A payroll target cannot establish what an agent can replace.
Can the 2027 workforce budget fall? That possibility deserves examination. Start by identifying work that has disappeared and work that remains. Faster drafts do not establish end-to-end replacement. If review and correction move to other employees, total organizational workload needs to be recalculated.
Part I examined investment signals and gaps in preparation. The departmental comparisons and decision process below are the author’s proposals, not a validated staffing model or a forecast of replacement rates.
Different departments need different evidence.
| Area | Examples to test with an agent | Capacity to verify before changing staffing |
|---|---|---|
| Development | Proposed code changes, test runs, documentation | Requirements, security review, incident response, maintenance |
| Marketing | Content variants, campaign setup assistance, results compilation | Customer fit, accurate messages, outcome validation |
| Planning | Research, option comparisons, initial plans | Assumption checks, cross-functional agreement, execution resources |
| HR | Documents, scheduling, policy lookup, evidence organization | Corrections, employee explanations, exception review |
| Finance | Transaction classification, reconciliation assistance, draft reports | Unusual transactions, internal controls, verification, approval |
These are candidates for testing, not claims that any current agent can perform them autonomously. Systems, data, permissions and risk determine scope. Manufacturers need to examine physical process constraints; software companies need to consider deployments and incidents; financial and professional services need to examine review and exception workloads.
Compare difficult days as well as routine work.
Compare the existing workflow with agent-assisted work using equivalent scope, demand and quality standards. Record reviewer time, corrections, reprocessing, unresolved work and contractor spending alongside completed output. There is no universal trial length or sample size. Include consequential conditions such as month-end closing, demand spikes and staff absence.
Consider a hypothetical development team whose agent reduces coding time but increases security review and incident work. Counting only the coding savings misses the burden on operations. Conversely, if total workload falls while quality and response capacity hold, there is evidence for reviewing staffing within that tested scope. This is an illustration, not a documented company case.
If people leave but their work remains, the cost returns somewhere else.
Distinguish released time from reduced expenditure.
Time saved, hiring deferred and cash spending reduced are different outcomes. Unchanged salaries do not become immediate cash savings because a task is faster. If the team absorbs additional demand, verify that contribution. Include software, integration, security, review, correction, contracting and transition costs within the same comparison.
Gartner’s autonomous-business survey found nearly equal workforce reduction rates among respondents reporting higher ROI and those reporting modest or negative outcomes. Fewer employees alone cannot establish better investment returns. The survey covered 350 executives at enterprises with at least USD 1 billion in annual revenue and included more than AI agents. It does not establish the outcome of reducing staff in a particular Korean department. [1]
| Observed result | Staffing option to examine | Evidence still needed |
|---|---|---|
| Less work, but unchanged spending and demand | Redeployment, training or quality improvement | Demand and skills for the new work |
| Existing staff and agents handle expected additional demand | Adjust the timing of some new hiring | Peak demand and exception capacity |
| A departing employee’s work has transferred reliably | Review whether that position needs replacement | Hidden coordination, review and customer-support gaps |
| Total work and required capacity have fallen sustainably | Examine restructuring options | Quality, recovery capacity, transition costs and applicable processes |
Hiring, replacement hiring, redeployment and workforce reduction are different decisions. This table does not determine which individual should leave. A person’s value and the workload attached to a role are not identical. Evidence of changed work can inform a review of roles and skills; actual personnel decisions require separate consideration of individual circumstances and applicable processes.
Executives should specify when their numbers must change.
Build three scenarios: a base plan using demonstrated agent capability; an expansion plan activated when a wider workflow can be handled reliably; and a response plan restoring human coverage when quality falls or costs rise. This is an author proposal, not a model predicting AGI arrival.
For each scenario, state the work scope, budget ceiling, review date and conditions for expansion or withdrawal. A model announcement alone cannot justify a staffing change. Examine whether outcomes have changed in the company’s own systems. Executives who set the initial target should also authorize revisions when the evidence changes.
Can the organization still deliver to customers?
The decision appears in a cost sheet. Its consequences appear in products and services. Can the company correct defects after a faster release? Can it resolve difficult cases after automated support? If throughput rises but customers must repeat their problem, what improved? These questions keep delivery capacity within the workforce plan.
A useful 2027 plan needs evidence of work the agent performs and work still requiring people. Unnecessary tasks can be removed. New capabilities may need funding. Neither decision follows from a headcount target alone.
Sources and limitations
Reviewed as of September 30, 2026. Source [1] is a 2026 release based on a third-quarter 2025 survey. Read the Korean investment and organizational preparation evidence alongside Part I. Departmental examples, the hypothetical development team, tables and scenarios are author proposals. They do not establish an agent’s capability, a workforce replacement rate or an amount of savings. This article is not legal, employment or accounting advice for a particular staffing decision.