Skip to content

Practitioner Guide: Capability Building Roadmap

What This Guide Builds

A capability-building roadmap is the operational plan that closes the gap between the talent an organization has and the talent its AI ambitions require. It turns the Talent & Capability Framework into a sequenced program: who needs to reach which literacy level, by when, through what mechanism, and how you will know it worked.

The urgency is documented and the default is failure-by-neglect. The Microsoft and LinkedIn Work Trend Index found that 75% of knowledge workers already use AI at work, yet only 39% of those users have received any AI training from their employer, and just 25% of companies planned to offer generative-AI training that year (Microsoft & LinkedIn, 2024; Microsoft & LinkedIn, 2024; Microsoft & LinkedIn, 2024). The same study found 78% of AI users are bringing their own tools to work ("BYOAI") — using AI with no guidance, no guardrails, and no shared standard (Microsoft & LinkedIn, 2024). A roadmap is how you replace ungoverned, self-taught usage with deliberate capability. This guide gives you the phases to build one.

Capability building is not a training catalog. It is a sequenced program with targets, owners, delivery mechanisms, and measurement — built on top of role redesign and feeding the talent-readiness assessment.

Phase 0 — Baseline and Target-Set

Before designing any program, establish two things:

  • Where people are. Run the talent-readiness assessment to baseline current literacy by role and population. Expect the Work Trend Index pattern locally — high usage, low training, lots of BYOAI (Microsoft & LinkedIn, 2024).
  • Where they need to be. Pull the capability deltas from role redesign. Set differentiated targets: the broad workforce to Fluency (Level 2), a meaningful minority to Builder (Level 3), and a small core at Architect (Level 4).

The output is a simple matrix — population × current level × target level × deadline — that the rest of the roadmap fills in. Note the regulatory floor: the EU AI Act's Article 4 AI-literacy obligation has applied since 2 February 2025, so a baseline-awareness program for anyone touching AI systems is a compliance requirement, not just good practice (European Commission, 2024).

Phase 1 — Foundational AI Literacy (Everyone)

The first program is universal and shallow: bring the entire workforce to Awareness. Cover what generative AI is and is not, the failure modes (hallucination, data leakage, bias), responsible-use and data-handling rules tied to your governance policy, and a first hands-on session with the sanctioned tools.

Keep it short, mandatory, and role-agnostic. The goal is a common floor and a common vocabulary, not expertise. This phase also satisfies the EU AI Act Article 4 obligation for staff operating AI systems (European Commission, 2024).

Phase 2 — Applied Fluency (Most Knowledge Workers)

This is the phase that produces most of the value, and it cannot be delivered by a generic course. Fluency is domain-specific — a marketer, an analyst, and a support agent need different prompts, examples, and workflows. Design it as function-specific, hands-on, and anchored in real tasks drawn from the workflow and role-redesign work.

Treat "prompt engineering" as a general literacy, not a specialist track. The standalone prompt-engineer role collapsed precisely because strong prompting became an expected baseline skill rather than a separate job (TechRepublic, 2025) — so teach prompting, verification, and calibrated trust to everyone in the fluency program rather than concentrating it in a few specialists. Reinforce that augmentation lifts less-experienced workers most (the ~34% novice gain in the support-desk study), which makes fluency programs especially high-leverage for early-career and frontline staff (Brynjolfsson et al., 2023).

Phase 3 — Builders and Specialists

A smaller program develops the Builder layer — people who compose agents, automations, and retrieval workflows on top of the platform. Builders are the multiplier between the specialist core and the fluent majority; under-investing here leaves the platform unused. Select for aptitude and interest surfaced in Phases 1–2, give them deeper hands-on training and sandboxed build time, and connect them to the architect/specialist core.

For the Architect/specialist layer, capability is usually closed by hiring and partnering rather than internal development, because the depth required is too slow to grow from scratch — and because the market premium for that talent is steep (a 56% wage premium for AI skills) (PwC, 2025).

Phase 4 — Champions Network and Communities of Practice

Programs deliver the baseline; champions sustain and diffuse it. Stand up a network of practitioners — one or more per function — who are a level ahead and who coach peers, surface working use cases, and translate central guidance into local language. This is the same engine the adoption program relies on, and it is the answer to the training gap: when 75% use AI but only 39% are formally trained (Microsoft & LinkedIn, 2024; Microsoft & LinkedIn, 2024), peer-to-peer diffusion is already how most learning happens — the champions network makes it deliberate and good rather than accidental and uneven.

Support champions with a community of practice: a recurring forum, a shared prompt-and-pattern library, and visible recognition. Decide your operating model — a central center of excellence, a federated champion model, or a hybrid — and resource it; champion time must be real and protected, not volunteered on top of a full job.

