DECISION BRIEF 31
AI Talent Strategy: Build Capability, Not Headcount
Decision Brief 31
An AI talent strategy should build the capabilities the company must retain, not accumulate fashionable job titles. The CEO’s job is to decide which capabilities are strategically differentiating, which can be sourced temporarily, and when external access must become internal operating competence.

The conventional management assumption is that an AI skills gap is mainly a recruiting problem. It is not. Hiring can add expertise, but it cannot substitute for redesigned work, clear decision rights, usable data, management learning, or the ability of internal teams to operate and improve the system after a vendor leaves.
Executive answer: use a portfolio. Build critical judgment and workflow ownership inside the company; buy scarce specialists selectively; borrow speed from partners under an explicit transfer plan; and redesign existing roles so AI capability lives where business outcomes are owned.
Key Data: the shortage is a capability mix, not one missing title
The evidence does not support a single “hire more AI experts” response. It points to a blended capability system: technical specialists, AI-literate leaders, analytical and social skills, workflow owners, and broad foundational fluency.
| Evidence signal | What it means for the CEO | Limitation |
|---|---|---|
| The World Economic Forum’s 2025 survey found 63% of employers saw skills gaps as a major transformation barrier; 85% planned upskilling, 70% planned to hire new skills, and 50% planned to move people from declining to growing roles. | Employers are combining build, buy, and redeploy rather than relying on one lever. | Global employer survey; intentions are not realized outcomes. |
| The World Bank’s October 2026 East Asia and Pacific update says AI skill demand is often bundled with mathematics, critical thinking, communication, and collaboration. Only 13% of regional jobs were described as involving the complex thinking and judgment where AI support is most effective, versus 39% in advanced economies. | The scarce asset is not only model expertise; it is the ability to combine technology with judgment and context. | Regional labor-market evidence, not a firm-level capability audit. |
| The OECD’s January 2026 workforce guidance separates foundational AI skills for all staff, strategic AI knowledge for leaders, and deeper technical skills for digital and data professionals. | Different populations need different depth; one generic course or one elite team is insufficient. | Public-sector evidence; private firms must adapt it to their operating context. |
| The 2026 Stanford AI Index reports rising demand for skills required to deploy systems at scale, including workflow management, automation, and scalability; Singapore had the highest AI-job-posting share among the listed markets at 4.69% in 2025. | Operating AI reliably is becoming as important as building models. | Job-posting data are a demand signal, not proof of productivity or an Indonesia estimate. |
| The ILO estimated that 21.7% of Indonesian employment had more than minimal potential GenAI exposure in 2025, while finding no evidence of large-scale job losses across ASEAN. | CEOs should redesign roles and learning pathways before treating exposure as a headcount plan. | Exposure is not adoption, displacement, or realized value. |
Geographic scope: the strongest current evidence is regional or global. It informs the decision architecture for Indonesian and Southeast Asian companies, but each company still needs a local capability map, labor-market evidence, and workflow baseline.
The decision question
The CEO should decide where capability must reside, not approve a generic hiring target. The practical question is: which knowledge, judgment, data context, and operating responsibility would create unacceptable dependence if they remained outside the company?
For each priority workflow, answer four sourcing questions:
- What must the company build because it differentiates the business or carries material accountability?
- What should it buy through selective hiring because the expertise is scarce, persistent, and central?
- What can it borrow from a partner because speed matters and the capability can be transferred?
- What should it redesign into existing roles because the work belongs close to customers, operations, or risk?
The options are complementary, but the proportions matter. A portfolio without a capability destination can become a collection of contracts, courses, and titles with no durable operating advantage.
Point of View
AI talent strategy is a make-or-buy decision for organizational capability. Headcount is only one input; the outcome is whether the company can choose, operate, govern, improve, and stop its AI-enabled workflows without hidden dependency.
This challenges the legacy pattern of starting with an organization chart: appoint an AI leader, hire a few engineers, buy training, and expect adoption to follow. The sequence should run in the opposite direction—from strategic workflow to capability, from capability to sourcing, and from sourcing to accountable roles.
The sharper rule is:
> Borrow expertise for speed, but never borrow accountability. If the capability is strategic, every external engagement must leave the internal system stronger.
