What really makes an AI agent work

By volume of code, a production AI agent is mostly scaffolding.

Around the core model, which supplies the reasoning, sits everything that turns intelligence into work: a loop that decides what happens next, memory that persists between steps, tools that act in other systems and gates that determine what each action may do.

Why do we care? Because that balance changes what you’re buying. Choosing an agentic system now means looking past ‘which model?’ to the system around it – the shape of the loop, the rules on each step and the evidence left at the end.

Our new Perspective maps the system, traces what it means for biopharma, insurance, manufacturing and sovereign workloads. And gives technical teams a practical guide.

So you can scale the scaffolding. Via our Substack (opens in a new tab) here.