Negative-Carry Features
Negative-Carry Features are software capabilities whose ongoing maintenance and compute costs exceed their revenue contribution.
“A feature that costs more to keep alive than it generates in customer value is not an asset; it is an unhedged financial liability.”
In the era of AI and token-based billing, features no longer have fixed marginal costs. A feature that costs 50 dollars per customer per month in LLM tokens but is bundled into a 30 dollar subscription creates negative unit economics at scale.
Multi-Hop Causal Traversal Engine
Explore how concepts dynamically feed into each other across 1-hop, 2-hop, and 3-hop transitive relationships. Click any node to navigate the causal highway.
Negative-Carry Features
Negative-Carry Features are software capabilities whose ongoing maintenance and compute costs exceed their revenue contribution.
Direct Relationships (4)
Transitive Neighbors (Connected via Hop 1)
Extended Causal Ripple Effects
Richard Ewing’s Research Thesis
We must enforce feature-level unit economics to protect enterprise SaaS gross margins.
Why This Specification Exists
Generative AI features with variable token consumption destroy SaaS gross margins.
Treating all engineering work as fixed OPEX.
No methodology for calculating individual feature gross margin contribution.
Negative-Carry Feature analysis linking usage COGS to subscription revenue.
What Changes If You Believe This?
Engineers optimize code paths for token efficiency and semantic cache hit rates.
Surfaces accurate gross margin contribution per feature for board reporting.
PMs evaluate financial sustainability before launching compute-intensive features.
Implements rate limits to prevent algorithmic denial-of-wallet attacks.
Recommended Action by Role
Mandate feature-level gross margin reporting to prevent variable inference expenses from outpacing fixed subscription revenue.
Transition compute-intensive capabilities from flat-rate subscription tiers to usage-based quotas and tiered volume pricing.
Instrument telemetry to measure per-query token and vector search costs across every customer account.
Implement semantic caching and smaller specialized models to push feature operating costs below contract price thresholds.
AUEB Framework
Calculates AI Unit Economic Breakeven across feature portfolios.
Latest Publications & Research Activity
The Software Factory Is Running 24/7 (And Nobody Wants the Output)
When foundational models become hyper-cheap and agentic tools run mouse and keyboard actions 24/7, code generation outpaces human review capacity by orders of magnitude. The inflation-deflation loop floods companies with synthetic work that nobody requested, shifting true enterprise value from feature production to ruthless deprecation, product discovery, and human boundary control.
The Engineering Bottleneck Illusion: What Copilot Adoption Taught Us
Typing code was never the primary constraint in software engineering. When enterprises deploy AI coding assistants like GitHub Copilot, they do not eliminate system bottlenecks, but shift them downstream into code review traffic jams, security and architectural drift, and staging validation delays. To capture real economic ROI, engineering leaders must measure deployment lead time, review cycle time, and defect escape rate, bounded by automated runtime allowlists and deterministic state checks.
Cursor vs Google Antigravity for Production AI Building
Examining the operational shift from unconstrained conversational AI coding assistants (like Early Cursor) to structured development environments (Google Antigravity). By enforcing immutable root rule files, modular step-by-step execution, and terminal-level zero-trust type verification, context loss incidents dropped by over 90% and debugging overhead was reduced from hours to minutes during the production engineering of Exogram.ai and CareerWin.ai.
Most Companies Shouldn’t Be Using Autonomous Coding Agents Yet
The technology is getting ahead of the environments we are putting it in. Autonomous coding agents operating in shared environments create investigation and cleanup bottlenecks that erase productivity. Before increasing agent autonomy, engineering teams must establish strict boundary controls, autonomous verification loops, and failure recovery harnesses.
Frequently Asked Questions
Q:What is a Negative-Carry Feature?
A software capability whose direct compute, API, and engineering maintenance costs exceed the customer revenue it generates.
Q:How do you fix a Negative-Carry Feature?
Through semantic caching, model right-sizing, rate limiting, moving to usage-based pricing, or executing the Sunset Protocol.
Canonical Specification Origin
Feature carrying costs must not exceed revenue contribution.
Corpus Interconnections
Richard Ewing artifacts developed around this canonical framework, including publications, execution tools, and diagnostic models.
External Adoption & Peer Citations
Documented instances where independent researchers, engineering teams, and publications have cited, implemented, or referenced this concept outside Richard Ewing’s ecosystem.
External Evidence: No independently verified references recorded yet.
This concept is part of Richard Ewing’s original baseline canon. External citations and implementations are added only upon rigorous empirical verification.
Inspectable Evidence Ledger
Classified evidence items supporting, extending, or refining this canonical research specification.
Recommended Citation
Ewing, R. (2026). "Negative-Carry Features." Richard Ewing Research Canon. Available at: https://www.richardewing.io/concepts/negative-carry-features
@article{ewing_negative_carry_features,
author = {Ewing, Richard},
title = {Negative-Carry Features},
journal = {Richard Ewing Research Canon},
year = {2026},
url = {https://www.richardewing.io/concepts/negative-carry-features}
}