Writing
Software Creatures: Imagining the Next Evolution of Systems
Software has always looked like infrastructure: services, jobs, databases, APIs, dashboards. Things you deploy and keep alive. Agentic systems may end up looking stranger than that. They form around a task, gather tools, act for a while, leave artifacts behind, and then disappear — or mutate into the next workflow.
Why "creature" is the right word
I keep reaching for the creature metaphor because nothing in the infrastructure vocabulary fits. A creature-like system has goals, memory, perception, a budget it metabolizes, and behavior that adapts to its environment. It is not alive, obviously. But it behaves less like a static machine and more like a bounded organism, and our operational instincts — built over decades of keeping stable services healthy — do not transfer cleanly.
Traditional services are deployed and monitored. Creatures are commissioned, supervised, and eventually dissolved. The operational contract changes shape: lifespan, resource budget, tool scope, success criteria, drift checks, and a way to snapshot what the creature learned before it goes away. The runtime has to manage behavior, not just uptime. Emergent behavior becomes an operations problem, not only a model-quality issue.
The control primitives
If you squint, a small set of primitives covers most of it. Lifespan: one-shot, scheduled, long-running, or event-triggered. Memory: none, task-local, project-local, or durable. Tooling: read-only, staged writes, approved writes, or — carefully — autonomous writes. Containment: sandbox, rate limit, budget, policy gate, kill switch. Every creature gets explicit boundaries, a metabolism, and death conditions: what it may consume, what it may do, and what it may leave behind. The artifacts and audit trails persist even when the creature does not.
This is speculation, and the metaphor will break somewhere. But if software does become more creature-like, engineering has to get better at something closer to ecology than administration: containment, observation, incentives, and controlled evolution. We spent decades learning to keep services alive. The next skill might be deciding what gets to live, for how long, and what it is allowed to eat.