I was three paragraphs into a security incident notification when Gmail interrupted me with a cheerful suggestion: "Want help writing this? Try: 'I hope this email finds you well. '" I did not want help. I wanted to inform a client that their API key had been rotated after a suspicious login attempt. That unsolicited AI tip wasn't just tone-deaf; it was an interruption in a high-stakes workflow. After a decade of defaulting to Gmail, that was the moment I started looking for an exit.
For years, Gmail was the obvious choice. The web client was fast enough, labels beat folders. And search was genuinely useful. But Google has spent the last two years injecting artificial intelligence into every corner of the product: Smart Compose, Smart Reply, Nudges, Help me write, and now a Gemini side panel that pops up uninvited. Each feature tries to predict what I want to say, when I want to say it. And how often I should be interrupted. As a senior engineer, I recognize these as product decisions that improve for engagement metrics, not user productivity.
The replacement wasn't a high-profile app like Outlook or Superhuman. It was a quiet, native macOS client called Mimestream - an app built specifically for Gmail accounts that uses the official Gmail API, stores data locally. And refuses to offer unsolicited AI suggestions. What started as a simple search for an email client without AI bloat ended up changing how I think about software ownership and user attention. In this article, I'll break down the engineering reasons why this shift matters, the data privacy implications, performance benchmarks. And what developers can learn from the growing backlash against intrusive AI features.
The Uninvited AI Assistant Is a Cognitive Load Problem
Gmail's AI features aren't inherently bad. Smart Compose can save keystrokes when it correctly predicts "Thanks, let me know if you need anything else. " The problem is that the system doesn't ask for permission. The "Help me write" button appears whenever you select text or leave a draft idle for a few seconds. The Gemini side panel expands automatically on certain messages. Each unsolicited suggestion forces your brain to process, evaluate. And dismiss it - a micro-interruption that adds up over a day of email triage.
Research on interruption science consistently shows that context switching costs more than the interruption itself. Gloria Mark's work at UC Irvine found that it takes an average of 23 minutes and 15 seconds to return to a task after an interruption. While an AI pop-up is smaller than a full context switch, the cumulative effect of dozens of daily micro-interruptions is measurable. For engineers managing on-call rotations or incident response, a false AI suggestion during a critical email thread isn't just annoying - it can delay an urgent response.
From a systems perspective, each AI feature adds round-trip latency. "Help me write" sends your draft context to Google's servers, runs inference. And streams back suggestions. That network call can take 300-800 milliseconds, during which your keystrokes may be blocked or delayed. In a local client that doesn't call home for AI, typing feels instant. The difference is subtle but real. And it compounds in long composing sessions.
Email Is Infrastructure, Not a feature Playground
Email is one of the oldest still-running distributed systems on the internet it's governed by
.
Need a Custom App Built?
Let's discuss your project and bring your ideas to life.
Contact Me Today →