AI Jobs Transition (6): Turn the framework into an operating plan

AI Jobs Transition · 6/6

A labor-market framework becomes useful inside a company only when it changes a decision.

The goal is not to assign every employee an automation-risk score. It is to understand a workflow well enough to decide whether AI should remove tasks, change the human role, expand the service, or remain under observation.

Here is a practical operating loop for doing that.

Start with work, not job titles

Choose one workflow with a clear customer, output, owner, and measurable volume. Map the tasks, handoffs, decisions, tools, and exceptions. Estimate where time is spent rather than relying on the job description.

Then mark four things:

1. tasks AI can perform reliably today;
2. decisions that require accountability, trust, or physical action;
3. demand that could appear if cost or waiting time falls;
4. current usage, including unofficial personal use.

This produces a transition hypothesis grounded in actual work.

Run a bounded delegation experiment

Do not begin with company-wide deployment. Pick a task class, define the inputs and completion criteria, limit permissions, and specify when the agent must escalate.

Measure cycle time, verified quality, human repair time, exception rate, and cost per successful outcome. A faster first draft is irrelevant if reviewers spend the saved time finding subtle errors.

The experiment should answer whether the workflow can be delegated—not whether the model can generate a plausible output once.

Redesign the role around the remaining bottleneck

If AI removes routine work, do not leave the role as an empty version of its old description. Identify what becomes more valuable: judgment, customer relationships, exception handling, system improvement, or physical delivery.

Update decision rights, workload limits, training, evaluation, and career paths. Otherwise the company captures throughput while workers inherit invisible supervision and accountability.

Test whether productivity can become growth

Before treating efficiency as a headcount plan, test the demand side. Offer faster delivery, a lower-priced tier, a new customer segment, or a more customized product. Watch whether additional demand is real and economical to serve.

If demand expands, invest in distribution and the new bottlenecks. If it does not, plan the workforce transition honestly instead of hiding it behind an augmentation narrative.

Build an early-warning dashboard

Track a small set of signals by workflow and role:

• share of task volume delegated to AI;
• verified success and human repair rates;
• output per worker and total output;
• demand, price, backlog, and customer mix;
• hiring, internal mobility, wages, and attrition;
• concentration of exceptions and accountability.

No single metric tells the story. Together they show whether change is automation, reorganization, expansion, or delayed adoption.

Assign an owner to the transition

AI transformation fails when model deployment belongs to one team, workforce planning to another, risk to a third, and no one owns the combined outcome.

Each workflow needs a directly responsible person who owns productivity, quality, worker impact, customer impact, and escalation. The owner should revisit the classification as models, regulation, demand, and organizational capability change.

The durable practice is not predicting the future correctly once. It is building a system that notices when the mechanism changes and can respond before the org chart becomes the last place to learn about it.

Sources

• Alex Martin Richmond, The AI Jobs Transition Framework, OpenAI Economic Research, April 2026.

AI Jobs Transition series

• Start: AI can do your job. That does not mean it will replace you.
• Previous: AI will automate, reorganize, expand, and sometimes barely change work