This paints a picture of an industry going through a state of chaotic experimentation. Adoption is happening but it is largely bottom-up, driven by enthusiasts finding their own efficiencies as they can. The danger here is not just inefficiency, inappropriate use cases or data insecurity, though those are genuine risks; the greater danger is homogenisation of output.
When AI use lacks a strategic overlay it risks drifting into a sea of sameness. In a recent conversation I was asked if a document I had produced had been 'created by AI'. It’s an interesting question without a single-word answer. I had indeed used Gemini to help me create the document, but the whole process had the full context of an enterprise AI solution and began with a clear instruction of "Let me start by giving you my opinions on..."
There is a subtle but critical distinction here that is easily missed. To me, the large language model (Gemini in this case) is a tool for ordering my thoughts, challenging the content, coming up with a first draft of the text and checking the final version for errors. It plays the role of a hardworking assistant. If someone with no expertise on the subject were to try to prepare an equivalent document the interaction would be quite different. They might approach the tool with little knowledge of the subject and simply ask it to source all the ideas and strategy, which of course Gemini would happily do.
The output in the second scenario will be very plausible, probably competent and cover the standard points of the subject. But it will also be what we might consider as the average response of the internet. It’s as if you just Googled the subject line and pasted in whatever came back. It’s unlikely to be the best answer to the question and it certainly won’t be distinctive. If agencies are simply using these tools to generate answers rather than to structure their own expertise they just become a middleman who won’t stay in the process for long. We’re all going to have to become accustomed to this idea and look beyond the simplistic question of ‘was this generated by AI’ because the reality is much more nuanced.
This is where the distinction between public tools and enterprise solutions becomes fundamental. When we use our internal enterprise instance of Gemini, we are not just 'Googling' an answer. We are using a retrieval-augmented generation (RAG) system which has access to our entire company knowledge base. This gives it the benefit of the Jaywing methodology which has been developed by working with clients over many years as well as the learnings from tackling a wide variety of specific client challenges. This information doesn’t exist on the internet therefore the answer is not a generic internet consensus; it is an answer retrieved from Jaywing’s extensive knowledge base of expertise and experience.
This all has implications for the agency commercial model. If the agency model, either knowingly or otherwise, enables a junior executive to generate a set of media campaigns which can be created in seconds and which are very plausible, it is risking its very existence. If they are implemented verbatim the client might as well have done it themselves and manage the whole process in-house.
AI has raised the bar. The value of an agency in the post-AI world is not the production of any given asset itself, but the insight and experience it has gathered which informs its creation and which AI can help to harness and use. The agent or prompt becomes a highly contextual entity that guides the creation process throughout and the output is interpreted by an expert who can enhance it with deep domain knowledge and can therefore deliver a superior solution that an LLM using generic public information cannot replicate.
The report suggests that many fear AI will erode their unique selling proposition or herald a descent into sameness. This is a valid fear for those who use AI as the oracle and the source of strategy. But for those who use it as an assistant to structure, present and use deep experience, it is an accelerator of quality output and will actually become a source of distinctiveness.
We all need to be transparent about this. There is no shame in admitting that an AI drafted a document or summarised a meeting. The currency of any agency is the insight, the strategy and the ideas that fuelled the propositions; AI isn’t best used to provide the expertise, it simply quickens the production process. For every opportunity these fantastic tools provide, the core challenge remains the same: delivering a better return on marketing spend. That is complex and achieving it is not as simple as asking in a prompt how to do it.
Before you ask, Gemini was involved in the creation of this article, but the original idea, opinions, observations and full editorial control all stayed with me even if some of the information gathering, typing and checking was done by my trusty assistant Gemini.
*(of an admittedly small sample including a good share of smaller agencies)