[Caveat: personal experience only, nothing more]
There was a moment (not that long ago, really) when it felt like we might actually escape the gravity of Big Tech's communication platforms.
Around 2015-2022, a whole ecosystem of alternatives emerged. DuckDuckGo was genuinely growing, not just surviving. Fastmail had loyal devotees. Hey came along with its opinionated approach to email. ProtonMail offered end-to-end encryption. Smaller calendar apps, notes apps, document tools, all carving out sustainable niches.
They were real businesses with real growth. Some driven by privacy concerns, others by better UX.manageable. The tradeoffs were acceptable. You gave up some convenience and integration, sure, but you got something back: privacy, a cleaner interface, a sense of control over your digital life.
That moment seems over.
The AI Integration Trap
Here's what changed: AI assistants actually got good enough to be useful. And useful AI assistants need to read your email, check your calendar, scan your documents, and understand your context.
Look at what happened between ChatGPT's launch in November 2022 and mid-2024. Nearly every major AI productivity tool now requires Google Workspace or Microsoft 365 integration. Superhuman? Gmail and Outlook only ... no Fastmail support despite years of requests. Motion, Reclaim.ai, and other AI calendar tools? Same story.
Want to use these tools with your privacy-focused email setup? You simply can't. It's just not an option! It's very common now to see "Connect to Gmail", or "Connect to Google Drive" or "Connect to Google Calendar".
This isn't an accident: If you're building an AI email assistant, you need all of
- Deep API access to email content
- Calendar integration that's actually reliable
- Contact management that doesn't break
- Document access that works across services
- And you need all of this to work together
Google and Microsoft offer this in a single integration. Google Workspace has 3 billion users on a unified platform. Building comparable integration across smaller alternatives means 10x the engineering work for a tiny fraction of the market.
So they don't.
And suddenly, your carefully constructed privacy-focused setup isn't just a personal choice anymore, it's a professional liability. It's the reason you can't use that AI assistant everyone's raving about ... the reason you're doing manual work while your colleagues automate.
This dynamic benefits not only the dozens of AI startups building on these platforms, but most of all, the platform owners themselves. Gemini in Google Workspace and Copilot in Microsoft 365 have privileged access no third-party tool can match (they're “built into the substrate”, seeing everything by default).
Privacy Takes a Back Seat (Again)
Remember when we cared about privacy? :-)
The privacy moment (roughly 2013-2021) was brief—between scandals becoming visible and AI becoming irresistible. We knew we were the product.
And then we collectively decided that having an AI read all our emails was fine, actually, as long as it could draft responses and summarize threads.
I'm not even being sarcastic. The tradeoff actually shifted. When Gmail was just a place to store email, giving Google access to all of it felt creepy and unnecessary. When Gmail becomes the substrate on which your AI assistant runs, well... that's the price of admission.
Privacy isn't dead. But it's no longer the deciding factor for most people. Capability won.
Integration Is the New Moat
Here's what Google and Microsoft figured out, maybe by accident: in the AI era, the moat isn't just having good products. It's having products that work together in ways that AI can exploit.
The old switching cost was your data—years of emails, contacts, calendar events. Painful but doable.
The new switching cost is your AI's effectiveness. It's not just your data anymore—it's your AI's training ground, context window, ability to help. Switch providers and your AI assistant gets dumber. It can't see your email, doesn't know your schedule, can't coordinate across services.
Those with existing data integration? Winning. Those with millions of users already on their platform? Winning. Those with the engineering resources to offer unified APIs? Winning. (and those who control the platform and can offer native AI assistants with privileged access are naturally winning the most)
This is a fundamentally different kind of lock-in
The Paradox of Decentralization in the AI Age
The alternative tools dream was decentralization—pick the best email, calendar, document editor. Mix and match. Keep data separate.
But AI works best with concentration. Large language models benefit from:
- Massive datasets in consistent formats
- Deep integration across services
- Unified APIs and authentication
- Scale that justifies infrastructure investment
What makes privacy-focused setups appealing to users makes them unappealing to AI developers. Every additional integration is exponential complexity.
Centralization is happening through genuine product advantages. AI works better when it can see everything. Assistants are more helpful with unified access.
But what about on-device AI?
Apple's on-device AI offers a different model: privacy is architectural, not policy-based. For personal use, this works (your phone can summarize messages, clean up photos, suggest responses).
But here's the catch: the most powerful professional and collaborative AI tools still require cloud integration. They require server-side processing, persistent context, and deep integration with cloud services.
So What Now?
I don't have a neat answer here. Maybe there isn't one.
The obvious take is cynicism: we're doomed to repeat the cycle, and Big Tech always wins. Privacy was a brief aberration, and we're back to surveillance capitalism, now with AI assistants.
But maybe it's more nuanced. Maybe we get:
- A mainstream that uses integrated AI on Google Workspace or Microsoft 365 because the capabilities justify the tradeoff
- A committed niche that pays the convenience tax for privacy and control
- On-device AI that handles personal tasks while cloud AI dominates professional collaboration
- Periodic regulation that forces some degree of interoperability
- New alternatives that emerge when AI reaches its next plateau
Or ... maybe ... we're watching the end of the era where you could meaningfully opt out of big tech platforms while remaining professionally competitive. Maybe AI integration requirements make that stance unsustainable for anyone who wants to participate fully in modern work.
The thing that strikes me most: we're making this choice really fast. The privacy-focused tools era lasted maybe 7-10 years. The AI reversal happened in about 18 months.
P.S. Here's an in-the-moment example
I’m editing this piece right now with Claude in the browser. Had to switch to Chrome to get the best experience for it, and ditched Safari, Zen, all the alternatives I’d been using. Got massive value out of the AI integration. But I had to go back to Chrome to use it. Another clear-cut case of the empire striking back.