The marketing tech stack is broken.
Especially for agents.
I believe the future is agents plugging into a marketing operating system (MOS).
Here's my dream system.
What is a marketing operating system?
A marketing operating system (MOS) is a layered architecture that gives AI agents the context, rules, and infrastructure they need to make autonomous decisions across marketing. Unlike a traditional martech stack (built for humans clicking through UIs), a marketing operating system is purpose-built for agents: it exposes clean identity data, encodes business logic as queryable policy, handles decisioning, executes across channels, and learns from outcomes in a connected system where each layer feeds the next.
The key distinction: a martech stack is a collection of tools. A marketing operating system is a system, one where the layers are load-bearing and the agent has everything it needs to act without a human making each call.
The six layers of a marketing operating system
The data layer
At the base of the MOS is a data layer that focuses on identity management and signal capture. This allows the agent to know who the lead is, what part of the buying group they represent, and what they are most interested in.
The lifecycle layer
Using this data MOS can then determine the lifecycle of the lead. The lifecycle here has to be based on identity. So the agent needs to know where the account is in the sales lifecycle. Then where the rest of the buying group is in the process. And then and only then where the lead is in the lifecycle.
The policy layer
Next up is a policy layer. Here the agent will need to know everything about my business. What's considered a good ICP fit. What types of leads go straight to AEs vs BDRs. Plus thousands of other parameters the agent will use to make decisions.
The decisioning layer
Arguably the most important layer to me is the decisioning layer. Here based on the combo of the lead's lifecycle stage and policies, the MOS will decide what's the next best action to take. This could be sending the lead directly to sales. It could be dropping them into a nurture campaign. It could be adding them to a paid social retargeting campaign.
The action layer
With the decision in hand the MOS will take action across multiple channels. Email has to be one of many channels to reflect the realities of the modern marketer. The MOS needs to be omnichannel.
The learning and measurement layer
Lastly, there's a learning and measurement layer, the holy grail of marketing. The MOS should see what's working where and for whom. As an example, the MOS realizes that a particular ICP responds better to retargeting, but copy & design need to change to improve results. This starts the process all over again to improve results.
What a marketing operating system needs to work
Now I've been thinking a lot about what's needed to make the system work. The MOS needs to leverage multiple LLM providers interchangeably. And it needs to be flexible enough to use different providers at the individual task level. I wouldn't want to bet my future on a single provider. So much is happening so fast that you need flexibility on both performance and cost.
Equally important will be an observability layer so that the MOS isn't a black box. It has to be easy to troubleshoot where in the stack things may have gone wrong.
So too must there be a governance layer to ensure reads and writes happening with MOS in a way that respects your security and data needs.
Lastly, you can't have a MOS without having a way to test changes made. You create a new policy, you want to see its impact before deploying. You make a change to a prompt, you want to see if it causes any regression.
What am I missing in the stack? Anything you'd change?
