google services enabled.
Gemini just didn't work. I couldn't get it to connect to my Google services.
But Claude and ChatGPT surfaced docs I totally forgot about, like a Q&A I completed for my wedding officient detailing how I met my wife to be. Old calendar entries. Forgotten email threads.
They each got some facts wrong—relationship statuses, birthplaces, dates—but the scaffolding was accurate.
It's a tangible example of where AI is going. More personal.
Agents are more valuable when they have access to meaningful, personal context. That could mean surfacing documents from your past, details from your preferences, or summaries of who you are.
It’s not hard to imagine how this could reshape tools we regularly use, productivity apps, and onboarding flows into new services.
There's a whole host of privacy issues we'll have to contend with, but as someone who relishes in nostalgia, I'm excited to use these tools to resurrect forgotten memories.
📺 Aaron Levie: White Collar Jobs, Future of SaaS, Agents, Concentration of Power
This conversation with Mark Bergman really hits the current tension around AI and white collar work.
Mark kicks things off by referencing two recent AI hot takes: one from Dario Amodei at Anthropic warning that AI could eliminate 50% of entry-level white collar jobs, and another from Amazon's CEO Andy Jassy suggesting similar workforce reductions.
Aaron doesn’t dismiss the tech or its capabilities. In fact, he’s bullish on AI agents and automation. But he’s skeptical of doomsday timelines.
His key point: even if agents can do more and more, they still trip up on basic things. They struggle with edge cases, lack contextual judgment, and need human supervision to get things over the line.
Or as he puts it: they’re not great at the last mile.
That means people are still needed—for now and likely for a while—to manage, review, verify, and redirect agent output. We might need fewer people, but we’ll still need people.
He also highlights something I don’t hear enough: not all work is just a sequence of tasks. There’s ambiguity, collaboration, and human nuance that agents aren’t close to replicating.
A human PM doesn’t just assign tickets. They rally stakeholders, mediate conflicting opinions, and sometimes bet on a risky idea no spreadsheet would approve.
Most people don’t want to spend their day managing a dashboard of bots. We like working with other humans. That won’t disappear overnight.
📺 Behind the Craft: Is This the First AI Analyst That Actually Works? | James Evans (Amplitude)
James Evans, Head of AI at Amplitude, demos their new analytics assistant—Amplitude AI.
Instead of fixating on superintelligence or agentic magic, he frames the product in terms of time. Not "what can AI do that humans can't," but "what can it do so humans don’t have to."
In his words: they’re not just shipping software. They’re shipping "people" in the form of agents.
Most folks don’t want to spend their days knee-deep in Amplitude queries. But that doesn’t mean there’s not demand for that kind of work. This is where AI agents can step in—doing the work that's otherwise neglected, or giving back time to do something higher value.
As someone at Pendo, I'm watching this product closely.