Skills-Based Workforce Planning Starts Below the Role

In February 2026 we brought HR leaders, workforce strategists, talent executives, and technology operators together in New York around a single question: how does AI fundamentally reshape enterprise value creation through work? What follows is the argument that came out of those panels, and what it means for skills-based workforce planning.

The central conclusion is also the least comfortable. The conversation has moved past experimentation, which means AI is no longer a technology discussion. It is an operating model decision. And across the panels one mismatch came through clearly: organizations approach AI at the role level, while value is created or destroyed at the task and workflow level.

Four structural shifts surfaced repeatedly. They are not independent themes but interlocking parts of one move: from job-centric design to task-centric value creation.

The unit of redesign is the task, not the role

Any job is the sum of the tasks it contains, which is why AI value does not materialize at the role level. The leverage sits lower, at the intersection of individual tasks and the workflows connecting them across teams and systems. Speakers kept returning to it: AI automates tasks, not jobs.

The Autor-Levy-Murnane task framework from labor economics gives this a usable foundation. Every role is a portfolio of activities spread across cognitive-routine, cognitive-non-routine, physical-routine, and physical-non-routine dimensions. With large language models and agentic AI increasingly capable of handling complex cognitive-routine work, and agentic systems beginning to operate across interconnected tasks rather than isolated activities, the decomposition has to happen at the task level.

Where to start is close to settled. Administrative friction is where time is consumed — compliance workflows, multi-system data entry, scheduling chains, reporting loops — and it is routine enough to suit LLM-augmented automation. Scheduling a hospital procedure means verifying insurance eligibility, checking clinician availability, entering data into several systems, and sending reminders. An agentic system can coordinate that end to end and flag exceptions, with staff stepping in only outside standard rules.

The measurement question follows immediately. There is a strong pull toward reading AI's impact as output per person; the argument made repeatedly was to drive value across the organization instead. Embed AI at each step of a workflow and the gains compound across functions, with individual productivity improving as a byproduct.

Governance belongs in the architecture

A consistent theme across the conference was that AI governance should not be treated as a compliance checkbox, but as a foundational design principle embedded into how enterprises build and operate — captured best by the line that just because we can does not mean we should. In practice that means controlled systems and human-in-the-loop frameworks. The EU AI Act, the NIST AI Risk Management Framework, and ISO 42001 were referenced as evidence that this is becoming structural, alongside enterprise safeguards such as LLM hardening, data sovereignty, zero trust architecture, and transparency in ethical sourcing.

Talent acquisition becomes a trust and judgment layer

AI is accelerating how organizations screen and hire, while recruiters bring judgment, observation, and contextual understanding that AI does not replicate. The notable development is that the human step is now described as a competitive differentiator rather than a legacy stage. AI-driven video interviews, piloted over the past two years, are being pulled back across industries — not because the technology failed, but because the candidate experience damaged employer brand.

Candidate use of AI is reshaping screening in parallel. When candidates clear interviews with AI assistance, the signals recruiters relied on — writing quality, prepared answers, case study responses — become less diagnostic. Skills-based assessments, live problem-solving, and structured behavioral interviews are gaining ground, because authenticity validation is now a design problem inside the hiring architecture.

That changes what TA should measure. A decade of optimizing time-to-fill, time-to-offer, and requisition aging shaped operating models, incentives, and technology purchases. Speed means little if decision quality is poor, and with AI shaping both sides of the process, activity metrics no longer indicate whether the right call was made. Quality of hire, skill alignment, and long-term fit survive that test.

The split between strategic and tactical recruiting has also turned structural.

• High-volume / tactical — Operating model: Speed-optimized, process-driven · AI role: Screening, scheduling, matching · Human role: Quality assurance, offer decision
• Specialized / mid-tier — Operating model: Balanced AI-recruiter model · AI role: Sourcing, market intelligence · Human role: Assessment, engagement
• Strategic / executive — Operating model: Relationship-led, long-horizon · AI role: Market mapping, compensation data · Human role: Full lifecycle advisory

Most TA teams already have AI-powered sourcing and screening available. The unresolved problem is how the tools get used: AI screening behaves much the same across roles and levels, while the real judgment calls happen between those categories. That is a capability gap, not a technology gap, and closing it is a development investment rather than a procurement decision.