What AI changes—and what remains human leadership
AI changes the shelf life and composition of skills. Technical methods evolve quickly, workflow-level deployment skills gain importance, and more employees can perform tasks that once required specialist access; none of this removes the need for human judgment about strategy, people, risk, and institutional responsibility.
| AI can augment or automate | Irreducible CEO and human responsibility |
|---|---|
| Map tasks, infer skill adjacency, and suggest learning paths | Decide which capabilities differentiate the company |
| Generate code, analysis, documentation, and training simulations | Set workforce principles and fair-transition commitments |
| Match internal talent to projects and detect capability gaps | Judge performance, potential, trust, and context |
| Monitor workflow quality, overrides, cost, and skill use | Assign decision rights and accept material risk |
| Accelerate onboarding and make expert knowledge reusable | Decide when external dependence is strategically unacceptable |
AI also changes the economics of expertise. A small specialist group can codify patterns for a wider workforce, while non-technical domain experts can build or supervise more sophisticated workflows. The operating model must therefore reward knowledge transfer and workflow outcomes—not specialist heroics or course completion.
Thought Process
The recommendation follows four explicit steps. Each step prevents the company from confusing access to talent with ownership of capability.
- Start with the value-critical workflow. Strategy should name the customer, operating, or risk outcome before HR names a role.
- Decompose the capability. Separate domain judgment, data knowledge, workflow design, technical engineering, change leadership, and governance.
- Choose a source for each component. Build, buy, borrow, or redesign according to criticality, scarcity, learning speed, and dependency risk.
- Require evidence of transfer. Scale only when internal owners can operate, evaluate, and change the workflow with acceptable external support.
This is an inference from the evidence, not a universal staffing ratio. A regulated bank, industrial manufacturer, public institution, and digital marketplace will keep different capabilities inside.
Four sourcing options and their trade-offs
No option wins in every context. The CEO needs a portfolio that maximizes speed and learning without outsourcing strategic judgment or creating a brittle concentration of expertise.
| Option | Best fit | What must be true | Main trade-off |
|---|---|---|---|
| Build internally | Capability is differentiating, recurrent, data-intensive, or accountability-sensitive | Protected learning time, real workflow access, strong managers, and a path to practice | Slower start; opportunity cost; risk that training remains theoretical |
| Buy through hiring | Scarce expertise is persistent and will be used across multiple high-value workflows | Clear mandate, credible technical environment, executive sponsorship, and role integration | Hiring premium; concentration risk; expert may become an isolated service desk |
| Borrow from partners | Speed matters, capability is new, and external patterns can shorten learning | Contracted knowledge transfer, data/control boundaries, documentation, exit rights, and internal shadow owners | Vendor dependence; incentives may favor continued complexity rather than transfer |
| Redesign existing roles | Domain judgment is central and AI can expand the leverage of current teams | Task redesign, decision rights, manager capability, measurement, and psychological safety | Adoption friction; hidden workload; inconsistent standards if support is weak |
Recommended default: borrow to accelerate, build to retain, buy only the scarce core, and redesign the surrounding work. A partner should compress the learning curve, not become the operating model.
Five CEO gates for AI talent strategy
The five gates turn a talent plan into a capability investment decision. A proposal should not pass because a candidate is impressive, a vendor is credible, or a training catalogue is large.
Gate 1 — Strategic criticality
Keep strategic judgment and accountable workflow ownership inside the company. External support may be deep, but the company must retain the ability to decide what the system should optimize, what risk it may take, and when it must stop.
Ask:
- Does this capability shape customer value, economics, trust, or regulatory responsibility?
- Would losing external access materially interrupt a priority workflow?
- Could a supplier’s incentives diverge from the company’s?
- Does the capability compound through proprietary data, context, or repeated learning?
If the answers are mostly yes, the destination is internal capability even when the starting source is external.
Gate 2 — Work and skill decomposition
Define the capability as observable work, not a title. “AI engineer,” “prompt expert,” or “AI product manager” can hide very different requirements across domain judgment, data, integration, evaluation, security, change, and governance.
Required evidence:
- the workflow and outcome;
- the tasks AI will augment or automate;
- the skills needed at foundational, leadership, and specialist levels;
- the human decision rights at normal and exceptional cases;
- the capability gap that remains after role redesign.
The gate fails if recruitment begins before the work and accountability are clear.
Gate 3 — Learning and transfer
Every borrowed capability needs a transfer mechanism and a measurable handover. Documentation alone is not transfer; internal people must demonstrate that they can operate, evaluate, troubleshoot, and improve the workflow.
Require:
- named internal shadow owners;
- paired delivery rather than vendor-only execution;
- reusable playbooks, evaluation sets, and decision logs;
- protected practice time;
- a transfer checkpoint before contract expansion.
The gate fails if only the vendor can explain why the system works, change it safely, or interpret its failures.
