L I B R A R Y · E C O N O M I C S
Agent Economics
Unit economics, token costs, ROI, pricing, capital, org design, and measurement frameworks for production agentic systems.
120 articles
10 clusters
Search this series
Start here
Cost-per-completed-task is the only AI metric that belongs on a P&L
A practical operator guide to Cost-per-completed-task is the only AI…: what changes in real workflows, how to design for production, and what to measure…
Baseline or it did not happen
A practical operator guide to Baseline or it did not happen: what changes in real workflows, how to design for production, and what to measure before you scale.
Model routing as a real-time procurement decision
A practical operator guide to Model routing as a real-time…: what changes in real workflows, how to design for production, and what to measure before you scale.
Who should own AI cost and AI value inside the enterprise
A practical operator guide to Who should own AI cost and AI value…: what changes in real workflows, how to design for production, and what to measure before…
Capital Allocation & Financing Reality 13
- 01The circular deal problem: when ecosystem-building becomes manufactured demandA practical operator guide to circular deal problem: when…: what changes in real workflows, how to design for
- 02Depreciation schedules as a multi-hundred-billion-dollar accounting choiceA practical operator guide to Depreciation schedules as a…: what changes in real workflows, how to design for
- 03How to value a loss-making AI company when traditional DCF breaksA practical operator guide to value a loss-making AI company when…: what changes in real workflows, how to des
- 04The revenue gap that has to close for the current AI capex cycle to pay offA practical operator guide to revenue gap that has to close for the…: what changes in real workflows, how to d
- 05Why neoclouds exist: arbitraging other companies’ utilization and depreciation uncertaintyA practical operator guide to Why neoclouds exist: arbitraging other…: what changes in real workflows, how to
- 06Free-cash-flow compression at the hyperscalers: what it signalsA practical operator guide to Free-cash-flow compression at the…: what changes in real workflows, how to desig
- 07The difference between committed compute and funded demandA practical operator guide to difference between committed compute…: what changes in real workflows, how to de
- 08GPU residual value risk: the number almost no one modelsA practical operator guide to GPU residual value risk: the number…: what changes in real workflows, how to des
- 09The difference between AI capex and AI opex in board-level decisionsA practical operator guide to difference between AI capex and AI opex…: what changes in real workflows, how to
- 10Vendor financing in AI: lessons from previous technology cyclesA practical operator guide to Vendor financing in AI: lessons from…: what changes in real workflows, how to de
- 11How to stress-test an AI infrastructure investment caseA practical operator guide to stress-test an AI infrastructure…: what changes in real workflows, how to design
- 12The balance-sheet difference between owned GPUs and committed cloud spendA practical operator guide to balance-sheet difference between owned…: what changes in real workflows, how to
- 13How circular financing affects the cost of capital for everyone elseA practical operator guide to How circular financing affects the cost…: what changes in real workflows, how to
Geopolitics & Supply Decisions for Business 11
- 01The US–China capability gap is now measured in months, not yearsA practical operator guide to US–China capability gap is now measured…: what changes in real workflows, how to
- 02Export controls: what they actually achieved and what they did notA practical operator guide to Export controls: what they actually…: what changes in real workflows, how to des
- 03Sovereign AI: why every country wants its own stack and why almost all of them still run on NVIDIAA practical operator guide to Sovereign AI: why every country wants…: what changes in real workflows, how to d
- 04Talent flows have reversed: what that means for capability buildingA practical operator guide to Talent flows have reversed: what that…: what changes in real workflows, how to d
- 05Open-weight models as a strategic supply-chain decisionA practical operator guide to Open-weight models as a strategic…: what changes in real workflows, how to desig
- 06Concentration risk in the AI supply chain: one foundry, one software stack, a handful of cloudsA practical operator guide to Concentration risk in the AI supply…: what changes in real workflows, how to des
- 07How to build an AI capability strategy that is robust to export-control and geopolitical shiftsA practical operator guide to build an AI capability strategy that is…: what changes in real workflows, how to
- 08Sovereign AI initiatives as both market opportunity and competitive threatA practical operator guide to Sovereign AI initiatives as both market…: what changes in real workflows, how to
- 09Building AI capability under technology-transfer constraintsA practical operator guide to Building AI capability under…: what changes in real workflows, how to design for
- 10The talent-location decision in a world of reversed flowsA practical operator guide to talent-location decision in a world of…: what changes in real workflows, how to
- 11National AI capability as industrial policy: implications for global companiesA practical operator guide to National AI capability as industrial…: what changes in real workflows, how to de
Measurement, Governance & ROI 11
- 01The only ROI questions a CFO should acceptA practical operator guide to only ROI questions a CFO should accept: what changes in real workflows, how to d
- 02Why governance is now a growth multiplier, not just a costA practical operator guide to Why governance is now a growth…: what changes in real workflows, how to design f
- 03Agentic AI is breaking traditional ROI models — here is the fixA practical operator guide to Agentic AI is breaking traditional ROI…: what changes in real workflows, how to
