The people side of AI implementation

Went to a great event last week on AI in L&D - lots of interesting perspectives and opinions in the room that really made me think about a few things.

One speaker made the point that in the past, the biggest barriers to change and growth were things like technology or investment. Now, it's people. Technology is evolving so quickly that it no longer slows progress down - it's the humans who are slower to adapt, apply and take it up who are the barrier. That resonated with me, because at Teamshaper we've focused heavily on the human side of AI implementation in recent months.

We're still only touching the tip of the iceberg when it comes to what AI can do to support public policy, planning and services, and it's great to see so many AI implementation roles and teams being created across the public sector in the past few months.

Our role in L&D is to look at the people side of this change. Here's what we're seeing right now:

1. Uneven uptake - some people are storming ahead while others are lagging behind. A substantial number still use LLMs almost entirely as a Grammarly replacement, and some don't even do that. Someone at the event mentioned seeing individuals "hogging" knowledge - a result of this uneven uptake, and a natural split between early adopters and those still holding onto the Nokia brick phone while everyone else has moved to smartphones. Part of it comes down to hybrid working and the disjointed way many teams operate now. Leaving people behind is in nobody's interest.

2. Over-reliance by some - there are people so impressed by what LLMs can do that they've already become over-reliant on them professionally, particularly in the past few months. Someone at the event pointed out that, over time, this risks people losing skill as they hand over key elements of their role to their LLM entirely. The more immediate problem we're seeing is that some of that delegation is done badly, without the right safeguards or quality checks, resulting in below-par outcomes: unintelligible copy, inaccuracies or mistakes. Most LLMs are intuitive to use, but learning to use them well is a separate skill worth having.

3. Over-scrutiny by others - just as some are blasé about trusting their LLM, others swing the other way: checking and re-checking every output until the time they saved evaporates, and the whole task ends up taking longer than if they'd just done it themselves. Knowing what to trust and when to scrutinise is crucial to using LLMs efficiently. Learning to use them well is what produces the biggest saving of all - it cuts down on rework and, over time, builds confidence that holds up, because it's based on doing things right.

4. Groundhog Day usage - plenty of users are still using their LLM ad hoc, without building things like agents, skills or even project memory. Using an LLM this way is the difference between hiring unskilled temporary staff for a single day and employing, training and developing a permanent team member you nurture and grow over time: any learning is lost, and every project starts from scratch. Teaching your LLM, and building a few skills and agents isn't difficult, and it makes a real difference. It often takes very little time to set up things like your style and preferences, information about your role and its purpose, or who your colleagues are and what they do. It's also fairly easy to build team, departmental or organisation-wide knowledge banks that support meeting tone, brand guidelines, policies and regulations, or learning from past projects.

The key to all this is accepting that AI implementation covers everything from basic technical training to a full rethink of processes and services. But the role of L&D in it is crucial, and it comes down to what L&D has always been about: building confidence in people.

Our Writing with Copilot programme is here.