I’ve Been in Tech 25 Years. Gmail Finally Broke Me.
I’ve debugged kernel panics at 2am, migrated production databases on a handshake agreement, and once convinced a UNIX box to behave using nothing but desperation and a man page. But Gmail — Gmail — is where I finally hit my limit. Not because it broke. Because it started deciding things for me.
This isn’t a nostalgia rant. I don’t miss the old web. I love good AI. I build with it every day. But there’s a difference between AI that amplifies your intent and AI that quietly replaces it. Gmail crossed that line, and I have the receipts.
The Moment Gmail Stopped Trusting Me
It wasn’t a dramatic failure. It was a nudge.
I’d intentionally left a cold pitch email unanswered. Classic cold outreach, nothing personal, just not relevant to anything I was working on. Gmail nudged me about it. Then again. Then a third time. A small banner at the top of the thread: “You haven’t replied to this. Did you mean to?”
Yes. I meant to. That was the point.
The quiet indignity of it stuck with me. The system had decided that my silence was an accident — an oversight to be corrected — rather than a deliberate choice made by someone who has been managing their own inbox since the late 1990s. Google’s own 2018 blog post announcing the Nudge feature framed it as “helping you stay on top of things.” That framing assumes you can’t. It assumes your inbox is managing you, not the other way around.
That’s paternalistic UX. It’s not a bug. It’s a philosophy.
Paternalistic design overrides or second-guesses user intent rather than amplifying it. Tools built for power users expose controls. Tools built for mass-market audiences hide them, make decisions on your behalf, and treat your preferences as temporary inconveniences to be reset after the next update.
Smart Reply arrived in 2018. Smart Compose followed. Then Nudges. Each one individually tolerable. Collectively, suffocating. By 2023, opening Gmail felt less like using a tool and more like negotiating with a well-meaning intern who kept finishing my sentences.
The Feature Creep Autopsy
Gmail launched in 2004 as something genuinely radical: 1GB of storage when everyone else was offering 4MB, conversation threading, and search-first design. It was built for people who received a lot of email and wanted to handle it intelligently.
Look at what it’s become.
The counts are my own tally from Google’s blog announcements and Gmail changelog entries. In six years, roughly nine significant AI and automation features shipped. Four user-control features made it through. The product’s centre of gravity shifted from your preferences to its assumptions.
2013 brought the tabbed inbox — the first major categorical override of how you sorted mail. 2018 gave us Smart Reply, Smart Compose, Nudges, and Confidential Mode. 2019 brought the embedded Google Meet push. 2021 folded in Chat and Spaces, transforming Gmail into a communication hub whether you consented to that transformation or not. By 2022–2023, Gemini AI summarization, generative reply drafts, and a persistent Workspace AI side panel had arrived.
Each year, the ratio of features I’d asked for to features imposed on me got worse.
The settings archaeology problem made it worse still. To fully disable Smart Compose, Smart Reply, Nudges, and autocorrect you need to navigate four separate settings panels — none grouped logically, none with a master off-switch. And after major updates, at least one of those settings would silently re-enable itself. There’s a longstanding thread on the Google Support Community with multiple reports from 2020 through 2023 documenting exactly this: users disabling Nudges, updates re-enabling them, repeat.
The message that sends is deliberate. Your preference to disable these features is treated as a temporary aberration, not a legitimate configuration choice.
Google’s stated goal is to reduce cognitive load with AI assistance. The actual effect on a deliberate email user is the opposite. Now you’re managing both the email and the AI’s interpretation of the email. Nielsen Norman Group research on automation surprises is instructive here — when automated systems act in unexpected ways, cognitive overhead spikes significantly. Users spend mental energy second-guessing the tool rather than doing the work.
The best analogy I’ve found: it’s like hiring an assistant who keeps finishing your sentences out loud, in meetings, to other people.
Email is still the professional nervous system. The Radicati Group’s 2023 data puts daily global email volume at 347.3 billion messages. Business email dominates professional async communication by a wide margin. When your primary async communication tool starts filtering your intentions through a machine’s assumptions, the cost compounds daily. This isn’t a minor UX quibble. It’s an argument about tool sovereignty.
What I Actually Needed
Before evaluating alternatives,
Twenty-five years of archived email history is a real asset. I use filters and labels as a personal knowledge system. I navigate by keyboard. I run multiple domains. I care about privacy posture I can verify, not just trust.
The actual requirements list looked like this:
- Full control over filtering logic — my rules, not suggested ones
- No AI drafting or replying without explicit invocation
- Search that retrieves, not interprets
- Clean keyboard-driven navigation
- IMAP and standards compliance — not locked to a proprietary interface
- Privacy posture backed by jurisdiction and transparency, not just policy language
That’s not an exotic wishlist. That’s a professional practitioner’s baseline.
The evaluation shortlist was honest about trade-offs.
Fastmail is privacy-forward, headquartered in Australia (outside US surveillance law reach), has a genuinely powerful filter system with Sieve scripting exposed, and supports JMAP. No AI features anywhere in the product roadmap language. Proton Mail offers end-to-end encryption and Swiss jurisdiction — their 2023 transparency report shows active warrant canary status. It’s more locked down than I needed for daily use, but the Bridge tool for desktop clients is solid. Hey by 37signals is the most opinionated rethinking of email I’ve seen — the Screener, Imbox concept, and threading model are genuinely coherent. But the opinionation cuts both ways; it replaces Gmail’s paternalism with its own. Mimestream on Mac is an interesting middle ground — it uses the Gmail backend but surfaces it through a native Mac client that strips away all the AI cruft. Self-hosted Postfix/Dovecot is the nuclear option: maximum control, maximum maintenance, maximum “you broke it, you fix it” energy.
