Give AI the rote work: anything repeatable with clear inputs and outputs, like processing, pattern recognition, summarization and task automation. Keep judgment work with people: relationships, creative decisions, cross functional calls, and the discernment to confirm that repeatable output still maps to the outcome you wanted.
Knowing where that line sits, and being willing to move it as your tools change, is the new core competency. Most teams get stuck because they sort work by function and job title, when the useful sort is by the nature of the work itself.
Rote Work vs. Judgment Work
Sorting work starts with one question about each task: does this have clear inputs, a clear output, and does it repeat? If yes, it belongs with a specialist, human or machine. If the answer depends on context, relationships or a call you'd want to defend in a room, it stays with a person.
Here's how I put it on the record:
"AI can manage a lot of that specialty work. The work with very clear inputs and outputs. Think processing, pattern recognition, summarization, task automation, anything that is wrote, anything that is repeatable, AI can often handle much better than a human."
Britt Gage, founder, oAT
| Task | Type | Who or what owns it | Reasoning |
|---|---|---|---|
| Summarizing a week of support tickets into themes | Rote | AI | Defined input, defined output, runs the same way every week |
| Deciding which ticket theme becomes a roadmap item | Judgment | Generalist operator | Requires context from sales, product and finance at once |
| Cleaning, tagging and deduping a lead list | Rote | AI | Pattern recognition at volume |
| Choosing which five accounts get founder attention this quarter | Judgment | Founder with an operator | Relationship read, not a data lookup |
| Drafting the recurring investor update from known metrics | Rote | AI | Template plus data |
| Deciding how to frame a bad quarter to the board | Judgment | Founder | Consequence lands on a human |
| Running an established onboarding checklist | Rote | AI or specialist | Consistency is the whole point |
| Designing onboarding for a team that just doubled | Judgment | Generalist operator | The process doesn't exist yet |
AI Is a Specialist
The clarifying move is to stop treating AI as a general purpose replacement for people and give it a job description.
"Because that's really what AI is. AI is a specialist."
Britt Gage, founder, oAT
A specialist is excellent inside a defined box: clear scope, clear inputs, repeatable output, measurable consistency. That describes a good data processor, a good transcriptionist, a good report builder. It also describes what current AI tooling does well. Treating AI as a specialist gives you three practical handles:
- Scope it. Write the box down. What goes in, what comes out, what good looks like.
- Review it. Specialists get their work checked against outcomes. So does AI.
- Retire it. When the work changes shape, the box changes or closes. No reorg required.
Research on human and machine collaboration points the same direction: the largest gains show up where companies design for complementary roles rather than straight substitution [1].
Traditional Design and Dynamic Design Ask Different Questions
The org chart used to be humans in specialized functions, sitting in boxes. That structure forces work into the shape of the chart. Dynamic teaming inverts it, because the work moves faster than any chart can be redrawn.
| Traditional org design | Dynamic org design | |
|---|---|---|
| Opening question | Who reports to whom in order to get this work done? | What does this work require, and who can deliver it? |
| What gets assigned | People already in boxes | Skills and capacity mapped to current work |
| How resources move | Through a reorg | Toward the work that needs them |
| What the chart represents | A chain of command | A living map of capability |
| Where AI sits | An add-on tool inside a function | A specialist resource in the mix |
Organizational barriers, siloed teams among them, are consistently identified as the reason AI programs stall rather than the technology itself [2].
how We Do: Spotting What Can Be Handed Off
Over the past five years we've been developing this as an operating model. The handoff is a repeating cycle inside each phase of work.
1. Form around the work in front of you
A project kicks off, a quarter starts, a workstream opens. Map the skills and capacity the work requires, then pull resources toward it. You're structuring the team around the work, and that includes deciding which pieces AI already handles.
2. Learn by doing, then get discerning
Midway through, you have evidence. Get reflective about it. Which steps came out the same way three times? Which inputs reliably produced the result you wanted? Which parts still needed someone to think?
3. Hand off what has proven repeatable
Take the steps that ran with consistency and give them to the specialized input: AI, or a person with clear roles and responsibilities. Write down the inputs, the expected output and the check. Anything you can't describe that precisely is not ready to leave human hands.
4. Flow when the work changes
At the end of the cycle the work has moved. Shift to the new phase and run the sort again. Nobody waits for permission to restructure. Reported AI adoption gains concentrate in organizations that keep redesigning workflows as capability improves [3].
What Stays With Generalists
"But what about the judgment calls? What about the relationships? What about those creative decisions? What about the cross functional work? That's what makes work go. That's the work that stays human."
