.avif){kind=small}
The problem is not the technology. It is the perspective. Most executive teams are applying 2026 AI tools to 2010 business playbooks.
Having guided PUSH through its transformation from a Google Ads performance agency into an AI-native organisation, and after spending a lot of time with agency leaders, CMOs and AI enterprise strategists, one thing has become clear: successful AI integration is not about rapid adoption. It is about structural workflow design.
Key takeaways
- Most agencies are applying 2026 AI tools to 2010 business playbooks. The problem is structural, not technological.
- The four traps to avoid: treating AI as a decision-maker, using it purely to cut headcount, automating broken tasks, and chasing every new model release.
- A single person can now manage an entire client account end-to-end. The diamond team model replaces junior execution layers with AI agents managed by experienced strategists.
- Effective AI deployment starts by protecting human judgement first, then automating everything around it.
- For every £1 invested in AI tools, organisations need to invest £2 in talent training and change management.
Here is what leaders must stop doing, and what to do instead.
The 4 AI traps to avoid
1. Treating AI as the decision-maker instead of the tool
George Sullivan, founder of The Sole Supplier, put it well: you cannot delegate critical thinking and core human judgement to a language model. AI can process millions of data points in seconds. It has no intuition, no empathy, no lived experience.
Einstein did not conceptualise the theory of general relativity by running a query. He thought his way to it. When leaders hand over strategic direction and creative vision to AI, the output becomes derivative. It has no distinct brand identity. AI is a tool. A very good one. It is not a strategist.
2. Using AI purely to cut headcount
Treating AI as a cost-cutting instrument is a one-dimensional play. Cost reduction is a one-time saving. Expanding capacity and building new services compounds revenue over time.
As Informa's Group Chief Commercial AI Officer Amin Mrini argues, focusing solely on minimising heads per dollar of profit is a defensive move. While you're trying to cut 10% of your workforce, AI-native competitors are using the same technology to increase revenue per head, launch new service lines, and take market share. The maths only runs one way.
3. Automating broken tasks
Rishad Tobaccowala, former Chief Growth and Strategy Officer at Publicis, said it plainly: "Doing pathetic things faster and more pathetically still keeps you pathetic."
Distributing Claude or ChatGPT accounts to employees without a structural plan is not an AI strategy. Generating a PowerPoint deck in three minutes instead of two hours adds no enterprise value if the surrounding meeting cadence and decision-making structures stay unchanged. Automating isolated, broken tasks pushes operational bottlenecks further down the pipeline.
4. Chasing every new model release
Unless you are running pharmaceutical drug discovery or complex aerospace engineering, you rarely need the absolute frontier model on day one. Stressing over every four-week LLM release cycle wastes focus. Business leaders should focus on riding the cost curve down across proven workflows, rather than paying top-tier pricing to chase capability improvements that may not move the needle for their actual use cases.
Can 1 person really manage an entire client account with AI?
During my recent conversation with Jordan Platten on the Agency Giants podcast, Jordan put me on the spot with a direct question, "if I were starting PUSH from scratch in 2026, how would I build it?"
My answer surprised him. I believe a single person could manage an entire client account end-to-end.
In the clip, Jordan and I work through what this shift looks like in practice.
The diamond team model
Traditional agencies rely on a pyramid structure, with a large base of junior executives handling manual execution. AI collapses that base. The business shifts into a diamond structure, where experienced strategists manage internal AI agents rather than junior staff.
Start with one process
When Jordan asked where agency owners should begin without getting overwhelmed, my answer was straightforward: pick one time-consuming internal process, such as ad copy generation or reporting, automate it using Claude or custom agents, and refine it before scaling further.
How leaders should deploy AI
1. Invert the workflow question
Stop asking which tasks can be delegated to AI. Ask instead: what parts of this process require human judgement, strategic reasoning, and relationship management?
At PUSH, we maintain human-in-the-loop review for campaign strategy, client communication and critical account decisions. Once those human control points are protected, the surrounding administrative, data-gathering and execution tasks are chained into automated background workflows. The most effective AI integration is invisible to the end client.
2. Build a normalised data layer
AI is only as effective as the data context it works from. Many agencies attempt to build complex AI agents without giving them structured data to query.
At PUSH, the shift came when we consolidated client performance metrics, meeting call transcripts, pitch frameworks and historical campaign data into a centralised database via our proprietary platform, Dial 360. Because our AI operating system, Dial Agents, queries a normalised context layer, it produces tailored pitch roadmaps and campaign frameworks in seconds.
3. Equip your people properly
AI will eventually become a baseline utility across every industry. Your real differentiator is your people.
For every £1 invested in AI software and infrastructure, organisations need to invest £2 in talent training, AI fluency and change management. Think of it this way: giving your team an AI "Iron Man suit" turns capable people into high-value strategists who can manage end-to-end client outcomes that previously required entire departments. The suit matters less than the person wearing it.
4. Expand beyond marketing
AI's reach goes well past ad copy generation or basic SEO. Performance agencies need to use it to bridge marketing, sales and operations.
At PUSH, we have expanded into custom AI workflow integrations, generative engine optimisation (GEO) to surface clients on LLM platforms, and AI-driven reactivation bots via SMS and WhatsApp. Moving into operational and sales enablement creates stickier client partnerships and increases lifetime value in ways that traditional retainer models rarely achieve.
Where does this leave you?
Change is difficult. Irrelevance is worse. The businesses that survive this shift will not be the ones that bought the most software seats or ran the most pilots. They will be the ones that rebuilt how they work around AI, not alongside it. That is a structural decision. It has to be made deliberately.
Most agency leaders I speak to already know what needs to change. The gap is not awareness. It is the willingness to do the uncomfortable structural work that real transformation requires. Buying tools is easy; rebuilding around them is not.
Frequently asked questions about AI for agencies
Q1: What is the biggest mistake agencies make with AI in 2026?
The biggest mistake is treating AI as a cost-cutting tool rather than a capacity multiplier. Cutting headcount saves money once. Using AI to expand service delivery and increase revenue per head compounds over time. Agencies focused purely on reducing costs are ceding market share to competitors who are using the same technology to grow.
Q2: What is the diamond team model for agencies?
The diamond team model replaces the traditional agency pyramid, where a large base of junior staff handles manual execution. AI takes over that execution layer, leaving experienced strategists at the centre managing AI agents rather than people. The result is a leaner team delivering higher-value output per head.
Q3: How should an agency start its AI transformation without getting overwhelmed?
Start with one time-consuming internal process, such as ad copy generation or performance reporting. Automate it using a tool like Claude or a custom AI agent. Refine it until it runs reliably, then scale the approach to the next process. Trying to transform everything at once is the most common reason AI projects stall.
Q4: Why is a normalised data layer important for AI agents?
AI agents are only as useful as the data they can access. Without a structured, centralised data layer, agents produce generic output that does not reflect the client's actual performance history, objectives or context. Consolidating client data, call transcripts, campaign history and pitch frameworks into one queryable source transforms what agents can produce.
Q5: How much should agencies invest in AI training versus AI tools?
For every £1 spent on AI software and infrastructure, agencies should invest £2 in talent training, AI fluency and change management. Tools without trained people produce inconsistent results. The agencies seeing the strongest returns from AI are the ones that treated the human transition as seriously as the technology build.
