Insight
Insight
Everyone can build with AI. Now what?
14 Jun 2026
Matt Cull
When technology becomes accessible to everyone, the difficult part is no longer building. It is knowing what should be built in the first place.
AI has moved remarkably quickly from something organisations were exploring to something employees are already using every day. People who have never written software can create applications. Teams can build agents without waiting for lengthy development cycles. Tasks that previously required specialist technical knowledge can increasingly be completed through natural language.
That democratisation creates enormous opportunity, but it also creates an entirely new set of problems.
At the latest Powering Agency Business breakfast at Soho House, Vannda brought together senior leaders from organisations including Omnicom, AVANTGARDE Group, Inner Group, Visual Data Media Services, Clarity, Havas Media Network, Manifest Group, Oliver, Podimo and Cardlytics. The closed-door discussion explored what happens when AI capability spreads across an organisation faster than its operating model, governance and people can adapt.
The overriding sentiment was neither pessimistic nor blindly optimistic. There was excitement about what AI is making possible, but equally significant concern about how organisations turn experimentation into something reliable, scalable and commercially valuable.
The question is no longer whether everyone can build with AI. Increasingly, they can. The question is what happens next.
When everyone can build, judgement becomes more valuable
One of the clearest themes from the morning was that technical capability is rapidly becoming less of a constraint. As one participant put it, “Coding is not the problem.” The challenge is understanding the business problem, the process around it and whether the thing being created should exist at all.
Giving everyone access to powerful technology does not automatically give everyone the expertise required to use it well. Someone may now be able to generate a contract without understanding contract law, build an analytical model without being a data scientist or create software without understanding the process it is supposed to support. The output can look remarkably convincing while still being fundamentally wrong.
This was particularly concerning when the conversation turned to younger or less experienced employees. Several attendees described situations where AI had accelerated the journey towards an answer while simultaneously removing some of the thinking that would previously have happened along the way.
“It looks right. It sounds intelligent. But actually, what does this say? Does this make sense?”
The risk is not that people use AI. It is that confidence in the output begins to replace critical judgement. As one attendee reflected, AI can make organisations “go fast and… go really slow” when employees spend significant amounts of time arguing with a model before returning with an answer that an experienced colleague can immediately see is wrong.
From the Wild West to an operating model
The same tension exists at organisational level. AI experimentation has often grown from the bottom up, with individual employees and teams discovering tools, creating agents and solving problems independently. That freedom can be incredibly valuable, but it can also produce a fragmented estate of overlapping technologies with little visibility or control.
One organisation around the table had recently audited its AI use across approximately 400 employees and discovered 135 different AI tools being used. The response was not to stop experimentation, but to begin establishing guardrails, thinking more deliberately about structured data and understanding how individual developments fitted into a wider roadmap.
That example captured a challenge shared by organisations of very different sizes. Leaders want people to experiment because innovation is unlikely to emerge entirely through a centrally controlled programme. At the same time, allowing hundreds or thousands of disconnected experiments to proliferate does not constitute an AI strategy.
The challenge is creating enough freedom for people to discover new possibilities while establishing enough governance to ensure those ideas can eventually become reliable business capabilities.
Evolution may not be enough
The conversation also exposed a more fundamental question about transformation itself. Should organisations use AI to improve the processes they already have, or should they rethink those processes entirely?
There is an understandable tendency towards evolution. A business identifies an existing workflow, introduces AI into one part of it and measures the efficiency improvement. It feels manageable because the organisation remains recognisable afterwards.
But several attendees questioned whether that goes far enough. If AI fundamentally changes what is possible, simply making yesterday’s processes faster may eventually become the wrong objective.
“It’s like revolution rather than evolution.”
The difficulty is that revolution requires organisations to imagine an operating model they cannot yet fully see. Employees often need evidence before they will embrace radically different ways of working, but the full benefit may only become visible once those new ways of working have been implemented. That creates a familiar transformation dilemma: people want proof before changing, while the proof depends on making the change.
The pragmatic answer may lie somewhere between the two. Organisations can take incremental steps while keeping sight of a much more fundamental destination, using individual use cases to learn rather than allowing them to dictate the final operating model.
AI still needs foundations
For all the discussion about new organisational structures, there was strong agreement around one familiar principle: AI cannot compensate for weak foundations.
Poor data remains poor data. Broken processes remain broken processes. Inconsistent governance does not become consistent simply because an intelligent layer has been added on top.
