With former Google CEO Eric Schmidt being booed for mentioning AI last week, during his University of Arizona commencement speech, I’m becoming a little hesitant to talk about AI in anything I write. It seems to have gone from being a topic of curiosity to being something that massively divides opinion.
I’m well aware of AI’s downsides. From intrusive data centres using acres of greenfield land, to the massive amounts of power and water they gobble up, to its misuse by authoritarian governments and public authorities. All deeply concerning. Yet from a personal point of view, I find it hard not to be fascinated by AI’s potential.
It’s no secret that I’m quietly working away on the development of a software application for L&D. And while most of what I’ve been doing over the last 12 months has focused on the frameworks, principles and thinking that would underpin said application, I am now at the beginning of the design and prototyping stages; and I’ve been using the AI assistant in a tool called Figma, to help get that process started.
And frankly, what I’ve been able to achieve using that assistant, would have been impossible had I just been trying to figure stuff out in Figma all by myself.
In case you are wondering, I mention all this more for background context than anything else. Just to reassure you, AI is definitely not the focus of today’s piece.
However, what I do want to write about today is deeply connected to the design and prototyping I’m doing. It’s about the conceptual thinking behind the design and prototyping, which also connects back to one of my previous Learning Re-Framed pieces.
The binary we default to
A couple of weeks back I was exploring the question of how people with different domain expertise might go about solving workplace performance problems. Writing the piece made me realise that our response to the question is often a binary one.
It frequently leads to people decide one of two things. Either there needs to be some kind of training solution put in place or there need to be changes to or refinements of the workplace process.
The latter might include looking at the:
process stages or related procedural steps
the tools used to carry out the process, or
the environment in which the process happens.
In this mode of thinking, you implement one or other of those solutions and your performance problem goes away.
However, if you’ve read the piece already, you’ll know that the more I looked at this, the more it seemed to me that the typical responses to that, ‘How do we improve performance?’ question might not be the whole story.
Possible solutions are likely a bit more nuanced than that. It could, for example be a combination of training and process refinement that’s required. Or (and this where I think things get interesting) it could be that there’s a ‘somewhere in-between’ solution that often gets missed.
I started to explore that middle ground in the previous article. But this week I’d like to dig a bit deeper and explain what I think it is all about.
Most of you reading this are more likely than not to have some kind of involvement in L&D. So, I wanted to start with the more familiar part of that binary response – the training solution.
The limits of preparation
In terms of workplace performance, this is what you might call prospective or preparatory learning. In other words, preparing people as best we can for future performance. And if done well, it can be highly effective.
But it does have its down sides. By its very nature, it typically has to be somewhat compressed and simplified. Unless you are teaching something that has a very narrow scope, what you will be able to cover during training has to be representative of workplace reality. It would be impossible to cover every single aspect of that reality.
Which means that our learners return to the workplace well-prepared for certain situations and eventualities but completely under-prepared for others.
If we are in the happy position of being able to do something about that (and, let’s remember, many people in L&D are not) then I think this is where our response might get a bit misdirected.
I think a typical (and perfectly logical) L&D response would be, ‘Let’s provide more training’. After all, the thinking would go, there are all these other situations we could teach the learners about that we didn’t have time to cover initially. So, let’s dig in and cover more in follow-up sessions.
And, of course, doing that would be way better than doing nothing. It definitely wouldn’t be a waste and it would almost certainly improve performance over time and help demonstrate the value of the additional training.
But I think it probably misdiagnoses what learners (who, by the way, now find themselves transformed into performers) might benefit from most.
When learning changes shape
And if you are feeling a little puzzled at this point, I totally get it. You may already be thinking, ‘But hang on a minute, Andrew, these guys are going to continue learning on the job anyway, aren’t they?
To which I would respond, ‘Absolutely’.
But I think the bit that is easy to miss is that the nature of learning as a performer is different. It’s no longer prospective, representative learning that is happening during workplace performance, it’s what we might call retrospective learning. In other words, learning that is embedded in actual instances of performance that have already happened or may be about to happen.
And that kind of embedded learning is inevitable, whether there is anything formal in place to help it along. And where it isn’t supported, it will often be uneven, unpredictable and messy.
While more prospective or preparatory learning, is not unhelpful, it is almost certainly the least useful kind of help we could provide at this point.
Supporting the inevitable
If retrospective, embedded learning is inevitable, then the question switches from, ‘What additional preparatory training can we provide?’ to, ‘How do we best support the inevitable?’
In other words, how can we best shape and support the embedded learning that kicks in once workplace performance is happening? How can we best reduce friction, unevenness, and unpredictability and make embedded learning more structured, accelerated and accurate?
