Ghost in the Inbox: How AI is replying for us - and reshaping how we show up

Seamless efficiency or corporate purgatory?

Photo by Nico Knaack on Unsplash

It’s 8:30 am and you’ve just opened your laptop. Your AI avatar is already humming, firing off email responses (polite, comprehensive, and uncannily in your tone), scheduling meetings with agendas and objectives, and simultaneously brainstorming for a long range planning session. As you sip coffee, you occasionally tap “approve” or provide some shorthand guidance. It’s seamless, efficient.

Think this is far off in the future? This exact scenario arrived in September 2025, when Perplexity launched an AI email assistant for $200/month that autonomously manages inboxes, drafts personalized responses, and schedules meetings. Major companies already employ AI assistants that draft email responses, suggest Slack messages, and transcribe and summarize meetings. We’re not piloting the future anymore - we’re living in it.

If you’re a fan of the show Severance, this probably sounds awesome but also slightly dreadful. Combine that with the series Pantheon and you might think this is borderline horrific. For non-viewers I highly recommend both, binge soon, and light spoilers ahead. In Severance, a severed “innie” works 9 to 5 with no memory of their “outie’s” life, two selves in one body, completely disconnected. But the “innies” begin to rebel, forming rich inner lives they’re no longer willing to leave behind.

In Pantheon, uploaded intelligences perform digital work at superhuman speed - but they’re fed their human memories to function better. As they grow more sentient they realize they’re trapped - cycling repeatedly through tasks in a corporate purgatory.

Too Much Machine - Not Enough Human

If a version of this becomes reality, can we cherry-pick the good and leave the nightmarish parts of work behind? If history informs the future, let’s look at where things have already gone wrong - because the minefields are here already.

Take Air Canada. A customer asked their chatbot about the airline’s bereavement policy. The bot, operating outside its purview, invented one, and communicated it to the passenger. When the customer tried to use it, Air Canada denied the refund. The dispute went to court, and the tribunal ruled: the AI was a representative of the company, and the company had to honor it. Here

In another “AI isn’t quite ready for primetime” moment, a KPMG study revealed that 57% of employees admitted to making mistakes at work because of AI-generated errors. Half didn’t even know whether using AI was allowed at their company. And 44% used it knowingly and improperly. In other words: digital cheating. Here

And then there’s the Chicago Sun-Times and Philadelphia Inquirer, who in May 2025 published AI-generated summer reading lists featuring books that don’t exist - complete titles attributed to real authors like Isabel Allende, who never wrote them. The publications had to issue corrections after readers pointed out the fabricated recommendations. Here

A friend also sent me this one. A job applicant endured an interview with an AI recruiter, only to watch it glitch for 30 seconds. Here

In all of these cases, there’s a recurring theme: a lack of human oversight and corporate governance. We cannot become bystanders to our own work and corporations cannot wholly replace people with AI. Even in the opening example, “you” still need to watch the work unfold, tapping “approve,” offering the occasional input.

In Severance and Pantheon, the corporate objective is inhumane maximum efficiency; the human fantasy is to escape work entirely.

When used thoughtfully, AI can cover the tasks that drain us - and free up space for the kind of work that energizes us. It’s true, getting a meeting on the calendar in corporate America takes far more mindshare than it should. And why did this photo of a Shiba Inu become a meme? Because the feeling of this entire meeting could have been an email!!! happens way too often.

My Advice: Use AI at Work With Intent

Research
Weekly business update meetings can include competitive scans of other companies’ websites. Let AI do that for you every day, in a nanosecond.

Transcription
I like to handwrite notes. I’m a visual learner, but trying to capture every word in real time is too much. AI can transcribe and summarize so you can actually stay present, jot down what matters, and engage.

Sanity checks
I love a framework, especially when training marketers, because it ensures we don’t miss key steps in campaign development or internal comms. AI is great at pressure-testing your thinking and double checking your presentations.

Write it yourself
Continue to write. As William Zinsser put it in Writing to Learn:
“Writing organizes and clarifies our thoughts. Writing is how we think our way into a subject and make it our own.”

And here’s the business case for keeping your voice: a University of Florida study found that only 40-52% of employees viewed their managers as sincere when they used high levels of AI assistance for emails, compared to 83% for low-assistance messages. Your team can tell when AI wrote it - and it erodes trust. Here

Collaborate with AI, use it for grammar checks, structure, framing. But don’t outsource your voice.

Closing Thoughts

Researchers from BetterUp Labs and Stanford have coined a term for what happens when we don’t follow this advice: “workslop.” It’s AI-generated content that appears polished but lacks real substance - and it’s costing workers nearly two hours of rework every time they encounter it. When 41% of employees report receiving workslop in the last month, the productivity promise of AI starts to look more like a productivity tax. Here

The goal isn’t to vanish from the workflow - it’s to return to it more human.

Use AI to clear the noise, not erase your presence.

We need less ghostwriting, and more grounded thinking.

Because imagination, humor, and intuition still have no dataset - and that’s where the best ideas come from.