Scaling Bespoke Work With Agents

One thing I keep running into again and again with agents is that very many patterns in the way we construct organizations and services are centered around maximizing the leverage and distribution of a resource-constrained set of domain experts. Everything from a BI dashboard to a SaaS product has this fan-in/fan-out style to encode that leverage into a single artifact.

But with agents, we can be much more point-to-point in how we deliver products and services by instead encoding that leverage in a machine that is able to produce arbitrary software and make arbitrary tool calls just in time for the user.

A constrained Data Science team will invest most of its effort in a set of golden fact tables and metrics dashboards that it expects can be adapted by other employees to answer 80% of their data questions. Members of senior leadership, though, will be able to yoink time from a member of the team to get bespoke metrics work done on their behalf.

In the world of agents, we can skip the intermediate step of the dashboard and instead have that Data Science team empower an agent to not only reproduce that dashboard, but understand the data ontology and business context around the metrics so the product the team ships to everyone is the same experience the SLT gets.

This is the same problem software engineers are solving today with harness engineering.