Where the conversation has got to
At our recent AI in Practice event, one thing stood out to me more than anything else. The conversation has moved on. A year ago, many discussions about AI revolved around definitions, possibilities and predictions. What is agentic AI? Which tools are worth paying attention to? Is this hype, or is something genuinely changing? Those questions have not disappeared entirely, but they are no longer the ones dominating the room.
What struck me during the event was how much more practical the discussion has become. People wanted to know which models were best suited to different tasks. They wanted to talk about products, internal tooling and implementation. They wanted to understand where value is actually being created. These are the questions of people who have moved beyond curiosity and into action.
I also noticed, both during the sessions and in the conversations afterwards, that there were two separate discussions taking place.
The first was about engineering – How AI changes software development, product delivery and technical workflows. This conversation feels well established. There is already a shared language around it.
The second was about business users. How people in finance, operations, HR, customer service, compliance and other teams can use AI in their day-to-day work. This conversation feels newer. It is also where much of the next wave of value will come from.
Why a build is the natural instinct
Most organisations now recognise that getting value from AI is just as much about people as it is about technology. The next challenge is deciding what to do about it.
When teams decide to act, their instinct is usually to build something. That makes sense. A build is tangible. It has a clear scope, a deliverable, a project plan and a finish line. There is something visible to show when it is done. Progress can be tracked. Investment can be justified.
For organisations used to delivering change through projects, building something feels familiar and measurable. The challenge is that not every AI opportunity needs a build.
There are actually two distinct types of opportunity, and only one of them necessarily requires bespoke development.
Understanding the difference between service-level work and individual work
Some opportunities sit within workflows that span multiple people, systems and approval stages.
Think about processes that cross departments, require sign-off, move information between systems or involve multiple handovers. This is service-level work.
In these situations, bespoke applications, automations or AI agents are often the right answer because the value comes from improving a shared process. The outcome affects a service rather than a single individual.
But not all work looks like that. A significant amount of value sits within tasks that belong entirely to one person.
These are the activities that happen every day or every week. The work people have developed their own way of doing over time. The tasks that take hours but would never justify a formal software project. This is individual work.
An operations manager summarising reports. A compliance professional reviewing documentation. A project lead preparing stakeholder updates. A business analyst pulling together information from multiple sources. No one would commission a bespoke application for each of these tasks. Yet collectively, they represent a huge amount of time and effort.
Getting AI into this work often does not require building anything. It means giving people access to capable tools and support to learn how to use them well. That is where platforms like Microsoft 365 Copilot, Claude and ChatGPT are creating value.
The challenge is that enabling people is harder to see than a build. It is harder to put on a project plan, harder to measure in the usual ways, and harder to call finished.
As a result, organisations often choose bespoke development even when user enablement could deliver more long-term value.
Making enablement as concrete as a build is some of the most important work happening right now. It is what unlocks real value from the tools already in place.
What I am learning from doing it
Much of this is not theoretical for me. I am seeing it play out in real conversations with teams that are trying to understand where AI fits into their organisation.
One pattern appears consistently. People naturally focus on the biggest problem facing their team.
I actually like that instinct. They have looked at something that has frustrated them for years and recognised that it might finally be solvable. That ambition is a positive thing.
But the opportunity is often broader than the biggest process challenge. Starting with smaller individual tasks can unlock value in places a traditional build may never reach.
Take onboarding as an example. Viewed as a whole, onboarding is service-level work. It spans systems, teams, approvals and responsibilities. It is the process where automation or bespoke tooling may eventually make sense.
But within that process are individual activities owned by one person from start to finish. Those can often be improved right away using AI tools already available.
By starting there, something interesting happens. People build confidence. They develop new habits. They find practical use cases. Patterns start to emerge. Over time, those lessons show where larger workflow improvements may be worth it.
In many cases, enablement and building are not competing approaches. Enablement helps reveal what is actually worth building.
The lesson software teams learned years ago
There is a lesson here that the software industry learned a long time ago. A system does not succeed simply because it has been delivered. Successful delivery requires training, adoption, support and change management alongside the technology itself.
Buying licences for AI platforms and rolling them out across a team follows the same principle. Nothing may have been built. There may not be a formal project. But if people are left to figure things out alone, the outcome is often the same.
This time, we have a chance to apply those lessons on purpose, not relearn them. That applies to organisations adopting AI and to consultancies helping them do it.
A practical way to start
For organisations looking to create value from AI, a few principles seem to work again and again.
Put the tools into people’s hands
Real understanding comes from practical use. Give people access early and help them build confidence through experimentation.
Create space for support
The most useful insights usually emerge through questions, discussions and shared learning. A support channel can become one of the richest sources of feedback available.
Start with everyday work
Look for tasks that one person owns, involve words or data, have a clear outcome and carry relatively low risk if things go wrong.
Separate individual work from service-level work
When a workflow crosses teams, systems or approval stages, identify the components that belong to an individual and start there.
Build with evidence, not assumptions
The more teams learn about how work is actually done, the clearer it becomes where automation, agents or bespoke tools will create the most value.
Focus on enablement, not dependency
You can analyse a team’s work, create recommendations and hand over a list of opportunities. Or you can help people build the capability to spot those opportunities themselves.
Only one of those approaches continues creating value after the engagement ends.
What you leave behind should be capability, not a dependency.
What finished looks like
An AI enablement programme has a clear destination. People are onboarded. Tools are set up. Skills are developed. Working practices evolve. Outcomes are measured.
But the capability itself should not stop growing when the engagement finishes. In fact, that is arguably the strongest indicator of success.
The goal is not to create a team that depends on external support every time a new opportunity appears. The goal is to create a team that keeps developing its own capability, spotting new opportunities and improving how it works long after the project ends.
That is why finding internal champions matters. That is why adoption matters. That is why enablement deserves as much attention as any technical implementation.
The deliverable is embedded capability, not a list.
If the last year of AI adoption has been about discovering what the technology can do, the next year may be about something else entirely – helping people use it well.
Written by Ryan Grey, CTO
If you want help shaping your own AI or digital strategy, our team at Marra is always happy to talk.