One participant described this as “phase zero”: the work that must happen before the exciting technology is introduced. That means understanding the process, fixing the data, establishing governance and ensuring the organisation knows what a correct outcome actually looks like. Without that foundation, businesses risk automating problems they have never properly understood.
This becomes even more important as AI moves from individual productivity tools towards orchestration and agentic systems. The more autonomy technology is given, the greater the importance of knowing that the systems, processes and information underneath it can be trusted.
The commercial model has to move too
AI is not only challenging how agencies work. It is challenging how their work is valued.
Agency economics have traditionally been closely connected to people and time. If AI allows work that previously required weeks to be completed in days or hours, the relationship between effort, cost and value begins to break down. Clients understandably expect efficiencies to be reflected in price, while agencies still need to fund technology, retain expertise and take responsibility for the quality of the outcome.
The discussion suggested that this tension will increasingly push agencies towards value and outcome-based models, but making that transition is difficult when organisations still carry the infrastructure, headcount and commercial structures of the previous model.
Expertise nevertheless remains valuable precisely because accessibility is increasing. A client may be able to generate something that looks similar to an agency’s output at a fraction of the previous cost, but appearance and reliability are not the same thing. Experience provides the ability to recognise when an answer is wrong, understand context and judge whether something will actually work.
Power to the people needs responsibility too
Perhaps the clearest conclusion from the morning was that AI is shifting power throughout organisations. People have access to capabilities that would previously have required specialist teams, considerable budgets or lengthy implementation programmes. That shift should be celebrated because it creates the potential for dramatically more innovation.
But democratising capability also democratises responsibility.
Organisations now need to develop critical thinking alongside AI literacy, governance alongside experimentation and operational foundations alongside new tools. They need to preserve the institutional knowledge required to challenge AI outputs rather than removing expertise simply because a model appears capable of reproducing them. The conversation repeatedly returned to the danger of losing exactly the people capable of determining whether an automated answer is right.
That may ultimately be the most important shift facing agency leaders. AI strategy is moving beyond a conversation about tools and becoming a conversation about organisational design: who makes decisions, where expertise sits, how work is valued and which processes should continue to exist at all.
Everyone can build with AI. The organisations that gain the greatest advantage will be those that know what is worth building, what should never be automated and how to create the foundations that allow both people and technology to perform at their best.
That is where the real work begins.
When technology becomes accessible to everyone, the difficult part is no longer building. It is knowing what should be built in the first place.
AI has moved remarkably quickly from something organisations were exploring to something employees are already using every day. People who have never written software can create applications. Teams can build agents without waiting for lengthy development cycles. Tasks that previously required specialist technical knowledge can increasingly be completed through natural language.
That democratisation creates enormous opportunity, but it also creates an entirely new set of problems.
At the latest Powering Agency Business breakfast at Soho House, Vannda brought together senior leaders from organisations including Omnicom, AVANTGARDE Group, Inner Group, Visual Data Media Services, Clarity, Havas Media Network, Manifest Group, Oliver, Podimo and Cardlytics. The closed-door discussion explored what happens when AI capability spreads across an organisation faster than its operating model, governance and people can adapt.
The overriding sentiment was neither pessimistic nor blindly optimistic. There was excitement about what AI is making possible, but equally significant concern about how organisations turn experimentation into something reliable, scalable and commercially valuable.
The question is no longer whether everyone can build with AI. Increasingly, they can. The question is what happens next.
When everyone can build, judgement becomes more valuable
One of the clearest themes from the morning was that technical capability is rapidly becoming less of a constraint. As one participant put it, “Coding is not the problem.” The challenge is understanding the business problem, the process around it and whether the thing being created should exist at all.
Giving everyone access to powerful technology does not automatically give everyone the expertise required to use it well. Someone may now be able to generate a contract without understanding contract law, build an analytical model without being a data scientist or create software without understanding the process it is supposed to support. The output can look remarkably convincing while still being fundamentally wrong.
This was particularly concerning when the conversation turned to younger or less experienced employees. Several attendees described situations where AI had accelerated the journey towards an answer while simultaneously removing some of the thinking that would previously have happened along the way.
“It looks right. It sounds intelligent. But actually, what does this say? Does this make sense?”
The risk is not that people use AI. It is that confidence in the output begins to replace critical judgement. As one attendee reflected, AI can make organisations “go fast and… go really slow” when employees spend significant amounts of time arguing with a model before returning with an answer that an experienced colleague can immediately see is wrong.
