Interview With Tobias Cummins, COO At Pencil On Agentic AI In The Marketing Context

What does agentic really mean in a marketing context?

I think there’s a lot of noise around the term “agentic” at the moment, and people often jump straight to the models themselves. But agentic AI is really about what happens when those models are able to take actions and move through workflows, rather than simply respond to prompts.

Most people are familiar with AI as a chat interface. You ask ChatGPT or Claude to do something, it gives you an answer, and the interaction ends there. Agentic AI goes a step further. It allows AI systems to complete multi-step tasks, make decisions within set guardrails, and connect into the wider tools and processes a business already uses.

In a marketing context, that could mean AI helping move work from briefing and production through to approvals, versioning, publishing and performance analysis with far less manual intervention.

The models themselves are things like GPT, Claude, Firefly or Veo, but it’s the harness around those models that gives them context and structure. That’s what allows them to operate more autonomously inside a real marketing workflow, while still respecting things like brand guidelines, approvals and business objectives.

Why do you think the current approach to agentic AI use in marketing often feels disjointed, and how can businesses take a more cohesive approach to implementation?

A lot of businesses fell into what I’d call “pilot purgatory”. You have organisations saying, “Right, everyone needs to use AI,” and then different teams running isolated experiments across the business. One team speaks to Adobe, another speaks to OpenAI, another builds something internally.

Those pilots might generate small pockets of efficiency, but when you aggregate them together, they often don’t amount to meaningful ROI at scale.

Three years ago, I do think this was probably the only thing businesses could reasonably do. The technology was moving incredibly quickly and nobody fully understood what large-scale implementation would really look like.

Now, though, we’re reaching a point where leaders understand that these things need to be orchestrated together. Businesses that are seeing the most progress have taken a much more holistic view of how marketing work actually flows through the organisation, from briefing and production through to launch, optimisation and performance.

To be successful, marketing teams need to move away from just using AI to generate assets faster, and towards embedding full AI systems that remove friction from the fragmented processes underneath.

What is causing the disconnect between the promises made by agentic martech vendors and how those AI models are working in reality? 

Part of it is simply the pace of change. This space is moving faster than anything most marketers have experienced before. You can’t reasonably expect every CMO to fully understand the nuances between every model provider, every release cycle or every capability update.

At the same time, vendors naturally have their own ambitions and are pushing their own ecosystems very aggressively. So marketers are being hit from every direction with messaging about the next big breakthrough or the next must-have platform.

The risk is that businesses lock themselves too tightly into a single ecosystem too early. We’ve already seen how quickly the landscape changes. The dominant player six months ago may not be the dominant player six months from now.

That’s why I think flexibility matters so much. Businesses need a way to leverage the best available models without constantly rebuilding their entire workflow every time the market shifts.
 

What does an agentic marketing operating system do to solve this problem?

An agentic marketing operating system acts as the layer that sits above the models themselves. The models are incredibly powerful, but on their own they’re a bit like lightning in a bottle. You still need something that points that power in the right direction and gives it context.

So that’s why the operating system becomes important. It connects the models into real marketing workflows, understands the nuances of how marketing organisations operate, and allows businesses to move much faster without constantly rebuilding processes from scratch.

One of the principles we built Pencil around was avoiding dependency on a single model provider. As new enterprise-grade models emerge, they can be integrated quickly, which means marketers aren’t forced into betting everything on one ecosystem or one vendor’s roadmap.

That dramatically lowers the risk for businesses while still giving them access to the latest capabilities.

What changes can businesses make to ensure marketers feel empowered and supported to use AI?

This is probably one of the most important conversations happening right now. AI transformation can very easily become something that feels threatening to people if it’s handled badly. I’ve always believed AI should be a tool for people, not a replacement for people. How you build these systems really matters because you can either proactively empower individuals or unintentionally disempower them.

One thing I find really interesting is the idea that marketers increasingly build their own operational IP through AI systems. The prompts, workflows, agents and processes they create become valuable assets in their own right. Imagine someone spending years refining highly specialised AI workflows that reflect their expertise and intuition. That knowledge becomes part of their value as a marketer.

Businesses that approach AI with empathy and transparency will ultimately get far more from it than the businesses that treat it purely as a cost-cutting exercise.

As AI takes on more responsibility in marketing workflows, how can marketers pivot to provide value?

The human part becomes even more important. A lot of the binary or repetitive work can absolutely be automated, but marketing has always relied heavily on intuition and understanding human behaviour. That doesn’t disappear.

To thrive, people need to understand how to shape these systems around real business challenges and creative nuance. The technology is only as good as the context and thinking that sits behind it.

I also think marketers shouldn’t underestimate the value of their lived experience. Sometimes when organisations introduce AI, it can feel like someone is arriving with a red pen to critique the way you’ve worked for the last ten years. That’s difficult for people.

But the reality is that the best AI systems still depend heavily on human insight. Understanding how work flows through an organisation, how creative decisions get made, how brands behave in culture – those things matter enormously.

Marketers should lean into that expertise and combine it with strong AI capability. That mix will become incredibly valuable over the next few years.