If you've been reading the headlines, you'd be forgiven for thinking we're all about to be replaced by a chatbot.

We've spent the last decade running Through The Glass Creatives — an internationally-awarded design and growth agency working with brands from Jacob & Co. to Nuvia Dental Implant Center. We employ designers, writers, strategists, and developers. We've also been through our own AI transition, which forced difficult decisions about who stays and who doesn't on a creative team. So we have an opinion on this question that's grounded in payroll, not punditry.

Here's the honest answer: most jobs won't disappear, but most jobs will change. The question of "what's safe" is the wrong frame. The right frame is what shifts make you valuable in the new structure.

What the actual data says

The most-cited research on this question is the World Economic Forum's Future of Jobs Report 2023 (published May 2023), which surveyed 803 companies employing 11.3 million workers globally. The headline finding: 23% of jobs are expected to change by 2027 — 69 million new roles created, 83 million eliminated, for a net loss of around 14 million jobs (roughly 2% of current employment). Not catastrophic. Not nothing.

McKinsey Global Institute's "Generative AI and the future of work in America" (July 2023) projected that up to 30% of work hours in the U.S. economy could be automated by 2030 under accelerated adoption — but emphasized that "automated" rarely means "the job is gone." Most often it means tasks within the job are automated, and the worker's role shifts to higher-value work.

Goldman Sachs (March 2023) estimated 300 million jobs globally could be affected by generative AI, but two-thirds of current jobs are "exposed to some degree of AI automation," not eliminated outright.

Pew Research's "Which U.S. Workers Are More Exposed to AI on Their Jobs?" (July 2023) made a more interesting finding: the jobs most exposed to AI aren't low-wage roles — they're high-wage white-collar jobs in information processing, finance, and management. That inverts the historical story of automation.

Jobs the research suggests are MOST exposed

Across the WEF, McKinsey, Pew, and Stanford AI Index 2024 datasets, the consistently exposed categories include:

Jobs the research suggests are LEAST exposed

The same datasets converge on a few categories where human work remains hard to displace:

What we've actually seen at TTGC

When we started rolling out AI-assisted workflows in late 2024, we expected resistance. We didn't expect how cleanly the team would split.

The designers and writers who treated AI as a tool — something to draft, iterate, and pressure-test against — became dramatically more productive. They shipped more, the quality went up, and their role evolved into something closer to senior creative direction. The team members who refused to engage with AI as a tool — who insisted on doing every iteration by hand — got slower relative to the rest of the team, and eventually their cost structure no longer made sense for the work they were producing.

We had to make difficult layoffs. Not because the people were bad at their jobs, but because their jobs (as they defined them) were structurally less competitive than the same role re-imagined with AI in the loop.

The question isn't "will AI take my job?" The question is "am I willing to redefine my job around AI?"

What "safe" actually means in 2025

Based on the research and our own experience, a job is meaningfully safer if it has three properties:

1. It requires judgment under ambiguity

Generative models are pattern-matchers. They struggle when the inputs are incomplete or the goal is genuinely novel. WEF 2023 calls this "analytical thinking" — the #1 skill employers cite as growing in importance.

2. It involves direct relationship management

Trust, accountability, and long-term client/patient relationships aren't transferable to AI. Most of our retainer clients work with TTGC because they know us — that intimacy isn't something a tool replicates.

3. It has a physical or embodied component

Hands matter. AI doesn't have any. Trades, healthcare delivery, and on-site work remain anchored to bodies in physical space.

Most jobs have some of these properties. Almost no job has none. Your job's safety is best measured by how much of your daily work involves judgment, relationships, or embodiment — and how much is repeatable pattern-matching that a model can do.

The honest take

Nobody can promise you your job is safe. We can't promise that to ourselves. What we can say, based on the data and our own ops, is: the people who lose their jobs to AI in the next five years won't lose them because AI is more creative or smarter than they are. They'll lose them because they refused to integrate the new tool.

The good news: integrating is learnable. The better news: the people who do are getting paid more, not less.

Sources