Gate 4 — Operating model and accountability
Place capability where business outcomes are owned, supported by a small enabling core. A centralized team can set standards and provide scarce expertise, but value realization belongs with the function that owns the workflow.
Verify:
- one business owner owns the outcome;
- managers know how roles and performance measures change;
- data, platform, risk, HR, and finance responsibilities are explicit;
- specialists have a community of practice rather than isolated reporting lines;
- stop and escalation rights are independent of delivery pressure.
The gate fails when an AI center becomes a queue that business teams can blame and managers can bypass.
Gate 5 — Economics and resilience
Compare the fully loaded cost and resilience of the portfolio, not only salary or day rates. Include recruiting delay, ramp time, partner cost, management attention, data/platform access, training, attrition, quality control, and exit costs.
Test:
- time to first accepted outcome;
- cost per accepted outcome;
- time for an internal team to make a safe change;
- single-person and single-vendor dependency;
- retention and internal mobility;
- the cost of pausing, replacing, or rebuilding the capability.
The lowest day rate can create the highest strategic dependency. The highest-paid hire can also fail if the operating environment prevents leverage.
Decision rights
AI talent strategy needs explicit decision rights because capability choices allocate capital, shape workforce trust, and determine who remains accountable. The CEO should retain portfolio choices while delegating delivery evidence.
| Right | Accountable role |
|---|---|
| Decide strategic capabilities, workforce principles, dependency tolerance, and scale thresholds | CEO / executive committee |
| Delegate capability delivery and workflow redesign | Business value owner with CHRO and technology partner |
| Instrument skills, accepted outcomes, transfer progress, cost, and dependency | HR, operations, finance, data/platform |
| Escalate material labor, data, security, customer, regulatory, or vendor risk | HR, risk, legal, security, business owner |
| Stop expansion when transfer, quality, safety, or integrity conditions fail | Named stop authority independent of the delivery team |
The board should ask whether the company is becoming more capable, not whether the AI team is becoming larger.
The smallest credible 90-day test
A 90-day test can reveal whether the sourcing portfolio creates a durable capability. Use one high-value workflow, one accountable business owner, one internal delivery pair, and one bounded partner or specialist contribution.
Days 1–15: map the capability
- Baseline the workflow’s accepted outcome, cycle time, quality, cost, exceptions, and risk.
- Decompose the capability into domain, data, workflow, technical, change, and governance components.
- Mark each component build, buy, borrow, or redesign.
- Name internal shadow owners and pre-register success, failure, and stop conditions.
Days 16–45: paired delivery
- Run a mixed internal-external squad on real but bounded work.
- Use the actual data and control environment at proportionate risk.
- Record decisions, evaluation criteria, exceptions, and changes.
- Measure accepted outcomes and the internal team’s contribution—not training attendance.
Days 46–75: transfer and independence
- Have internal owners operate the workflow and make at least one safe change.
- Move the partner from executor to reviewer.
- Test absences, handoffs, difficult cases, and documentation.
- Measure time to diagnose, change, recover, and explain.
Days 76–90: decision
- Scale if the workflow meets its business and quality thresholds, internal owners can operate and modify it, fully loaded economics remain acceptable, and no critical dependency lacks an exit path.
- Redesign if outcomes improve but transfer, manager adoption, role clarity, or economics miss the threshold.
- Stop for a material data, safety, labor, customer, or integrity breach; inability to explain or reverse the workflow; persistent vendor-only knowledge; or no evidence that capability is compounding internally.
The test deliberately avoids universal numeric thresholds. The owner must set thresholds before launch using baseline variation, risk, and the company’s strategy.
Decision gate
The CEO should fund scale only when five statements are true. If one is false, the next decision is to narrow, redesign, or stop—not to add more people or another vendor.
- We can name the business capability and accepted outcome.
- We know which elements must remain inside and why.
- Internal owners can operate, evaluate, and change the workflow.
- Decision rights, workforce implications, controls, and stop conditions are explicit.
- The fully loaded economics and dependency profile are defensible.
Implications for decision-makers
CEOs should treat talent as a capability portfolio tied to a small set of strategic workflows. Broad “AI headcount” targets invite activity without clarifying value or dependency.
Boards should ask for a capability and dependency map. The critical questions are what the company could not operate without a named person or supplier, what learning has transferred, and whether risk ownership remains internal.
CHROs should move from course catalogues to skill evidence in real work. Foundational literacy, manager capability, specialist depth, and internal mobility need different measures and learning environments.
