(Editor’s Note: What follows is an excerpt from a new book by Jesper Lowgren, Agentic System Design. It introduces radical new thinking and doing, and sets a target state of what he believes architecture needs to support in the future.)
Human-centered design has been the dominant discipline for decades. It taught us to start with the user, to map journeys, and to design systems that are intuitive for human senses and human timing. This worked because humans were always the primary actors. Even when automation was present, it was subservient to human intent.
Agentic systems inhabit a different physics. They do not see what we see, they do not operate on human time, and they do not rely on human meaning. Treating them as fast humans with APIs attached creates friendly wrappers around brittle cores. The design looks elegant on a whiteboard but fails in the wild because the system behaves in ways the designers never anticipated.
The familiar playbook breaks for three reasons:
- Perception gap: Agents perceive data directly, not through human senses. They register signals invisible to us and ignore cues we consider obvious. A maintenance agent may act on vibration data long before a human would hear a strange noise. The cues that matter to us are irrelevant to it, while the cues it relies on may not be legible to us at all.
- Semantic gap: Agents interpret meaning through statistical and symbolic models, not through lived experience. Their understanding is structured differently from ours. A human sees a customer complaint and recognises emotion, tone, and intent. An agent sees vectors, tokens, and probabilities. The translation works most of the time, but when it fails, it fails without warning.
- Temporal gap: Agents act on machine time, which is often instantaneous and continuous. They do not wait for human cycles of attention. A trading agent can make thousands of micro-decisions in the time it takes a human to blink. This creates both opportunity and risk: outcomes compound before oversight can intervene.
These gaps mean that design cannot remain a human-first discipline with agents bolted on. It must be reframed as agentic design: systems conceived from the start as environments where agents act, and only afterwards adapted to include human touchpoints.
Jesper Lowgren is a leading voice in enterprise architecture and the emerging discipline of agentic AI. He develops frameworks and methods that help business and technology leaders design, govern, and scale intelligent systems where autonomy and emergence reshape risk and opportunity.