Intent engineering crystallized this week as the architectural discipline above context and harness, the layer that prevents agents from optimizing the wrong thing at scale. Five independent sources across consulting, executive press, practitioner frameworks, and vendor governance converged on the same answer, which is the signal itself.
Coforge formalized the three-layer stack: prompt, context, intent. The grounding is a sixty-million-dollar cautionary tale. A fintech agent resolved tickets five times faster than humans, took on the work of roughly seven hundred agents, and was rehired a year later. The measurable proxy (resolution time) and the actual goal (customer outcomes) had come apart. A well-functioning system optimized faithfully for a target that did not match the organization’s real objective. That is the characteristic failure mode of the intent gap.
Forbes extended the parallel to Andy Grove’s OKRs. OKRs aligned people to strategy for a generation. Intent engineering now aligns people and agents, and agents to each other. Leaders have five years, not twenty, because once the infrastructure is built, organizations will add fifty agents, then five hundred, that operate continuously. Intent engineering is a management operating model, not an IT initiative.
The empirical evidence is already here. Schellman’s 2026 State of AI Governance survey of five hundred-plus US organizations found autonomous agents in production at roughly fifty percent of surveyed orgs, but fewer than one in three have operational governance maturity. Governance maturity doubles production probability. The Harness State of Agent DLC, polling seven hundred engineers, found seventy-five percent in production with confidence in the mid-seventies, yet eighty-seven percent had a security event this year and sixty percent overspent their budget. Confidence without the controls that justify it.
The North Star Problem is no longer how to prevent intent fragmentation. It is how quickly business architecture can operationalize the intent layer before the next fifty agents ship without it. The discipline belongs in business architecture, not IT. The organizations that recognize that will be the ones whose agents do the right thing at scale. The rest will have agents that do something, fast.