From 70/20/10 to 90/10
A new L&D operating system for the AI Era?
Hey folks 👋
This week I want to share a hypothesis I’m increasingly convinced of: that we are entering an age of the 90/10 model of L&D.
90/10 is a model where roughly 90% of “training” is delivered by AI coaches as daily performance support, and 10% of training is dedicated to developing complex and critical skills via high-touch, human-led learning experiences.
Proponents of 90/10 argue that the model isn’t about learning less, but about learning smarter by defining all jobs to be done as one of the following:
Delegate (the dead skills): Tasks that can be offloaded to AI.
Co-Create (the 90%): Tasks which well-defined AI agents can augment and help humans to perform optimally.
Facilitate (the 10%): Tasks which require high-touch, human-led learning to develop.
Within this model, the [new] role of L&D is to lead this “capability triage” and define for each role what is delegated to AI, what is co-created with AI (the 90%) and what is facilitated and enabled by high-touch human support (the 10%). After that, the task of L&D is to design and maintain complex hybrid systems which enable strategic task delegation, task co-creation and deep and meaningful human facilitation.
In this week’s post I’ll walk through the 90/10 model in detail, discuss why it’s already emerging and explore what this will likely mean in practice for L&D teams, roles and key skills.
Let’s dive in!
The Rise of the 90/10 Model
If an AI-powered workplace and AI-powered L&D sounds like hot air and hype, the data suggests otherwise.
Just this week, tens of thousands of jobs have been cut amid what employers like Accenture are calling, “a major AI re-skilling drive”. Meanwhile, Microsoft’s 2024 Work Trend Index and McKinsey’s 2024 State of AI report showed that generative AI adoption in the workplace has more than doubled year on year for the last two years. More recently, Stanford’s 2025 AI Index found that 78% of organisations deployed AI at scale in 2024, up from 55% the year before.
Meanwhile, just last week amid a growing discussion about the use of AI coaches in the workplace, a report by BCG and Harvard Business Review found that an AI coach deployed in the flow of work can deliver positive skills outcomes roughly 23% faster than classroom methods, with impact rates rising to 32% for employees starting from a lower baseline.
Taken together, the data suggests that AI is reshaping not just how work gets done, but also how we enable people to do their work effectively.
So if AI at work is now both real and material, the natural question for L&D is: how do we design for it? The short answer is to stop treating learning as an event and start treating it as a system.
In a 90/10 world that system has two gears: everyday performance support where AI sits inside the flow of work, and high-touch human practice for the so-called irreducibles — i.e. complex and nuanced skills and mission-critical strategic decision making.
Let’s explore what this looks like in practice.
The 90%: Performance Support in the Flow of Work
The bulk of the emerging L&D model—the 90%—is about re-coupling work and learning through AI-powered performance support. In practice, this means embedding support and “productive friction” within the workflow itself rather than locating it classrooms or LMSs.
Exactly how this plays out is to TBD, but on the ground at the “bleeding edges” of L&D experimentation I already see a commitment to reducing investment in online courses and in person workshops, in favour of AI “copilots” integrated directly where work happens.
In Teams/Slack channels, docs and CRMs, AI is on hand to help employees to draft artefacts, consider alternative approaches, weigh-up decisions and retrieve information from the organisation’s knowledge base using retrieval-augmented generation (RAG).
The goal here is both simple and ambitious: optimise individual performance, minimise performance variability and accelerate time-to-output.
For the L&D team, this part of the 90/10 model requires a number of fundamental shifts in roles, responsibilities and skillsets, including:
Course builder → product owner: the L&D team decides how support and “productive friction” show up in different tasks within different roles.
Content producer → workflow designer: the L&D decide where support shows up in the tools people use, what it looks like and how we judge quality.
LMS admin → data analyst: the L&D team tracks who’s working with AI and track data such as first-time-right rates, reworks and overrides to inform system design.
In short: L&D’s attention shifts from creating content about the work to designing the work itself. Success no longer looks like courses produced and completion rates but the extent to which learning happens drives both the velocity and quality of each employee’s work.
The 10%: Mastering the Irreducibles
If the 90% re-couples work and learning, the remaining 10% delivers a new set of deeply complex human capabilities required by the age of AI.
In practice, this will likely mean investing in in-person hours to go deep on strategy and practice and hone skills such as:
Exercising judgment with uncertainty, e.g. trade-off analysis and ethical reasoning.