From the Wild West to an operating model
The same tension exists at organisational level. AI experimentation has often grown from the bottom up, with individual employees and teams discovering tools, creating agents and solving problems independently. That freedom can be incredibly valuable, but it can also produce a fragmented estate of overlapping technologies with little visibility or control.
One organisation around the table had recently audited its AI use across approximately 400 employees and discovered 135 different AI tools being used. The response was not to stop experimentation, but to begin establishing guardrails, thinking more deliberately about structured data and understanding how individual developments fitted into a wider roadmap.
That example captured a challenge shared by organisations of very different sizes. Leaders want people to experiment because innovation is unlikely to emerge entirely through a centrally controlled programme. At the same time, allowing hundreds or thousands of disconnected experiments to proliferate does not constitute an AI strategy.
The challenge is creating enough freedom for people to discover new possibilities while establishing enough governance to ensure those ideas can eventually become reliable business capabilities.
Evolution may not be enough
The conversation also exposed a more fundamental question about transformation itself. Should organisations use AI to improve the processes they already have, or should they rethink those processes entirely?
There is an understandable tendency towards evolution. A business identifies an existing workflow, introduces AI into one part of it and measures the efficiency improvement. It feels manageable because the organisation remains recognisable afterwards.
But several attendees questioned whether that goes far enough. If AI fundamentally changes what is possible, simply making yesterday’s processes faster may eventually become the wrong objective.
“It’s like revolution rather than evolution.”
The difficulty is that revolution requires organisations to imagine an operating model they cannot yet fully see. Employees often need evidence before they will embrace radically different ways of working, but the full benefit may only become visible once those new ways of working have been implemented. That creates a familiar transformation dilemma: people want proof before changing, while the proof depends on making the change.
The pragmatic answer may lie somewhere between the two. Organisations can take incremental steps while keeping sight of a much more fundamental destination, using individual use cases to learn rather than allowing them to dictate the final operating model.
AI still needs foundations
For all the discussion about new organisational structures, there was strong agreement around one familiar principle: AI cannot compensate for weak foundations.
Poor data remains poor data. Broken processes remain broken processes. Inconsistent governance does not become consistent simply because an intelligent layer has been added on top.
One participant described this as “phase zero”: the work that must happen before the exciting technology is introduced. That means understanding the process, fixing the data, establishing governance and ensuring the organisation knows what a correct outcome actually looks like. Without that foundation, businesses risk automating problems they have never properly understood.
This becomes even more important as AI moves from individual productivity tools towards orchestration and agentic systems. The more autonomy technology is given, the greater the importance of knowing that the systems, processes and information underneath it can be trusted.
The commercial model has to move too
AI is not only challenging how agencies work. It is challenging how their work is valued.
Agency economics have traditionally been closely connected to people and time. If AI allows work that previously required weeks to be completed in days or hours, the relationship between effort, cost and value begins to break down. Clients understandably expect efficiencies to be reflected in price, while agencies still need to fund technology, retain expertise and take responsibility for the quality of the outcome.
The discussion suggested that this tension will increasingly push agencies towards value and outcome-based models, but making that transition is difficult when organisations still carry the infrastructure, headcount and commercial structures of the previous model.
Expertise nevertheless remains valuable precisely because accessibility is increasing. A client may be able to generate something that looks similar to an agency’s output at a fraction of the previous cost, but appearance and reliability are not the same thing. Experience provides the ability to recognise when an answer is wrong, understand context and judge whether something will actually work.
Power to the people needs responsibility too
Perhaps the clearest conclusion from the morning was that AI is shifting power throughout organisations. People have access to capabilities that would previously have required specialist teams, considerable budgets or lengthy implementation programmes. That shift should be celebrated because it creates the potential for dramatically more innovation.
But democratising capability also democratises responsibility.
Organisations now need to develop critical thinking alongside AI literacy, governance alongside experimentation and operational foundations alongside new tools. They need to preserve the institutional knowledge required to challenge AI outputs rather than removing expertise simply because a model appears capable of reproducing them. The conversation repeatedly returned to the danger of losing exactly the people capable of determining whether an automated answer is right.
That may ultimately be the most important shift facing agency leaders. AI strategy is moving beyond a conversation about tools and becoming a conversation about organisational design: who makes decisions, where expertise sits, how work is valued and which processes should continue to exist at all.
Everyone can build with AI. The organisations that gain the greatest advantage will be those that know what is worth building, what should never be automated and how to create the foundations that allow both people and technology to perform at their best.
That is where the real work begins.