Applying Intelligence Late

Traditional software applies a great deal of intelligence early, during product design. Developers anticipate what customers might want, select which features belong together, encode workflows and build interfaces. Customers inherit those choices.

A growing alternative is to apply more intelligence late, at the point of use.

The user's agent can know the actual goal, constraints and available resources. It can select a suitable capability, configure it, combine it with others or ask for something new to be built. Decisions that once had to be guessed for a market can be made for a particular situation.

Discovery becomes part of operation

Today, people often search for software, compare products, register, integrate and learn them before they can solve a problem. Agents could progressively take on much of that work.

Imagine asking for a small report, and the agent finding a specialist transformation service, evaluating its terms, using it and returning the result. The user may not need to know the service's name. This is a change in distribution as much as interaction: agents become software discoverers.

Composition happens later too

Instead of buying a predetermined application suite, a person or business could acquire only the capabilities needed for its current job. An agent can orchestrate those capabilities and revise the composition when requirements change.

This does not mean that all intelligence migrates into general-purpose agents. Specialised algorithms, databases and domain-specific engines may still embody deep expertise. The shift concerns, especially, which capabilities to use, how to connect them and how to operate them.

A continuing process

Late intelligence need not be a single setup step. Requirements change; new services become available; costs and constraints shift. An agent can revisit the composition repeatedly.

The long-run consequence could be software disaggregation: features that once had to be sold together can compete separately. It may also become easier for an agent to create a missing capability than to find an existing product that almost fits.

There are new risks: discovery can be biased, opaque or captured by a powerful intermediary. Permissions, provenance, cost transparency and the ability to inspect or override an agent's choice will matter.

The central idea is not that product designers become irrelevant. It is that more decisions can be delayed until their context is actually known.