Computational Work Conservation
Abstract
Section titled “Abstract”This series has argued that artifacts represent units of computational work and that artifact graphs encode the structure of that work. This final note derives the principle that follows from these observations.
Computational systems produce artifacts representing completed computational work. When these artifacts persist, the work required to produce them remains available to future computation. When they are lost, the work must be performed again.
This document introduces the principle of Computational Work Conservation, which states that completed computational work persists only through the artifacts produced by that work.
Systems that preserve artifact availability therefore accumulate computational work over time, while systems that lose artifacts repeatedly destroy and recreate previously completed computation.
Recognizing this principle reframes artifact preservation not as a storage concern, but as a fundamental requirement for computational systems that seek to retain and extend the results of prior work.
Opening Observation
Section titled “Opening Observation”Modern computational infrastructure routinely destroys the results of computation.
Processes terminate.
Temporary outputs disappear.
Intermediate results are discarded.
When artifacts produced by computation are not preserved, the work required to produce them must be repeated.
In many modern systems, recomputation is treated as normal.
However, recomputation is not an inherent property of computation itself.
It is a consequence of lost computational artifacts.
Computational Work and Artifact Graphs
Section titled “Computational Work and Artifact Graphs”Previous notes established that artifacts represent units of computational work and that artifact graphs encode the structural relationships between those artifacts.
Taken together, these graphs represent the accumulation of computational work performed by a system.
The Loss of Computational Work
Section titled “The Loss of Computational Work”As discussed earlier in this series, traditional storage systems do not preserve artifact graphs across distributed computation. When artifacts disappear, the computational work represented by those artifacts is lost and must be recomputed.
The loss of artifacts does not merely remove stored data — it erases previously completed computational work.
The Principle of Computational Work Conservation
Section titled “The Principle of Computational Work Conservation”If artifact graphs represent computational work and losing artifacts destroys those graphs, a general principle follows.
Computational Work Conservation
Section titled “Computational Work Conservation”Completed computational work persists only through the artifacts produced by that work.
Systems that preserve artifact availability accumulate work over time. Systems that lose artifacts repeatedly destroy and recreate previously completed computation.
Artifact Availability and Work Preservation
Section titled “Artifact Availability and Work Preservation”The Artifact Availability Layer introduced earlier in this series provides the infrastructure required to preserve artifact graphs across distributed computational environments and maintain the results of prior computation.
When artifacts remain available:
- computational lineage remains intact
- derived artifacts remain reproducible
- prior computation remains reusable
Preserving artifact availability therefore preserves the structure of computational work.
Over time, artifact graphs evolve into durable representations of accumulated computation.
Implications for Computational Infrastructure
Section titled “Implications for Computational Infrastructure”When artifact graphs persist across systems and time, computational systems gain a new property.
They can accumulate computational work.
Each new artifact extends the graph of prior computation.
Future systems can build upon previously completed results rather than repeating the same work.
In large distributed ecosystems this property becomes increasingly important.
Autonomous systems continuously generate artifacts representing analysis, synthesis, transformation, and reasoning.
Preserving these artifacts allows computational ecosystems to extend and refine prior computation rather than recreating it.
Recomputation as a Symptom
Section titled “Recomputation as a Symptom”Much of today’s computational infrastructure implicitly assumes recomputation.
Data pipelines rerun.
Model training restarts.
Analyses are repeated.
In many cases, recomputation is not necessary because the computational work has already been performed.
The true cause of repeated computation is often the loss of artifacts that represent prior work.
Recognizing this distinction clarifies that recomputation is often not a computational necessity but a consequence of lost artifacts.
A Shift in Infrastructure Thinking
Section titled “A Shift in Infrastructure Thinking”Historically, computing infrastructure evolved to preserve data.
Filesystems persist documents.
Databases persist application state.
Object storage persists digital assets.
Autonomous computational systems introduce a new requirement.
Infrastructure must preserve computational work itself, not merely the data it produces.
Artifacts represent the results of computation.
Artifact graphs represent the structure of that work.
Preserving artifact availability therefore preserves the work performed by computational systems.
While these implications apply broadly to computational infrastructure, they are particularly significant in agent ecosystems where artifacts are continuously produced and reused.
Implications for Agent Ecosystems
Section titled “Implications for Agent Ecosystems”In distributed agent ecosystems, many independent systems may perform related computation.
Without preserved artifacts, these systems repeatedly reproduce similar intermediate work.
When artifact availability is preserved, artifacts produced by one system can become inputs to future computation performed by others.
Over time, computational ecosystems begin to behave less like isolated processes and more like collaborative systems extending a shared body of computational work.
Conclusion
Section titled “Conclusion”Computational systems produce artifacts that represent the results of completed work.
Artifact graphs capture the structure of that work across chains of computation.
When artifacts are preserved, computational work persists.
When artifacts are lost, computational work disappears and must be performed again.
Computational systems that preserve artifact availability therefore accumulate work over time, while systems that fail to preserve artifacts repeatedly destroy and recreate their own computational history.
Recognizing and preserving artifact availability allows computational ecosystems to evolve through the accumulation of prior computation.
In this way, computational systems become capable not only of performing work, but of conserving and extending the work that has already been done.
This shift enables a new computational paradigm: Cumulative Computing.
In cumulative computing systems, computation produces artifacts that persist, form artifact graphs, and enable future computation to build upon prior work.
Agent Artifact Availability (AAA) defines the reliability property required to support such systems.
Discussion and Feedback
Section titled “Discussion and Feedback”The ideas presented in this document conclude the initial exploration of Agent Artifact Availability (AAA) and the architectural implications of preserving computational artifacts in distributed systems.
Comments, critiques, and alternative perspectives are encouraged.
The broader paradigm emerging from this series is explored in Cumulative Computing.
Citation
Section titled “Citation”If referencing this work, please cite:
Kopcho, Rich. Computational Work Conservation.
Agent Artifact Availability (AAA) Series. Technical Note, March 2026.