AI agents are becoming a new way to build software and automation.

To adopt them effectively, teams need a clear approach for what to use from existing platforms, what to buy, and what to build, so effort stays focused on business value.

The 80–15–5 rule

A practical breakdown for adoption is:

• 80% of applications should come from existing platforms the organization already uses daily.
• 15% should come from specialized vendor solutions that solve industry-specific problems.
• 5% should be custom-built in-house, where it creates real competitive advantage.

This framing keeps teams focused. It prevents over-investing in custom work where commodity solutions are good enough.

Focus on core business value

Custom AI should be reserved for areas closest to the organization’s unique value proposition.

It is rarely worth building custom solutions for commoditized functions when the market already provides strong options that integrate well with your existing stack.

Build internal expertise through hands-on work

Building AI agents is not only an extension of data science. It is a new craft that requires experimentation, research, and iteration in real use cases.

A practical approach is to create small task forces that:

• partner with business units,
• co-create pilots that solve real problems,
• learn what works through repeated cycles,
• build reusable patterns over time.

Build the foundations that make agents work

Agents need more than models. They need: