From 70/20/10 to 90/10: Redesigning learning for the AI era

The conversation about AI in Learning & Development has reached an inflection point. After two years of rapid AI adoption we're seeing a fundamental shift in how learning happens at work.

Dr. Philippa Hardman recently articulated what many of us have been observing: a move from the traditional 70/20/10 model to what she calls 90/10. In this emerging framework, roughly 90% of learning happens through AI-powered performance support embedded directly in workflow, while 10% focuses on developing complex human capabilities through high-touch facilitation.

The model resonates because it names and operationalizes what's already happening in forward-thinking organizations. 

Understanding the 90/10 model

The 90/10 split reflects a fundamental redesign of how we enable performance at work:

-The 90%: AI-powered performance support in workflow
This is about re-coupling work and learning by embedding support directly where work happens. This is the approach we are taking with our AI coach, Pascal, who joins employees in Microsoft Teams, Slack, or  Zoom. AI sits in the flow, offering real-time guidance, retrieving information from knowledge bases, and providing nudges at decision points.

The goal is simple but ambitious: optimize individual performance, minimize variability, and accelerate time-to-output. Success isn't measured by completion rates but by first-time-right rates, reduced rework, and velocity of work.

-The 10%: High-touch development of irreducible skills

This is where L&D invests in intensive, human-led practice for capabilities that remain uniquely human: judgment under uncertainty, ethical reasoning, designing and supervising AI work, and navigating change. These skills improve with deliberate practice, feedback loops, and human coaching. You can't automate your way to better judgment. 

The capability triage no one is doing

Here's what's missing from most AI adoption strategies: a clear framework for deciding what humans should still do. L&D teams need to perform "capability triage" for every role. Dr Hardman describes it  as a decision tree:

• Delegate: Can AI execute this task independently with minimal risk? If it's rule-based, procedural, and errors are machine-catchable, automate it fully.
• Co-create (the 90%): Does human-AI collaboration materially improve both speed and quality? Can we add confidence measures and mandatory checkpoints? Then augment the task with AI support in workflow.
• Facilitate (the 10%): Does excellence require instant retrieval, nuanced judgment, or capabilities that must be internalized through practice? Then invest in high-touch human facilitation.

Most organizations treat AI adoption as a technology rollout rather than a redesign of work itself. They deploy tools, announce training, and hope for behavior changes. What gets skipped is this hard work of triage and determining what AI should handle, what humans do better with AI support, and what requires intensive human development.

Why play and product thinking unlock the 90%

At Pinnacle, we've emphasized that adoption requires play, not mandates. This matters even more in a 90/10 world where the bulk of learning happens in workflow.

Sandboxes, peer demos, and hackathons outperform slide decks because they build confidence through experiential learning. When employees can test AI on realistic tasks in controlled environments, hesitation fades quickly. People need structured space to experiment safely and share discoveries.

Gail Fierstein's recommendation to run HR like a product organization is critical here. Product teams focus on experience, work iteratively, and measure value rather than activity. When L&D embeds AI support in workflow, they're doing product design.

What this means for L&D teams

The transition to 90/10 requires L&D to rebuild their own capabilities:

For the 90%: Shift from course builder to product owner. Design how support shows up in different tasks and roles. Become workflow designers who determine where AI sits in tools people use. Track performance data, rework rates, first-time-right metrics, not completion rates.

For the 10%: Shift from curriculum planner to capability planner. Identify what belongs in the 10% using clear tests: instant retrieval requirements, transfer across contexts, judgment that improves with practice. Become practice architects who turn real work into learning experiences.

Just as 70/20/10 helped us move beyond classroom-only learning, 90/10 offers a framework for the AI era.