On August 20, 2026, Training Journal editor Jo Cook argued that AI is making individuals faster while organisations fail to convert those gains into business outcomes. That AI productivity gap is now L&D’s problem to solve as much as IT’s.
What Training Journal says about the AI productivity gap
Cook’s piece on Training Journal sets a blunt expectation: AI-created time savings won’t show up on the balance sheet until teams change workflows, targets and accountability. The argument tracks what many learning leaders see—people get quicker at emails, documents and analysis with AI assistants, yet cycle times, quality and customer outcomes barely move. Training Journal’s point is clear: treating AI like a personal timesaver, without redesigning work, leaves value on the table.
The publication’s focus on organisational performance sets a useful bar for L&D. If programs stop at tool tips and prompt craft, they stall. If they connect skills to process change, measurement and risk controls, they move the needle. That’s the line between a skills catalog and a performance system—and it’s where the AI productivity gap either widens or closes.
Why L&D teams aren’t yet seeing the lift
Three friction points show up again and again in enterprise rollouts:
- Workflows weren’t redesigned. People bolt AI onto legacy steps, so cycle time improves in one task but waits and handoffs erase the gain.
- Targets didn’t change. Managers still grade on output volume, not first-pass quality or customer resolution, so behavior reverts.
- Risk fears freeze progress. Without clear guardrails, teams avoid using AI on high-value work, keeping it parked on low-stakes tasks.
Each friction point sits in L&D’s remit as much as operations. It means courses alone won’t cut it. Programs have to package skills with a defined way of working, a measurement plan, and a safe-to-try boundary. That’s the work of enablement, not just education—and it aligns with the NIST AI Risk Management Framework, which pairs capability building with governance and monitoring.
Closing the AI performance gap: a 90‑day playbook
If the goal is to turn AI skills into team performance before 2027 budgeting locks in, L&D can lead a three-track push that mirrors Training Journal’s call to action.
- Redesign two workflows, end to end. Pick one revenue-facing (e.g., proposal creation) and one operations-facing (e.g., incident triage). Map current steps, then rewrite the flow with AI in the loop. Remove or combine steps made redundant by generation, summarisation or retrieval. Document who does what, which model is used, and where a human must sign off.
- Instrument outcomes, not just usage. Set a minimal telemetry plan tied to business goals: cycle time from request to delivery, first-pass quality, rework rate, backlog burn-up, and customer or stakeholder satisfaction. Usage minutes and prompt counts are fine, but they don’t prove value. Publish a weekly dashboard for the two pilots.
- Set crisp guardrails. Define allowed data, red lines, and approval points. Keep it on one page per workflow. Borrow language from your legal team and from regulators like the UK’s ICO guidance on monitoring workers to avoid accidental surveillance when tracking outcomes.
Package those three elements—the new way of working, the metrics, and the guardrails—into a “workflow playbook” that ships with short practice tasks and office hours. This is the missing middle between tool training and transformation, and it’s where the AI productivity gap starts to close.
What to train, and what to change in management
Skills still matter, but they’re table stakes. Focus training on four capabilities that make the redesigned workflows stick:
- Problem framing and decomposition. People need to break work into AI-able chunks without losing context.
- Source hygiene and citation. Teach how to ground outputs in verifiable sources and record them for audit.
- Review heuristics. Give teams checklists for catching common AI errors tied to their domain, not generic warnings.
- Exception handling. When the model stalls or hallucinates, what’s the fallback? Make the path obvious.
Then change how managers measure. Swap activity measures for outcome measures in weekly check-ins on the two pilot workflows. A short manager kit—new questions to ask, how to read the dashboard, how to approve exceptions—does more than a town hall. This is where “sponsorship” turns into operating rhythm. It aligns with Training Journal’s insistence that productivity wins are meaningless without organisational follow-through.
Metrics L&D can ship now
You don’t need a data lake to prove value. Start with a lightweight set that teams can collect in spreadsheets or existing tools:
- Cycle time: hours from intake to handover for the entire workflow.
- First-pass quality: share of outputs accepted without rework by a defined reviewer.
- Rework rate: average number of revisions per item before acceptance.
- Backlog burn-up: items completed per week in the pilot lanes versus baseline.
- Customer or stakeholder satisfaction: one-question pulse after delivery.
Track adoption too, but separate it from impact. A shared scorecard prevents the common trap Cook flags in Training Journal: celebrating tool uptake while business results stay flat. If your HR team is formalising human capital reporting, consider mapping these to the spirit of CIPD guidance on generative AI at work so metrics survive audit and board scrutiny.
Risk and ethics aren’t blockers—they’re design inputs
Concerns over data leakage, bias and over‑monitoring often freeze pilots. Treat those concerns as design requirements. For each workflow playbook, list prohibited data sources, bias checks relevant to the domain, and the minimum review needed for sensitive outputs. Assign a named process owner. This approach mirrors elements in the NIST AI RMF and keeps ethical issues visible without halting progress.
Training Journal’s August 20 article is right to set a higher bar: without changes to work design, measurement and management, the AI productivity gap widens even as individual skills improve. The fix is close at hand. Pick two workflows, ship playbooks, and publish outcomes weekly. Do that, and by January 2027 you won’t be arguing about pilots—you’ll be budgeting for scale, with evidence that the AI productivity gap is closing where it matters most. For more on this, see bloomberg.com and nytimes.com.