Phase 5 — Build, Buy, or Partner for Delivery

Each program above can be delivered three ways, and the choice should be explicit per layer:

  • Build (in-house). Best for fluency, which must be tailored to your functions and tools. Generic courses underperform here.
  • Buy (off-the-shelf training and certification). Efficient for the universal awareness floor and for standardized technical curricula. Large enterprises have committed at scale — Amazon's Upskilling 2025 pledged more than $1.2 billion to provide skills training to 300,000 employees (up from an initial $700 million for 100,000 in 2019) (Amazon, 2021) — but spend is not the point; fit and follow-through are.
  • Partner (external providers and integrators). Best for the specialist/architect layer and for accelerating early phases while internal capability is still thin.

The governing rule from the framework holds: build the broad middle, hire the deep core, partner for the spikes and the bridge.

Measuring the Roadmap

Track capability the way you track any program — against the baseline, on a cadence:

  • Coverage — share of each population that has reached its target literacy level (re-run the assessment).
  • Application — is trained capability showing up as changed work? Cross-reference adoption and workflow metrics; training that does not change behavior is waste.
  • Outcomes — does capability connect to value? Tie back to the measurement track.

Checklist

  • [ ] Baseline run and capability deltas pulled from role redesign (population × current × target × deadline matrix)
  • [ ] Phase 1 universal awareness program live and mandatory (satisfies EU AI Act Article 4)
  • [ ] Phase 2 fluency program built function-specific and hands-on, with prompting taught as general literacy
  • [ ] Builder track selecting and developing the multiplier layer; architect/specialist gaps closed by hire/partner
  • [ ] Champions network and community of practice stood up, with protected champion time
  • [ ] Build/buy/partner decided explicitly per layer
  • [ ] Coverage, application, and outcome metrics defined and on a cadence

Key Takeaways

  • Replace BYOAI with a roadmap. 75% use AI, only 39% are trained, and 78% bring their own tools — ungoverned by default (Microsoft & LinkedIn, 2024; Microsoft & LinkedIn, 2024; Microsoft & LinkedIn, 2024).
  • Sequence by layer: universal awareness → function-specific fluency → builders → champions, each with differentiated targets.
  • Fluency is domain-specific and where the value is; teach prompting as a general literacy, not a specialist role (TechRepublic, 2025).
  • Champions make peer diffusion deliberate — the realistic answer to a 75%/39% usage-training gap (Microsoft & LinkedIn, 2024; Microsoft & LinkedIn, 2024).
  • Build the broad middle, hire/partner the deep core. Scale spend (e.g. Amazon's $1.2B for 300,000) matters less than fit and follow-through (Amazon, 2021; PwC, 2025).
  • Measure coverage, application, and outcomes — training that does not change work is waste.

Sources

  • Microsoft & LinkedIn — 2024 Work Trend Index Annual Report, 2024. 75% of knowledge workers now use AI at work. View source · verified 2026-06-20 · primary
  • Microsoft & LinkedIn — 2024 Work Trend Index Annual Report, 2024. Only 39% of people who use AI at work have received AI training from their company; 66% of leaders say they would not hire someone without AI skills. View source · verified 2026-06-20 · primary
  • Microsoft & LinkedIn — 2024 Work Trend Index Annual Report, 2024. Only 39% of users have received AI training from their company and only 25% of companies expect to offer it this year. View source · verified 2026-06-20 · primary
  • Microsoft & LinkedIn — 2024 Work Trend Index Annual Report, 2024. 78% of AI users are bringing their own AI tools to work (BYOAI). View source · verified 2026-06-20 · primary
  • European Commission — EU AI Act (Regulation (EU) 2024/1689), Article 4 - AI literacy, 2024. Providers and deployers of AI systems shall take measures to ensure, to their best extent, a sufficient level of AI literacy of their staff and other persons dealing with the operation and use of AI systems on their behalf; per Article 113, Chapters I and II (which include Article 4) shall apply from 2 February 2025. View source · verified 2026-06-20 · primary
  • TechRepublic — Forget Prompt Engineering: Companies Are Now Hiring These AI Specialists, 2025. Prompt engineering is now basically obsolete; the career path, once predicted to be highly lucrative, has faded because generative AI can essentially prompt itself, and companies are hiring AI trainers, data specialists, and AI engineers instead. View source · verified 2026-06-20 · primary
  • Brynjolfsson, Li & Raymond — Generative AI at Work (NBER Working Paper 31161), 2023. Access to the tool increases productivity, measured by issues resolved per hour, by 14% on average, including a 34% improvement for novice and low-skilled workers but with minimal impact on experienced and highly skilled workers. View source · verified 2026-06-20 · primary
  • PwC — Global AI Jobs Barometer, 2025. Jobs requiring AI skills offer a wage premium in every industry analysed, with the average premium hitting 56%, up from 25% the previous year. View source · verified 2026-06-20 · primary
  • Amazon — Upskilling 2025, 2021. Starting in 2019 and through 2025 we are dedicating over $1.2 billion to provide 300,000 employees with access to free training programs. View source · verified 2026-06-20 · primary