Volume I · Measure value, not activity.

The Economics of AI-Assisted Software Engineering

Teams optimize token bills while the real costs — context rebuilding, review time, rework — stay invisible. This volume builds the economic vocabulary of the discipline, from Session ROI to cost per merged outcome, on the honest evidence base.

Chapters

Published

01The Real Cost of AI Coding Isn't Your Token BillTeams optimize token billing rather than engineering value. Framework: the AI Engineering Pyramid.9 min readPublished

02The Productivity ParadoxThe METR pair, cited as a pair: slower while feeling faster — and what changed by 2026.8 min readPublished

03Context DebtLost conversational context as a form of debt slowing teams. Framework: the Context Lifecycle.8 min readPublished

04Session ROIValue created vs. time and cost invested, at the session level.7 min readPublished

05The AI Cost IcebergVisible costs above the waterline; context rebuilding, review and rework below.7 min readPublished

06Cost Per OutcomeCost per merged feature as the unit metric — and why harness design moves it 3–4×.7 min readPublished

07The Review BottleneckMore PRs, flat delivery metrics, longer reviews: downstream verification absorbs generation gains.8 min readPublished

08The Quality LedgerDuplication up, refactoring collapsing: AI code as unsecured debt — with the critiques adjudicated.8 min readPublished

09Engineering SignalsEvery AI session is a source of data. Framework: the Telemetry Schema.7 min readPublished

10The Case for AI ObservabilityInstrumenting AI interactions: token tracking, pricing registries, unified telemetry.7 min readPublished

11From Costs to IntelligenceCapstone — the hand-off from economics to the measurement discipline.4 min readPublished

Key sources: METR 2025 RCT + 2026 update · DORA 2025 — State of AI-assisted Software Development · GitClear code-quality research · Faros AI 10k-developer dataset

Related