COOs and business leaders should own workflow outcomes and create practice opportunities. People cannot build operating capability in a classroom while production work remains unchanged.
CAIOs, CIOs, and CTOs should build standards, platforms, evaluation methods, and communities of practice that make distributed capability safe. They should not become permanent owners of every business use case.
Testable hypotheses
These hypotheses make the brief falsifiable. They should be reviewed at day 90 and again after six months.
- A capability-sourcing portfolio tied to one workflow will reach an accepted business outcome faster than a role-first hiring plan.
- Paired delivery with a transfer checkpoint will reduce vendor-only dependency without materially slowing the initial pilot.
- Role redesign plus targeted specialists will produce wider adoption and better exception handling than an isolated expert center.
- Measuring internal time-to-change will reveal strategic dependency that salary and contract-rate comparisons miss.
- Foundational fluency for all staff, strategic knowledge for leaders, and deep skills for specialists will outperform one generic training programme.
Evidence ledger and source notes
The evidence supports a blended capability strategy, not a universal staffing formula. It was reviewed on 26 October 2026.
- World Bank, East Asia and Pacific Economic Update, 6 October 2026. Regional evidence that AI adoption remains below advanced-economy levels, AI skills are bundled with analytical and social skills, and adoption/adaptation of existing tools is the near-term pathway for most countries. https://www.worldbank.org/en/publication/east-asia-and-pacific-economic-update
- World Bank, “East Asia and Pacific Sees Steady Growth Amid Global Uncertainty,” 6 October 2026. Reports that 13% of regional jobs involve the complex thinking and judgment where AI support is most effective, versus 39% in advanced economies. https://www.worldbank.org/en/news/press-release/2026/10/06/east-asia-and-pacific-holds-steady-amid-global-uncertainty-but-harnessing-ai-s-ful
- World Economic Forum, Future of Jobs Report 2025, 7 January 2025. Survey of more than 1,000 global employers representing over 14 million workers; employer intentions are not realized outcomes. https://www.weforum.org/publications/the-future-of-jobs-report-2025
- OECD, “Building an AI-ready public workforce,” 19 January 2026. Distinguishes foundational skills for all staff, strategic knowledge for leaders, and technical depth for digital/data professionals. Public-sector evidence requires private-sector adaptation. https://www.oecd.org/en/publications/building-an-ai-ready-public-workforce_b89244c7-en/full-report.html
- Stanford HAI, 2026 AI Index Report, Economy chapter, 2026. Job-posting evidence shows growing demand for deployment, workflow, automation, and scalability skills; geography and posting coverage limit direct transfer to Indonesia. https://hai.stanford.edu/assets/files/ai_index_report_2026_chapter_4_economy.pdf
- International Labour Organization, “AI may affect nearly 80 million workers in the ASEAN region,” 8 July 2026. Estimates 21.7% of Indonesian employment had more than minimal GenAI exposure in 2025 and reports no evidence of large-scale job losses across ASEAN. Exposure is not displacement. https://www.ilo.org/resource/news/ai-may-affect-nearly-80-million-workers-asean-region-large-scale-job
- OECD, “Empowering the Workforce in the Context of a Skills-First Approach,” 24 June 2025. Skills-first practices may widen talent pools, but weak implementation can introduce bias, erode standards, and accelerate skill obsolescence. https://www.oecd.org/en/publications/empowering-the-workforce-in-the-context-of-a-skills-first-approach_345b6528-en/full-report/practical-considerations-for-a-skills-first-approach_88eb89d7.html
- Microsoft, “Work Trend Index 2026: 33% of Indonesian Workers are at the Forefront of AI Adoption,” 30 June 2026. Indonesian user survey highlights critical thinking and AI-output quality control; vendor-sponsored and based on self-reported users, so it is directional. https://news.microsoft.com/source/asia/2026/06/30/microsofts-work-trend-index-2026-33-of-indonesian-workers-are-at-the-forefront-of-ai-adoption/
Continue Reading
- AI Workforce Strategy: Redesign Tasks Before Cutting Jobs
- Chief AI Officer: When the CEO Should Create the Role
- Agentic AI Operating Model: Redesign Work Before Scale
- Digital Transformation: Who Owns the Business Value?
Discuss
If your company can access AI expertise but cannot operate, evaluate, or change the workflow without the same expert or supplier, it has rented capability rather than built it. Start a discussion with Antovany Reza.
About the author
Antovany Reza writes CEO Decision Lab for leaders turning AI and digital transformation into business value, stronger judgment, and accountable execution across Indonesia and Southeast Asia.
Leave a Reply