Skip to content
Subscribe
Rafael Cordeiro, colunista

Agents across companies and the paradox of invisible decisions

Rafael Cordeiro · 2 min read
Share

We are trying to connect agents across companies that still cannot explain how they make their own decisions.

This may be one of the biggest paradoxes of AI adoption in large organizations.

More data, the same misalignment

A recent BCG study on collaboration between consumer packaged goods companies and retailers lays the problem bare.

For years, the answer was to share more data. POS, inventory, sell-out, forecast, demand.

Today, much of that data already flows between companies. Even so, the misalignment persists.

Why? Because sharing data does not mean sharing decisions.

The retailer operates on short-term demand signals. The manufacturer has to turn those signals into production, inventory and logistics decisions that may be made months in advance.

These are two organizations looking at related data, but operating with different contexts, goals and decision horizons.

That is exactly where Agentic AI can change the game.

Editor's context. CPG stands for consumer packaged goods, the food, beverage, personal care and household products industry. POS and sell-out are point-of-sale data, sales to the end consumer.

The size of the opportunity

According to BCG, experience with large manufacturers and retailers shows potential for a 2% to 4% improvement in product availability and more than 60% productivity gains in planning.

To give a sense of scale, for a retailer with revenue between R$ 500 million and R$ 3 billion, a 2% to 4% improvement in product availability could represent somewhere between R$ 10 million and R$ 120 million in potential additional sales.

The warning sign

But there is an even more important data point. About 75% of the consumer goods companies BCG surveyed are still piloting or exploring AI, according to BCG's June 2026 survey with the Consumer Goods Forum. And more than half of the companies surveyed do not formally measure the ROI of their AI investments.

That should set off an alarm.

We are discussing agents able to negotiate inventory, planning, demand and execution between companies. But first we need to solve something more fundamental inside them.

  • What context led to a given decision?
  • Which data was considered?
  • Which rules and constraints were involved?
  • Who made the decision, a human or an agent?
  • What was executed?
  • And what outcome did that decision actually produce?

Before autonomy, operational intelligence

This is the point I believe is being underestimated in the race for agents.

Before the autonomy layer, we need to build an operational intelligence layer. A layer that connects data, context, decisions, execution and outcomes.

Because connecting two agents is relatively easy. The hard part is ensuring they are making the right decision, with the right context, within the right rules. And proving that it created value.

Perhaps the next leap in enterprise AI will not come from smarter agents. It may come from companies that finally manage to make their decisions visible, governable and measurable.

In short

  • Sharing data does not mean sharing decisions.
  • Before the autonomy layer, an operational intelligence layer is needed.
  • The next leap may come from companies that make their decisions visible, governable and measurable.

Further reading

Originally published on LinkedIn on August 19, 2026. See the original post. This version has been revised and expanded with context and sources. The views are those of the author, a co-founder of OperOS AI.

Read next