I went with Fastmail as primary, Thunderbird as desktop client.
The Sieve filter scripting is the real reason. My filtering logic is non-trivial — I’ve built a personal tagging and routing system over years, and Sieve handles it cleanly without hiding the logic behind a friendly-but-limited GUI. The import tooling for Gmail history migration is genuinely good. And the absence of AI suggestions everywhere created no void whatsoever. Turns out I was never struggling to compose sentences.
The Migration: What It Actually Takes
Ninety-day parallel run. Both accounts active. No heroic cutover.
The label-to-folder mapping took thought. Gmail’s label system is many-to-one (an email can have multiple labels); Fastmail’s folder structure is hierarchical. I used Thunderbird with both accounts connected during the transition period, which let me drag-and-drop selectively and verify the structure as I went.
Filter reconstruction took a weekend. I exported Gmail filters as XML, used that as a reference document, and rebuilt them in Sieve from scratch. It’s worth doing manually — I found a dozen filters I’d forgotten existed and another dozen that were doing nothing useful.
graph TD
A[Gmail Account Active] --> B[Open Fastmail Account]
B --> C[90-Day Parallel Run]
C --> D[Export Gmail Filters as XML]
D --> E[Rebuild Filters in Sieve Scripting]
C --> F[Map Labels to Folder Hierarchy]
F --> G[Import Email History via Thunderbird]
G --> H[Verify Folder Structure]
E --> I[Test Routing Logic]
H --> J[Update MX Records on All Domains]
I --> J
J --> K[Monitor for 30 Days]
K --> L{Issues Found?}
L -- Yes --> M[Adjust Sieve Rules]
M --> K
L -- No --> N[Disable Gmail Forwarding]
N --> O[Migration Complete]
The MX record cutover for custom domains is the moment of commitment. I did it domain by domain, starting with the lowest-volume one. No drama.
What nobody tells you about migrating away from Gmail: Google Contacts is its own project. Export as vCard, import into Fastmail’s contact system, then audit the duplicates. Budget two hours minimum.
The Privacy Dimension Nobody Talks About Plainly
Google’s business model is attention and data. That’s not a conspiracy theory — it’s in their SEC filings. Gmail has been used to serve targeted advertising. The company says that changed in 2017 when they stopped scanning emails for ad targeting directly. But the integration of Gmail data into broader Google account profiling continues.
Proton’s 2023 transparency report documented 6,378 legal orders received that year; it contested 407 and complied with 5,971 — but because of end-to-end encryption, complying means handing over limited account metadata, never readable message content. That’s Proton. Google receives a substantially larger volume of legal process each year. Their transparency reports confirm thousands of government requests fulfilled annually across products.
I’m not doing anything that warrants legal attention. But data sovereignty isn’t just about hiding things. It’s about not having your professional communication infrastructure owned by a company whose interests are orthogonal to yours.
Australian privacy law — under which Fastmail operates — differs meaningfully from US law in terms of surveillance access. It’s not perfect. Nothing is. But it’s a different threat model.
Six Months Out: What I Actually Notice
The things I expected to miss: Gmail’s search quality, the mobile app polish, the Google Calendar integration.
Gmail’s search is genuinely excellent. Fastmail’s is good. The delta is smaller than I anticipated — for anything older than a few years, I’m searching by sender and date range anyway, and both handle that cleanly.
The mobile experience took adjustment. Fastmail’s iOS app is functional but not as polished as Gmail’s. I use Mimestream on Mac as my primary interface and the Fastmail app on mobile. This is fine.
The Google Calendar integration I replaced with Fastmail’s calendar for personal items and Fantastical for the unified view. Not a regression.
What I didn’t expect: how much mental space returned when the AI stopped generating text near my name. No suggested replies to dismiss. No Nudges to override. No Gemini summaries of emails I was about to read anyway. Just email.
The radar chart tells the story honestly. Gmail wins on mobile polish and migration ease (staying is always easier than leaving). Fastmail wins everywhere a deliberate practitioner actually cares about.
Who Should Actually Make This Switch
Not everyone. I want to be honest about that.
If you use Gmail casually, get value from Smart Compose, and don’t have complex filtering needs — stay. The product is genuinely good for median-user workflows. Google has built something impressive for that audience.
This switch makes sense if you’re a power user who’s noticed the settings archaeology problem. If you’ve ever been Nudged about an email you were deliberately ignoring. If you’ve turned off Smart Reply four times and found it re-enabled. If the idea of Gemini summarizing your professional correspondence before you’ve read it makes you instinctively uncomfortable.
It makes sense if you build things for a living and you’ve started noticing that your tools are developing opinions about how you should work.
It also makes sense if you’re the kind of person who, after 25 years in tech, still believes that a tool should do what you tell it and stay out of the way when you don’t.
The Bottom Line
Gmail didn’t break. It evolved — toward an audience that isn’t you.
The product is optimizing for the median user: someone overwhelmed, reactive, and grateful for an AI to reduce the burden of communication. If you’re reading a 25-year practitioner’s tech blog, that’s definitively not you.
The lesson I took from six months of migration and reflection isn’t that Gmail is bad. It’s that tool sovereignty is a professional discipline. The email client, the code editor, the note-taking app — every tool you use shapes how you think. When those tools develop paternalistic opinions about how you should work, the compound cost is real, even when each individual nudge seems trivial.
The best tools disappear into the work. The worst ones keep reminding you they’re there.