For the last three years, most conversations about enterprise AI have centred on one question: what can it do?
At the Microsoft Frontier Tour in London on the 21st October (the event formerly known as the Microsoft AI Tour), it felt like the conversation had shifted.
AI is becoming more capable at an incredible pace. Agents can handle increasingly complex tasks, work across multiple systems and continue operating without someone constantly guiding them. At the same time, organisations are gaining access to more models than ever before, giving them greater flexibility to balance speed, accuracy and cost.
But as the technology matures, capability is becoming less of the challenge.
The bigger question now is whether organisations are ready to use it effectively.
Where does AI genuinely add value? How do you govern it? How do you measure success? Who owns it when something goes wrong? And perhaps most importantly, how do you move beyond experimentation and into something that creates lasting business value?
Those were some of the themes that stood out most throughout the event.
AI proofs of concept only tell part of the story
The experimentation phase has been important.
Without it, many organisations wouldn’t have developed the confidence to explore AI in the first place. Pilots and proofs of concept have helped teams understand what’s possible and sparked ideas that might never have emerged otherwise.
The challenge is that proving something can work isn’t the same as proving it can work at scale.
Most proofs of concept happen in relatively controlled environments. The data is cleaner. The users are engaged. The complexity is limited.
Real business processes tend to be messier.
Legacy systems, inconsistent data, changing requirements, governance concerns and budget pressures all have a habit of appearing once something moves into production.
That’s why a successful pilot should be viewed as the start of the conversation, not the end of it.
One example shared by Microsoft really brought this to life. When looking at opportunities within its cloud supply chain, the focus wasn’t initially on where an agent could be deployed. Instead, the team started by understanding the process itself and the decisions that sat behind it.
The result was cycle-time reductions of up to 75% in selected workflows.
The important lesson wasn’t that AI improved the process. It was that the process was understood and redesigned first.
Human review can’t be your only safety net
There’s a big difference between AI that creates content and AI that takes action.
If AI drafts an email or creates a summary, there’s usually a person reviewing it before anything happens.
Once an agent starts updating records, triggering workflows or interacting with customers, that layer of oversight can quickly disappear.
At that point, agentic AI governance can’t rely on someone spotting issues after the fact. It needs to be built into the way the process operates.
Microsoft’s investments in areas such as agent identity, lifecycle management, auditing and observability show how seriously this challenge is being taken.
But technology can only provide part of the answer.
Organisations still need to decide where agents should operate, what decisions they can make and where human involvement remains necessary.
The more capable AI becomes, the more important those conversations become too.
Start with the problem, not the technology
This sounds obvious, but it’s still surprisingly common to see organisations start with a tool and then go looking for a problem it can solve.
The stronger approach is usually the opposite.
One of the recurring messages from Microsoft was that no single model is best at everything. Different models have different strengths, different costs and different trade-offs.
That’s leading to a much more pragmatic approach, where the technology is selected based on the task rather than the other way around.
Take something like invoice processing.
You might use AI to interpret poorly formatted documents. Traditional automation might handle approvals. An AI agent might only become involved when an exception needs investigating.
In reality, many successful AI solutions look less like an AI transformation and more like a well-designed combination of technologies working together.
Sometimes the smartest AI strategy involves less AI than people expect.
The constraint has moved
Perhaps the biggest takeaway from Microsoft’s Frontier Tour wasn’t any individual feature or announcement.
It was the growing sense that technology is no longer the primary barrier.
For much of the last three years, organisations were limited by what AI could do.
Increasingly, they’re being limited by how effectively they can implement, govern and scale it.
The organisations that create lasting value from AI are unlikely to be the ones running the most experiments.
They’re more likely to be the ones that can repeatedly identify the right opportunities, apply the right technology, measure the outcome and operate it responsibly at scale.
That’s where the next stage of the AI journey is heading. And for many organisations, that’s where the real work begins.
If these questions sound familiar, join us at Frontier Forum on 24 November at the Barbican, London. We’ll explore what’s holding AI progress back and how to turn ambition into practical, measurable results, with fascinating keynote speakers, real customer stories and live demos. Find out more and apply for your place.
By Trent Powell
Head of AI Solutions