And that, I think, is where things get interesting. Because if prospective learning is where we design what people need to know before they act, retrospective, embedded learning is where we shape what people learn while and after they act.
I’d like to finish by focusing on three examples from the framework I mentioned earlier, which, I hope, help to illustrate the idea of shaping what people learn while and after they act.
Testing work against reality
The first relates to something I’ve been thinking about in the aforementioned framework, connected to reviewing and evaluating completed work.
Take a sales proposal, for example. Imagine a relatively inexperienced account manager preparing a proposal for a major client. They’ve completed the training on how to write proposals.
But now they’re sitting in front of a real piece of work that actually matters. Without any support, what often happens at this point is surprisingly superficial. A quick proofread. A vague sense-check. Maybe a moment of uncertainty followed by, ‘Well… I think it’s okay.’
The embedded learning opportunity is there, but it’s weak and inconsistent.
However, imagine something slightly different. Before submission, the performer is guided through a small set of structured prompts connected to the realities of successful proposals:
Have you clearly justified the proposed solution from the client’s perspective?
Does the proposal directly address the commercial risks the client raised earlier?
If challenged, could you explain why you prioritised these recommendations over alternatives?
Notice what’s happening here. The support isn’t ‘teaching evaluation’ in the abstract. It’s helping the performer learn from a real output by testing it against reality.
Without support, that learning is often vague, biased or skipped entirely. With support, it becomes more deliberate and structured. The performer starts gradually building better judgement through repeated exposure to real work and real evaluation criteria.
Learning through variation
A second example sits slightly differently. One of the easiest traps for performers to fall into is overgeneralising from limited experience. Something works once or twice and quickly becomes interpreted as ‘the way this works.’ But real-world situations rarely repeat themselves neatly.
Take a customer service environment. Imagine someone learning to handle difficult complaint calls. Early in their experience, they may discover that reassurance works well with frustrated customers. But eventually they encounter a customer who becomes more irritated precisely because the response feels too scripted or too passive. Suddenly the learner hits a variation they weren’t prepared for.
Again, more generic training might help a bit. But something else may be more useful at this point. Imagine being able to compare several short examples of similar complaint situations side-by-side:
one where reassurance calms the customer
one where speed and decisiveness matter more, and
another where escalation is the right response almost immediately
But importantly, the performer isn’t just ‘seeing more examples.’ They are guided through what changes and what stays the same across real situations.
Without support, people often build narrow mental models from limited experience. With support, they start developing more deliberate pattern recognition. They begin to understand not just procedures, but variation. And, I suspect, that distinction matters enormously in judgement-heavy work.
The fragility of judgement
The third example is perhaps the most interesting of all because it deals with something surprisingly fragile: our ability to judge our own performance accurately.
Most of us are probably familiar with situations where someone becomes overconfident after a few early successes. Equally, we’ve all seen capable people lose confidence completely after one or two difficult experiences. In both cases, the issue isn’t really competence. It’s calibration.
Take a first-time manager running difficult one-to-one conversations with team members. After one meeting goes badly, they may conclude: ‘I’m terrible at these conversations.’
Another manager might have the opposite reaction after one conversation goes well: ‘I’ve got this sorted.’ Both interpretations may be inaccurate.
Now imagine if, after key conversations, the manager was guided through a very lightweight review process:
Which parts of the conversation appeared to move things forward?
Where did the employee become more engaged or more withdrawn?
What evidence suggests the outcome was positive rather than simply comfortable?
What would you adjust next time?
Again, this isn’t really ‘reflection’ in the loose sense of the word. It’s helping performers test whether their internal judgement matches external reality.
Without support, people often drift into either overconfidence or self-doubt. With support, they gradually correct and refine their internal models.
The ‘in-between’ space
It’s worth noting that none of these examples are really about delivering more training content. Nor are they about redesigning the workplace process itself.
They sit somewhere in between. They are ways of shaping the quality of embedded learning that emerges, while people are actually performing.
And the more I explore this territory, the more I suspect that this ‘in-between space’ could become increasingly important for L&D – especially in a world where nuanced, judgement heavy, work (the kind that can’t easily be replaced by AI) becomes ever more significant.
The world of work where a huge amount of capability development happens after the course ends — in the messy, variable, imperfect reality of the work itself.
Until next time,
Andrew
PS One of the things I’ve increasingly noticed while exploring topics like workflow learning, judgement, transfer and performance support is how interconnected they all are.
Much of the thinking behind this September’s public Learning Re-Framed programme is really about helping L&D teams think more carefully about what happens after formal learning ends — when people are back in the messy, variable reality of the work itself.
If you’re curious, you can find more about the programme here.