From AI Pilots to Everyday Practice: What I’m Learning About Embedding Agents at Scale

An AI version of me looking more youthful!

There’s no shortage of commentary right now about AI and leadership. Everyone has a framework, a prediction, or a hot take.

But here’s what I’m noticing from inside the work itself:
the conversation changes completely once AI stops being a pilot and starts becoming infrastructure.

At work, we now have over 3,000 agents in the business.

That sounds impressive.
It is impressive.

And it’s also the point at which things get far more complex.

Because building agents is no longer the hard part.
Embedding them into how people actually work is.

I’m currently building what we’ve started to call an Agent Factory – enterprise-grade agents, built at scale, and designed to be embedded into day-to-day workflows rather than sitting on the sidelines as “nice to have” tools.

And for those of you new to agents…in simple terms an agent is a ‘specialist’ that you have built with specific instructions and knowledge sources that can perform a task, mine information, or automate a workflow.

I want to share what’s genuinely working well so far, what’s proving much harder, and what I’m still figuring out in real time.

This isn’t a finished story. It’s very much a journey.

What’s Working Well

1. Treating agents like mini humans, not tools

One of the biggest shifts we’ve made is changing how we think about agents.

We don’t see them as bits of tech anymore.
We see them more like mini employees.

That means:

• They have a “hiring date”

• They need to be registered and reviewed

• Their usage and performance is tracked

• They have a lifecycle, not a one-off build moment

Just like people, agents need ongoing care and attention. The maintenance matters as much as the creation.

This mindset shift alone I hope will move the organisation away from “build and forget” behaviour and towards something much more sustainable.

2. Building agents systematically, not creatively every time

When lots of people start building agents, creativity is wonderful… until it creates chaos.

We’ve learned quickly that scale demands de-duplication and consistency.

So w’re putting a set of guiding principles in place that all agents should adhere to:

• Only build if it doesn’t exist already

• Common design standards

• Shared assumptions about tone, scope, and safety

• Clarity on what an agent should and should not do

This means we can hopefully have many builders across the organisation, without ending up with hundreds of wildly different experiences for users.

Freedom within a framework will be key.

3. Getting much closer to the business

This has been one of the most powerful changes.

Rather than centralising all agent building in one team, we’re placing agent builders directly into business units.

Why?
Because proximity matters.

When builders understand the real day job – the pressures, the workarounds, the frustrations – they build better agents.
Agents that genuinely relieve pain points rather than just demonstrating capability.

The closer the builder is to the work, the more likely the agent is to stick. That’s the idea. Let’s see if it works!

What’s Harder (and Still Unproven)

This is the part I’m spending most of my time on now.

1. The embedding problem

To date, using agents has largely been optional.

That’s fine at the start.
But optional tools rarely become embedded tools.

If we want agents to genuinely speed up how we work, then a shift is required:

• from optional

• to expected

• and in some cases, replacing core parts of the workflow

That’s a much bigger leadership ask.

2. Carrot and stick (not one or the other)

I’m increasingly convinced that embedding agents at scale requires both encouragement and accountability.

The carrot:

• praise and recognition for teams who actively embrace agents

• visibility of good practice

• celebrating time saved and friction removed

The stick:

• tracking adoption and usage

• understanding where agents are available but not used

• more active nudging (and cajoling) where resistance shows up

This isn’t about punishment.
It’s about being honest that behaviour change doesn’t happen just because something is available.

3. Drumbeat, dashboards and senior leadership air cover

Embedding won’t happen quietly in the background.

It needs:

• A drumbeat of communication

• Leader boards and dashboards to make progress visible

• Senior leaders actively endorsing – and in some cases mandating – the change

People take their cues from what leaders consistently talk about, not what they casually mention once.

4. Resistance (and naming it honestly)

Of course there is resistance.

Some people dislike change.
Some feel threatened by what AI represents.
Some simply don’t want to learn another new way of working.

Pretending this doesn’t exist doesn’t help.

What does help is being crystal clear on the message:

This is human plus agents. Always.

Not replacement.
Not redundancy.
Augmentation.

The work here is as emotional as it is technical – and that’s something leaders often underestimate.

“That All Sounds Great… But How Do I Actually Do It?”

Well, we are definitely still figuring it out! But my honest answer is:
you don’t embed everything at once.

You pick:

• A small number of genuinely great agents

• Embed them deeply into core workflows

• Remove alternative routes where possible

• And support the change relentlessly

Momentum comes from depth, not breadth.

Where I’m Heading Next

I’m going to keep sharing what I’m learning as this evolves:

• What finally tips agents from “interesting” to “indispensable”

• What leadership looks like when AI becomes normal, not novel

• And where I get it wrong along the way

If you’re grappling with similar challenges – especially the embedding, not the building – I’d love to compare notes.

Because AI at scale isn’t really a technology problem.

It’s a leadership one.