Britt Gage, founder, oAT
The work that stays with people has a common signature: it spans boxes, it carries consequence, and the inputs are never fully specified in advance.
- Judgment calls. Choosing between two defensible options with incomplete information.
- Relationships. Clients, candidates, investors, teammates. Trust doesn't transfer to a tool.
- Creative decisions. Naming the thing, framing the offer, deciding what to make.
- Cross functional work. The connective tissue between marketing, sales, ops and product, where nobody owns the whole.
- Discernment about AI output. Confirming that consistent, repeatable output is still mapping to the outcome you're after.
That last one is the piece leaders underestimate. Handing work to a specialist creates a new job: making sure the specialist's consistency is pointed at the right target.
Who Maintains the Handoff
Every piece of rote work you move to AI needs an owner for the judgment layer around it. Without that owner, you get consistent output drifting away from the outcome, quietly, for a quarter.
- Input owner. Someone responsible for the quality of what goes in.
- Outcome check. A named cadence where output gets compared against the result it's supposed to produce.
- Escalation path. A person who decides when the box needs rewriting.
- Retirement call. A person willing to say this handoff stopped working.
This maintenance work sits naturally with generalists, because judging whether marketing output is serving a revenue outcome requires seeing both.
What Redrawing the Line Changes
Moving the boundary between rote and judgment work changes decisions you may think of as settled:
- Who you hire. Fewer roles defined by a repeatable output, more capacity for judgment and cross functional range.
- How you staff. Resources flow toward current work, including people, time, attention and AI capacity.
- How you measure performance. Consistency measures the specialist. Outcomes measure the generalist.
- What technology you choose. Tool selection follows the shape of the work you've mapped.
Analyses of skills based operating models describe a similar shift away from the fixed job as the unit of organizing [4].
Where Embedded Generalists Fit
oAT places embedded generalist operators inside companies. An operator joins your team, works in your tools and covers core functions so you stop carrying all of it personally. Functions we cover include marketing, GTM, sales, business ops, customer success, product, AI workflows, community and chief of staff support.
In a dynamic teaming context, an embedded generalist does three things at once:
- Owns the judgment work that spans functions
- Builds and maintains the AI handoffs for the rote work underneath it
- Reforms around the next phase when the work moves
Advice is cheap and outsourcable to AI. We'd rather be the MVHs, most valuable humans in the room.
FAQ
What counts as "rote work" that AI should own?
Rote work has clear inputs, a clear output, and repeats with consistency. Processing, pattern recognition, summarization and task automation are the common examples. A useful test: if the inputs, the expected output and the quality check can all be written down precisely, the work is ready for a specialist, human or AI.
What is "judgment work" and why can't AI do it?
Judgment work involves incomplete information, competing defensible options, and consequences that land on people. Relationships, creative decisions and cross functional calls fall into this category, along with deciding whether automated output is still serving the intended outcome. The inputs are never fully specified in advance, which is exactly what a specialist function requires.
Is AI replacing specialists or acting as one?
AI functions as a specialist: excellent inside a defined scope with clear inputs and repeatable outputs. Framing it that way lets a leader scope the work, review the output against outcomes, and close or rewrite the scope when the work changes. It also clarifies that the judgment layer around any specialist still needs a human owner.
How do you know when repeatable work is ready to hand off?
Repeatable work is ready when it has run the same way several times and someone can document the inputs, the expected output and the check that proves it worked. Work that resists that documentation is still judgment work. The mid-cycle moment in a project, after a team has learned by doing, is usually where the candidates become visible.
What role do generalists play once AI takes over rote tasks?
Generalists own the judgment work that spans functions and the maintenance of every handoff made to AI. That includes confirming that consistent output still maps to the intended outcome, deciding when an automated scope needs rewriting, and reforming the team around the next phase of work. Handing rote tasks to a specialist increases the amount of judgment work, and someone has to hold it.
References
- Wilson, H. James, and Paul R. Daugherty. "Collaborative Intelligence: Humans and AI Are Joining Forces." Harvard Business Review, July 2018. https://hbr.org/2018/07/collaborative-intelligence-humans-and-ai-are-joining-forces
- Fountaine, Tim, Brian McCarthy, and Tamim Saleh. "Building the AI-Powered Organization." Harvard Business Review, July 2019. https://hbr.org/2019/07/building-the-ai-powered-organization
- McKinsey & Company. "The state of AI." https://www.mckinsey.com/capabilities/quantumblack/our-insights/the-state-of-ai
- Deloitte Insights. "Global Human Capital Trends." https://www2.deloitte.com/us/en/insights/focus/human-capital-trends.html