AI, the future of learning and π-shaped people
Only 30% of AI's value comes from the technology. The other 70% is the human side. Sana Labs on building "π-shaped" capability that's already forming inside your organisation.
You may have heard the headline story: AI is coming for the workforce, and entire departments will disappear. It's compelling. It's also, according to the evidence so far, not quite what's happening.
Research from Anthropic, the AI company behind Claude, tested this properly, measuring what AI is actually used for rather than what it could theoretically do, tracked against employment data since before ChatGPT's launch. The finding: no measurable rise in unemployment among the most AI-exposed workers.
The real AI story isn’t job loss, it's job change
AI's theoretical capability runs well ahead of actual deployment. Even in the most exposed category, computer science and mathematical roles, real-world usage covers only around a third of what AI could, in principle, already do - held back by legal constraints, integration costs, and the time organisations need to change how they work.
In addition, according to BCG, over the next two to three years, roughly half of US jobs are expected to be meaningfully reshaped by AI, but only 10–15% face real risk of elimination, over a five-year-plus horizon. The dominant story isn't job loss. It's job change.
What has job change meant for learning?
Traditionally, skills for a professional used to be generalist or specialist. As the nature of skills and expertise changes with AI, enter the age of I-shaped, T-shaped and π-shaped people, though the real story here isn’t how many things someone knows. It’s the judgement that connects them, and that judgement shows up at every level of an organisation. Here’s how I see different skillsets in the AI-age:
- I-shaped - deep expertise in one area, with little ability to judge how it connects to anything else.
- T-shaped - one deep area, plus enough general awareness to collaborate across teams.
- π-shaped - the modern upgrade: multiple areas of capability, joined by the judgement to bridge them, decide when to trust a system’s output, and know when a human needs to step in.
This judgement isn’t reserved for high-skill knowledge work; a frontline retail associate deciding when to override an AI-driven stock recommendation, or a call centre agent judging when a scripted response won’t land with a frustrated customer, is exercising the same kind of bridging judgement as a π-shaped consultant but just applied to a different kind of complexity. As earlier BCG research shows, it’s this judgement layer that AI struggles to replace.
The depth that forms each leg of the π in a π-shaped individual’s toolkit doesn’t have to come from a formal domain or credential. It can be operational fluency, procedural mastery or judgement on how to solve the same job’s problems in a different way - connecting dots the same way a T-shaped specialist connects one area of expertise to a second discipline.
(Side note: this is the kind of talent I love working with - and at Sana, we hire for exactly this: audacious, independent thinkers who work like π-shaped people.)
I'd argue that AI is creating π-shaped people by default: multiple domains of expertise, increasingly connected by a broader AI fluency that cuts across all of them. These are people who build bridges between functions that used to sit in separate silos. This behaviour is already happening, informally, everywhere.
You can see it in how work has shifted too. When asked about which aspect of work have been changed by AI, 47% of workers now say that their role has shifted towards managing and directing AI agents. Somewhere in your organisation right now, someone in account management is teaching themselves enough AI tooling to direct work that used to need a specialist. The LPI found 4 in 5 workers already want to learn more about AI to stay relevant. What's missing, though, is a way for that appetite to actually help organisational learning, rather than staying locked in one person's workflow.
How should organisations build AI skills at scale?
BCG estimates that only 30% of AI's value comes from the algorithms and systems themselves; 70% comes from how organisations manage the human side of the change.
So, how should organisations approach building AI skills and learning at scale? Here is what we have observed:
- Upskilling has to sit in the flow of work. The strongest performers embed learning into real tasks, with real tools and real feedback, rather than treating training as a separate scheduled event. Increasingly, this means people can learn on demand - searching for an answer in the moment they need it, treating their learning environment as something to draw on mid-task, not a library visited separately.
- People need a clear development path. Without a visible route from "role being automated" to "role with a future," anxiety fills the gap left by ambiguity. That means skills development needs to be tracked, with visibility into who has which skills and where the gaps sit. It matters: 72% of people say AI has already changed what skills they're expected to have. Visibility only changes behaviour, though, when it's tied to something people actually care about - progression, opportunity, achievement - not a course completion.
- Use AI to preserve institutional knowledge. Capture how people are already using AI in their real workflows, and fold it into the organisation's collective knowledge, rather than letting it evaporate the moment someone changes role. When someone in finance teaches herself enough of your CRM's AI layer to rethink how proposals get built, that's a second leg of a π forming. Naming it and crediting her turns her individual workaround into organisational capability.
- Learning has to be personalised, not generic. Two people in the same role can have entirely different skill gaps, yet most learning content is still a static set of courses everyone works through in the same order. The organisations pulling ahead favour adaptive journeys, often supported by AI-guided tutoring, that adjust to what someone already knows and where they're actually stuck. A generic curriculum can't keep pace with a workforce where everyone's starting point and destination look different.
The answer is to stay people-centric
The AI wave was always going to come. What matters is who is ready for it.
AI's theoretical capability is already miles ahead of where most organisations are using it. The constraint isn't the technology - it's organisational. And in my view, the answer is to always stay relentlessly people-centred.
The organisations quietly building π-shaped capability - one workaround, one workflow, one adaptive learning path at a time - aren't doing anything exotic. They're simply noticing capability that's already forming inside their own walls, and choosing to name it and grow it.
Being people-centric means bringing people into shaping the change early, so they can see themselves in it, rather than announcing a new tool and hoping adoption follows. It means treating every informal workaround as a signal: evidence that a second leg of expertise is already forming.
The technology will keep advancing regardless of what any of us choose. Whether it produces a workforce of π-shaped people comes down to a much more human decision: whether organisations build around their people, or simply bolt AI onto the systems and habits already in place.
By Kyla Tan and Emily Ong from Sana Labs
Find out more at sanalabs.com