Designing & supervising AI work, e.g. defining what to delegate, co-create and facilitate, prompt↔task design.
Change navigation & adoption, e.g. shaping norms for safe AI use, and defining and agreeing what it means to introduce “productive friction”.
The goal here is equally simple and ambitious as in the 90%: build AI-enhanced systems of work, ring-fence human-critical skills and ensure reliable and ethical AI supervision in the workplace.
This part of the 90/10 model also requires a number of fundamental shifts in roles, responsibilities and skillsets for L&D team members, for example:
Curriculum planner → capability planner: the L&D team identify what belongs in the 10% using clear tests (instant retrieval, transfer across contexts, irreducible judgment, improves with practice). They also map role-specific irreducibles and write T-shaped profiles which each role must internalise.
Content creator → practice architect: the &D team turn real artefacts (contracts, tickets, PRDs, incidents) into deliberate practice exercises and drills. They also define rubrics for quality and design tight feedback loops that sharpen judgment and AI supervision.
In short: L&D’s attention shifts to engineering the conditions in which to practise and develop complex skills, with the goal of ensuring that the 10% compounds the value of the 90%.
The 90/10 L&D Decision Tree: Delegate, Co-create or Facilitate?
In a 90/10 world, the focus of L&D teams shifts from building content about the job to designing how the work is executed. In practice, L&D’s responsibilities within the 90/10 model lie in “capability triage”, i.e. determining for every role:
What gets delegated to AI > “dead skills”
What gets co-created with support from AI > the 90%
What must be human-facilitated in order to be mastered > the 10%
Here’s how that framework might play out in practice for L&D teams:
Stage 1: Delegate — which tasks can AI execute independently?
For every task within every role, L&D asks four questions: Is the task rule-based, procedural and structured, with clear rules? Does it require minimal contextual judgment? Are errors machine-catchable? Are the risks of error low?
If you answer yes to all four, automate the task.
L&D design the automation and error handling, communicate scope, and move on.Stage 2: Co-Create - which tasks benefits from AI support but require human judgement & steer?
L&D ask: Does human + AI collaboration materially raise both the speed and quality of this task? Can we add confidence/provenance measures to response and mandatory “stop and check” stop-points?
If you answered yes to both, augment the task (the 90%).
L&D design the human-AI interaction, set up mandatory stop-points where a person must review/approve before the workflow continues, define the cues that tell humans when to step in and what to look for.Stage 3: Facilitate - what tasks are mastered only through human-facilitated practice?
Does a task require instant retrieval which AI is not capable of? Does the task require nuance of judgment? Does excellence depend on instant retrieval humans must internalise?
If you hit two or more, facilitate the task (the 10%).
L&D design practice sessions, Case Labs, feedback plans and performance rubrics. Success is measured by the rate of transfer (workplace performance), not just course completion (LMS or attendance stats).
Conclusion: Designing Work, Not Courses
Just as the 70–20–10 model helped us move beyond classrooms in the LMS era, the 90/10 model offers a new way of designing and delivering learning for the AI era. 70–20–10 stuck because it named and operationalised what people already felt. Signals suggest that a similar process is underway with a shift towards 90/10.
More and more organisations are exploring and will likely adopt 90/10 because it operationalises at scale a new reality that technolgy has already started to create — a world in which:
Repeatable work can be delegated to AI
Situational work is both more efficient & effective when co-created with AI
Complex work is best practised and internalised through human-to-human interaction.
The shift to 90/10 isn’t inevitable because it won’t be easy. Change will demand new mindsets, new governance models and real re-skilling—first and foremost for L&D teams themselves. As L&D professionals, we will need to learn how to design hybrid workflows, how to prompt and validate AI responsibly and how best to design high-value, high-touch practice environments for a new set of deeply complex skills.
If we get this right, L&D teams become more—not less—business-critical. Within the 90/10 model, the value proposition of L&D shifts from “we design & deliver learning content” to “we design & deliver optimal performance at scale.” The critical task ahead is not simply implementing AI, but cultivating the human wisdom to manage it—a task that sits squarely with L&D.
Happy innovating!
Phil 👋
PS: Want to explore how AI is reshaping L&D with me and a group of like-minded L&D professionals? Apply for a place on my AI & Learning Design Bootcamp where we dig into these concepts and get hands-on to develop key, future-ready skills.