The Last Mile of Office Work Disappeared
The one thing you need to know in AI today | AI-Ready CMO
We remember our early days at Weber Shandwick, a global PR firm where we both began our careers. We were juniors. We had to fill out timesheets. The more junior you were, the more hours went into “admin,” which was a collective term for work that was not billable to clients. The more senior you were, the less admin you had in your timesheet.
Admin could be anything from collecting data to building spreadsheets to massaging PowerPoint decks. The “ground work” we call it now. The last mile for all the work we did.
This last mile has just disappeared now.
Three announcements landed within days of each other. Microsoft launched Copilot Cowork, an AI agent that reads, analyzes, and acts on files directly on your computer. Anthropic updated Claude to maintain shared context across Excel and PowerPoint simultaneously — pull data from a spreadsheet, generate the finished presentation, in a single session. OpenAI gave ChatGPT write access to Google Docs, Sheets, Outlook, and your calendar.
Each of these updates targets the same thing. Not the data. The output. The formatting, packaging, slide-building, and document writing that every marketer does, regardless of their title, specialization, or seniority. The last mile of work. The admin. The ground. The boring.
Quick sidenote:
It is worth noting that all three of these announcements came from different companies in the same week. Microsoft, OpenAI, and Anthropic are not coordinating. What this reflects is that they’ve all arrived at the same conclusion simultaneously — whoever owns the workflow layer owns the enterprise relationship.
What’s the damage?
There are many studies on this type of work. From our own experience, we can safely say that around 20-30% of the entire office workload (marketing or not) was “admin.” Which, by the way, rhymes with the projected %s of layoffs due to AI.
1/3 of the workforce will not be necessary because the work they do has simply disappeared. Not lost its meaning. Not lost its revenue potential. Not outsourced or done by others. Simply just not there anymore.
Some managers will tell you that doing admin builds character. That you need to earn your way through the spreadsheets before you’re allowed to think strategically.
We’ve never bought that.
You don’t learn architecture by hand-drawing floor plans when CAD exists. You don’t learn finance by refusing calculators. The tool removes the friction. What’s left is the actual work — the judgment, the analysis, the decision. That’s what we should be training people for. Always was.
What can we do?
Pre-AI, this admin work had two layers.
First, the data mining. Pulling numbers, cleaning them, and getting them into a format (sheet). Also known as “reporting.”
The second was output building. Taking the reported numbers, building up findings, and putting them into a format (deck). Also known as “insights.”
The first layer was already massively affected by AI. Not sure if you have seen, like, anyone, crunching numbers on a spreadsheet manually in 2025. It’s a rare breed. Dying breed.
The second layer disappeared just about now. We knew this would happen because we had early signs in startups like Gamma, and others. But now, it is natively built into most LLMs you use anyway. It’s seamless. Automated. Easy.
So what becomes the new admin? Honest answer: we don’t fully know yet. But the early shape of it is visible.
It’s the work of directing the machine — writing the prompts precisely enough to get usable output, reviewing AI-generated work for accuracy and judgment, maintaining the data sources the machine pulls from, and owning the quality bar when no human built the thing from scratch.
In other words, the new admin is in oversight. And oversight, done well, is not low-skill work. It requires knowing what good looks like, catching what the machine gets wrong, and making the judgment calls it can’t make. The difference is that one person can now oversee outputs that used to require ten people to produce.
That changes headcount math. It also changes what you hire for.
— Torsten and Peter
Share with a colleague. Earn paid membership months.
Everything we know about AI marketing. Free.
Courses, workshops, frameworks, tool reviews, and implementation guides — all free, all on-demand. Built for senior marketers who don’t have time for theory.
3 AI Marketing Tools To Try Today
Gumloop
Gumloop is a no-code AI automation platform that lets you visually chain together AI models, web scrapers, and data tools to build complex workflows. Use it to automate repetitive processes like lead enrichment or content pipelines without writing a single line of code.
Instantly
Instantly scales cold email outreach with unlimited sending accounts, built-in AI warmup, and deliverability optimization. Use it when you’re sending 1,000+ emails per week and can’t afford to land in spam.
Reclaim AI
Reclaim AI automatically schedules focus time, optimizes meetings, and protects work-life balance across Google Calendar and Outlook. Use it to reclaim 7+ hours per week by letting AI manage your calendar around actual priorities.