AI has removed L&D’s safety net. For years, we could prove our value by creating courses, workshops and content that helped people access knowledge. But when AI can deliver information faster, the question changes. If performance still varies when answers are everywhere, maybe the real issue was never access to knowledge. Maybe it was how the work was designed.

The threat isn’t that AI makes L&D irrelevant. The opportunity is that AI makes performance problems impossible to ignore. When everyone is experimenting, duplicating effort and getting wildly different results, the issue isn’t motivation or effort. It’s that the work hasn’t been defined clearly enough for people — or AI — to do it consistently.

The B-17 accidents weren’t caused by careless pilots. They were caused by cockpit controls that made the wrong action too easy. Alphonse Chapanis didn’t solve the problem with more training. He redesigned the environment. That distinction matters for L&D, because performance rarely improves by asking people to try harder inside flawed systems. Skills matter, but the environment comes first.

If performance depends on people remembering, interpreting and improvising, results will always vary. That’s why skills alone aren’t enough — especially when AI needs clear instructions, context and guardrails to perform well. Before we train people to “think critically”, we need to define what good looks like and bake that guidance into the work itself. Environment first. Skills second.

Learning matters, but learning and performance are not the same thing. Learning is the means; performance is the outcome. And if AI is exposing the limits of knowledge transfer and capability building, L&D has a choice: keep defining itself around helping people learn, or step into a broader role focused on designing the conditions that help people perform.

The shift from learning professional to performance partner doesn’t start with a new job title, team name or org chart. It starts internally — with the decision to see learning as one tool inside a broader performance toolkit. And once you see work design as the starting point, not the afterthought, it’s hard to unsee it.

Sometimes the problem isn’t missing knowledge, weak skills or badly organised content. It’s the absence of a clear, repeatable way to do the work. In this example, new salespeople didn’t need more product training — they needed guided workflows that helped them prepare, run and follow up on demos consistently, using the wisdom already trapped in experienced people’s heads.

Even when L&D spots a clear opportunity to improve performance, the system around us often pulls us back towards visible, scalable, people-positive solutions — programmes, platforms and training libraries. That isn’t because practitioners lack ambition; it’s because many organisations incentivise employee experience over performance. And until that changes, we’ll need to keep pushing the conversation from learning activity towards measurable business impact.

Designing work sounds simple until you hit the reality of complex organisations: disconnected tools, fragmented ownership, undocumented workflows and best practice trapped in people’s heads. An AI prompt library is a perfect example — useful in theory, optional in practice. If prompts, guidance and standards aren’t embedded into the flow of work, performance still depends on memory, motivation and heroic effort.

Before I ever thought of work design as an L&D problem, I was trying to build a satnav for my own work — a system that showed me what to do next, surfaced the right context, and stopped everything disappearing into a hot mess of tags, reminders and good intentions. Looking back, those productivity experiments were my first attempt at designing an environment that improved performance. I just didn’t know it yet.

Notion was the first tool I’d found that let me design the system around the work, rather than squeeze the work into someone else’s system. By creating my own task database, properties and filtered views, I could finally surface what I needed, when I needed it — the elusive satnav for work. And that’s when the organisational dots started connecting.

Notion templates changed the dashboard from a task list into something more powerful: guidance embedded directly into the work. Instead of resources dying in SharePoint or hiding inside eLearning modules, checklists, examples, links and instructions could appear at the exact moment they were needed. For the first time, “in the flow of work” felt like more than a slogan.

Once databases could talk to each other, the system became more than a way to manage tasks. It became a way to capture what people learned while doing the work, then surface those insights for the next person at the exact moment they needed them. The wisdom usually trapped in heads, retros and forgotten documents could finally become part of the workflow itself.

Some workflows aren’t triggered by a request, a problem or an external event. They’re triggered by time. A quarterly tax return is a simple example: infrequent enough to forget, important enough to matter, repetitive enough to systemise. Instead of figuring it out from scratch every three months, the workflow appears when it’s needed — with the links, prompts and guidance already built in.

Work degrades unless it’s maintained. Email lists go stale, standards drift, useful habits disappear, and eventually someone pays the price. Recurring workflows give us a way to prevent that decay by surfacing the right checks, prompts and reflections on the right cadence. Not as admin for admin’s sake, but as a self-maintaining system that keeps performance from falling apart.

Cyclical workflows aren’t just useful for housekeeping. Used well, they create space for the conversations and reflections that usually get swallowed by urgency. Goals, risks, expectations, neglected priorities, fragile processes — these things rarely announce themselves. They need to be surfaced deliberately. The difference between a reactive organisation and a proactive one isn’t effort. It’s whether the system repeatedly creates room for the right questions.

Once a workflow is defined, it becomes easier to repeat, improve and delegate. That used to mean handing work to another person. Increasingly, it will mean handing parts of the work to AI. But AI agents won’t magically infer the right steps, context, judgement and guardrails. If businesses want AI to perform reliably, the work has to be defined first.

Designing work for humans and AI requires the same basic ingredients: context, instructions, guardrails, blueprints and templates. That matters because it means performance improvement expertise is becoming more valuable, not less. If we can define what someone needs to complete a task reliably, we’re also defining what AI needs to do that work well.

Not every task deserves the same level of design. Some have far more impact on the outcome than others, which means we need to triage where our effort goes. And once workflows are documented, outcomes are defined and results are measured, tasks become levers we can adjust deliberately. That’s the shift from work design to workflow engineering.

Workflows don’t just help people perform the work. They help organisations understand how the work was performed. Instead of judging performance only by outcomes or managerial opinion, we can see the steps taken, decisions made and outputs produced. That gives us a fairer, more objective basis for evaluation — and a clearer way to improve the system when results fall short.