Skills-based workforce planning treats capability as the balance sheet

Traditional workforce planning operates at the job-category level, and that altitude misses where AI's impact sits. A role that looks stable on the org chart may have had a third of its tasks automated in the past year, and a function that appears overstaffed may be under-skilled for the work that remains.

Capability modeling treats skills and task portfolios, not headcount, as the planning variables — skills as the balance sheet, repositioned from an HR exercise to an enterprise asset. As external hiring slows in highly automated areas, remaining gaps must be closed through internal development or redeployment.

The early signals are readable now. Job descriptions are getting denser as routine tasks are automated; whether that is healthy evolution or a warning depends on whether roles are being redesigned or responsibility is simply layered onto fewer people. Individual contributor ratios are rising as coordination and entry-level tasks get absorbed, and hiring is slowing in highly automatable areas while accelerating where judgment is required.

Two disciplines follow. Experimentation becomes an instrument of skills-based workforce planning — no experiment, no readiness — because pilots that test AI-augmented roles generate signal historical forecasting cannot. And pipeline becomes a planning responsibility: AI orchestration, human-AI workflow design, and cross-domain integration are not sitting in today's applicant pools. Without long-term pipelines, planning is forecasting decline.

Skills architecture has to be maintained, not completed

The pattern several organizations described is familiar. A mapping exercise is completed, the taxonomy validated, the model integrated into skills-based workforce planning tools — and then roles evolve, AI reshapes task composition, and the data drifts out of alignment with actual work. Durable architecture requires continuous calibration: validating skills data against real task execution, refreshing taxonomies, and putting governance around updates and ownership.

It also requires answering a more basic question: what qualifies as a skill. Whether something counts as a skill, a competency, or a tool proficiency determines how roles are profiled, how gaps are measured, and where learning money goes. Credibility comes from internal validation; models built from vendor taxonomies describe the market's view of a role, not the organization's work.

The capability baseline is shifting too. As routine cognitive work is automated, what remains is harder: judgment under complexity, validation, cross-functional coordination, ethical reasoning. Three capabilities are becoming baseline expectations rather than differentiators:

• Critical thinking — interrogating assumptions, evaluating the quality of evidence, and resisting the pull of AI-generated plausibility. The task is to teach people to challenge AI, not simply to use it.
• Learning agility — the capacity to learn, unlearn, and relearn as conditions shift.
• AI output review — judging when AI confidence matches actual accuracy, identifying embedded assumptions, and overriding a recommendation when human judgment says otherwise.

The choice in front of executive teams

What makes this wave different from earlier automation is not raw technical capability. It is structural impact. When tasks are reorganized into AI-enabled workflows, enterprise economics shift: productivity changes, skill requirements change, governance models change. Equally clear was that full automation is neither the goal nor the advantage. The human layer is a deliberate design choice rather than a fallback.

The paper closes with five executive imperatives for 2026 to 2028. Mandate work redesign: require priority functions to map work at the task level and rebuild workflows with AI embedded by design rather than layered on top. Shift workforce planning to capability modeling, treating skills and task portfolios as strategic assets and moving to quantified capability gaps and build plans. Redefine enterprise productivity metrics around workflow velocity, decision quality, and cross-functional impact rather than individual output. Embed governance into AI architecture from the design stage, aligned to regulatory frameworks and risk controls, with clear human accountability points. And build long-term AI-native talent pipelines internally — AI orchestration, validation, human-AI workflow design — rather than relying solely on external hiring.

The organizations that outperform will not be the ones that automate fastest. They will be the ones that redesign work most deliberately, embedding AI into workflows while preserving human oversight where it strengthens trust. The move from tasks to talent is architectural, and architectural decisions compound, which is why sequencing matters more than speed.