Does Platform Engineering Reduce Cognitive Load for Al Agents?

Agentic Al is taking over. It's all fine (or is it?) when used within the scope of a single team's mission.

The question is now what happens when these agents cross organizational boundaries? Do we use the same interaction modes as we did for human-only teams? How can we allow them to go fast while keeping reasonable guardrails on what they are allowed to do? And do agents have cognitive limits as well?

Turns out platform engineering might just have an even more critical role in this new reality. The internal platform guardrails and direction might just be the last line of defense between going fast and being reckless.

Of course, we (and our leaders and our leaders' leaders, you get the drift) want our organizations to successfully adopt Al at scale and accelerate the business but without putting the business at risk! So how do we do that?

We need first of all to provide actionable and accurate context for agents to thrive with adequate risk, security, compliance and even ethical concerns in place.

Secondly, agents need sufficient leeway to find faster, more efficient (and often more localized) ways to get things done but also sufficient direction and paths that have been paved by human judgement before.

Finally, agents are born out of human decision (for now). This changes the interactions between systems but not the need for intentional interactions between the teams evolving those systems. You still need collaboration between teams in (end) product and platforms to take place so you can evolve the interactions between product and platform agents to become X-as-a-Service!

What's the logical place for all the above? You're absolutely right, an internal platform that: 1. provides sufficient guardrails (not more); 2. favors descriptive over prescriptive paved paths; and 3. focuses on reducing cognitive load for agents and humans alike.

What about the teams working inside the platforms? Well, they become more like orchestrators than builders - actively learning how product teams are using Al, identifying what's working, and feeding those approaches back into the platform for others to benefit from.

And of course teams working on the platforms should use Al to reduce their own cognitive load. But that's a whole other talk...