AI Possibility + Peril: A Two-Part Series on the Future of Knowledge Work
Note: This article is part of a two-part series on the future of knowledge work.
• Part 1 — Possibility: My One-Year-Old Just Became a Principal Consultant: Meet Auggie — how the new model of Humans + Virtual Employees + Agents raises the floor and opens new leverage.
• Part 2 — Peril: Elevate the Best. Automate the Rest. What Now? — what business leaders and LLM model-makers are really saying about jobs, workflows, and teams — and why leaders should act today.
We recommend reading both together: one frames the upside of leverage, the other the risks and changes leadership teams should be acting on now — and that employees are already thinking about, often with concern.
September 30, 2025
Part 2: Elevate the Best. Automate the Rest. Now What?
In my companion piece, My One-Year-Old Just Became a Principal Consultant: Meet Auggie, I wrote that we’re entering a new operating model for knowledge work.
That piece leaned into Possibility. Today is about Peril—and a plan.
The floor rises. The ceiling rises faster.
A WSJ piece on “work–life balance” kicked off a three-generation family text thread—not about balance, but about AI at work. In the rapid-fire back-and-forth, I typed, “We’re quickly moving to a world of elevate the best, automate the rest.” I didn’t love that I said it. But it still feels probable.
In pilot after pilot, average work got average—fast. Cleaner copy, tidier slides, fewer errors. That’s the Virtual Employee + Agent combo raising the floor.
At the same time, the people who could orchestrate—who knew where to point an Agent, when to lean on a Virtual Employee’s memory, and when a Human must decide—pulled ahead. They didn’t grind harder; they worked differently.
Apprenticeship work thinned out, too. First drafts, basic research, formatting—the glue that used to train junior talent—now handled by Virtual Employees and Agents. That creates a new responsibility: if the bottom rungs are automated, leaders must build new on-ramps so people can still learn the craft alongside these digital coworkers.
Inside teams, the shift clicked when we stopped debating jobs and started designing workflows with the trio in mind: Humans for judgment and trust; Virtual Employees for context, memory, and orchestration; Agents for speed and consistency. Rebundle the process into a closed loop—intake → action → QC → delivery—and layers compress naturally. Titles follow the work. Jobs are bundles; workflows are systems.
Three truths to act on today
Mix change beats headcount change. Even if totals stay flat, composition won’t. Expect fewer traditional roles and more AI-adjacent ones.
Workflow math drives org math. When work gets rebundled into closed loops, titles and layers compress naturally.
Trust is the throttle. Keeping a human in the loop, plus guardrails and variance monitoring, is what lets you scale faster without breaking trust.
”Elevate the best, automate the rest” isn’t cruel—it’s candor.
Say the hard part plainly so we can protect the human craft work by rationalizing the glue work with Virtual Employees and Agents.
What the Leaders are saying (and how to read it)
Walmart’s Doug McMillon: AI will “change literally every job.”
Ford’s Jim Farley: AI could replace “literally half” of white-collar roles.
Amazon’s Andy Jassy: some jobs shrink, others grow, but overall corporate headcount will fall.
JPMorgan’s Marianne Lake: ~10% reduction in operations over five years.
Anthropic’s Dario Amodei: stop sugarcoating—entry-level roles will disappear, and we need new on-ramps.
Where the LLM labs are steering
The supply side is clear: agentic, tool-using models that reason across steps and take actions (your Agents), supported by systems that hold state and context over time (your Virtual Employees). Real-time, multimodal assistants that see, remember, and work across apps—i.e., the new fabric your Humans will direct.
In every platform shift, those who wait typically end up renting someone else’s playbook. This time, the first to figure out workflows will own the margins, the talent, and the trust. So, treat “half of white-collar” as a stress test. Use ~10% as a realistic baseline. Assume composition change everywhere—fewer traditional roles; more AI-adjacent roles (builder/operators who orchestrate Virtual Employees + Agents).
The Honest Conversation To Have With Yourself
Here’s where empathy means being direct. Sit down and ask yourself: where is AI already better than me at the work I’m doing today? Break it down task by task.
If you assume AI could take your job, you’ll get far more clarity about what you need to do to keep it—or how to evolve into something even stronger. You’ll see which tasks to hand to Agents, which to anchor in a Virtual Employee, and where your Human edge must stay.
The most empathetic move isn’t reassurance. It’s honesty that gives people a chance to thrive.
How to help yourself (start today)
Know where AI is already better than you today—and where you’re uniquely valuable. Use Virtual Employees + Agents to strip away glue work so you can spend more time on the parts that make you, you.
Map your week.
Pick one workflow and script it (input → process → check → send).
Build a tiny bench that mirrors the model: a Virtual Employee prompt that carries your context across tasks, plus two Agents (Researcher, Rewriter/Checker).
Add guardrails.
Measure before/after.
How to help your team (if you haven’t started)
Run one 3-week micro-pilot in a customer-facing workflow.
Staff it simply: ops lead + embedded builder + domain SME. Set targets (50–70% cycle-time cut, 30–40% throughput lift).
Protect trust by keeping a human in the loop, enforcing style guardrails, and monitoring variance with a dashboard.
Design the runbook with Humans + Virtual Employees + Agents from day one.
A Pragmatic Framework Approach
Work redesign to tasks, not titles.
Rebundle into closed loops. Explicitly assign the trio: Humans → judgment/relationship; Virtual Employees → memory/orchestration; Agents → repeatable execution.
Build capability and capacity where the work lives.
Certify agent/VE builders inside business units. Give them a real problem, a small toolset, and 2–3 weeks to ship.
Controls Govern what you scale.
Keep a human in the loop for high-impact outputs; add brand guardrails and data/IP policy; use a variance dashboard so errors surface—and get fixed—fast.
Measure Outcomes That Matters.
Track cycle time, error rate, and cost per unit before/after. Reinvest gains into pipeline, CX, and quality.
Bottom line
The peril is real—and leaders should be acting on it today. The labs are shipping agents; CEOs are reshaping orgs. Employees are already wondering where they stand. Your move is to redesign the work on the Humans + Virtual Employees + Agents model, build builder capacity, and govern what you scale—so people and systems deliver outcomes neither could achieve alone.
If you want to explore how this model fits your business or see virtual employees in action, please connect, michael@pvaiconsulting.com.
Onward,
Michael DeNunzio
Managing Director
Go back to Part 1 — Possibility: My One-Year-Old Just Became a Principal Consultant: Meet Auggie
Read it alongside this piece for the full picture: the upside of leverage and the risks and changes leadership teams should be acting on now — and that employees are already thinking about, often with concern.
michael@pvaiulting.com