- 04Baseline or it did not happenA practical operator guide to Baseline or it did not happen: what changes in real workflows, how to design for
- 05The governance minimum viable product for agentic systemsA practical operator guide to governance minimum viable product for…: what changes in real workflows, how to d
- 06Leading vs lagging indicators for AI valueA practical operator guide to Leading vs lagging indicators for AI value: what changes in real workflows, how
- 07The four categories of AI value that most measurement systems missA practical operator guide to four categories of AI value that most…: what changes in real workflows, how to d
- 08Why “AI-generated productivity” claims need a counterfactualA practical operator guide to Why AI-generated productivity claims…: what changes in real workflows, how to de
- 09The one-page AI initiative contractA practical operator guide to one-page AI initiative contract: what changes in real workflows, how to design f
- 10Leading indicators that predict AI program success or failureA practical operator guide to Leading indicators that predict AI…: what changes in real workflows, how to desi
- 11The difference between AI activity metrics and AI impact metricsA practical operator guide to difference between AI activity metrics…: what changes in real workflows, how to
Moats, Value Capture & Industry Structure 14
- 01Foundation models are commoditizing: where the actual moat movedA practical operator guide to Foundation models are commoditizing:…: what changes in real workflows, how to de
- 02The wrapper test: three things that create real defensibilityA practical operator guide to wrapper test: three things that create…: what changes in real workflows, how to
- 03NVIDIA’s real moat is software and systems, not siliconA practical operator guide to NVIDIA’s real moat is software and…: what changes in real workflows, how to desi
- 04Mapping the AI value chain: where durable value actually accumulatesA practical operator guide to Mapping the AI value chain: where…: what changes in real workflows, how to desig
- 05Vertical AI vs horizontal platforms: where the business-model advantage sitsA practical operator guide to Vertical AI vs horizontal platforms:…: what changes in real workflows, how to de
- 06Porter’s Five Forces applied to the AI inference marketA practical operator guide to Porter’s Five Forces applied to the AI…: what changes in real workflows, how to
- 07Data flywheels vs model weights: which one compoundsA practical operator guide to Data flywheels vs model weights: which…: what changes in real workflows, how to
- 08Switching costs in the age of multi-model routingA practical operator guide to Switching costs in the age of…: what changes in real workflows, how to design fo
- 09Distribution as the last durable moat when models and interfaces commoditiseA practical operator guide to Distribution as the last durable moat…: what changes in real workflows, how to d
- 10The three layers where switching costs still exist in 2026A practical operator guide to three layers where switching costs…: what changes in real workflows, how to desi
- 11Why “we have better data” is the most abused claim in AI business plansA practical operator guide to Why we have better data is the most…: what changes in real workflows, how to des
- 12Open-weight strategy as competitive weapon rather than ideologyA practical operator guide to Open-weight strategy as competitive…: what changes in real workflows, how to des
- 13The “can the model provider ship this next quarter” testA practical operator guide to can the model provider ship this next…: what changes in real workflows, how to d
- 14Why integration depth is becoming more valuable than model depthA practical operator guide to Why integration depth is becoming more…: what changes in real workflows, how to
Operator Decision Frameworks 14
- 01The dual ROI lens: tangible vs strategic returnsA practical operator guide to dual ROI lens: tangible vs strategic…: what changes in real workflows, how to de
- 02The 90-day minimum-governance pilot that actually reaches productionA practical operator guide to 90-day minimum-governance pilot that…: what changes in real workflows, how to de
- 03Human vs AI cost-per-task: the only substitution arithmetic that mattersA practical operator guide to Human vs AI cost-per-task: the only…: what changes in real workflows, how to des
- 04When to build, buy or partner for AI capabilityA practical operator guide to When to build, buy or partner for AI…: what changes in real workflows, how to de
- 05The three complementary constraints that determine whether cheaper tokens helpA practical operator guide to three complementary constraints that…: what changes in real workflows, how to de
- 06The pre-mortem that prevents most AI pilot failuresA practical operator guide to pre-mortem that prevents most AI pilot…: what changes in real workflows, how to
- 07Make vs buy for fine-tunes and agents: a simple decision treeA practical operator guide to Make vs buy for fine-tunes and agents:…: what changes in real workflows, how to
- 08The kill criteria every AI project needs before it startsA practical operator guide to kill criteria every AI project needs…: what changes in real workflows, how to de
- 09The minimum viable measurement system for an AI pilotA practical operator guide to minimum viable measurement system for…: what changes in real workflows, how to d
- 10How to decide the right level of autonomy for a given workflowA practical operator guide to decide the right level of autonomy for…: what changes in real workflows, how to
- 11The “who owns the failure” test for AI systemsA practical operator guide to who owns the failure test for AI systems: what changes in real workflows, how to
- 12The decision rights map for human-agent systemsA practical operator guide to decision rights map for human-agent…: what changes in real workflows, how to des
- 13How to set the sampling rate for human reviewA practical operator guide to set the sampling rate for human review: what changes in real workflows, how to d
- 14The “default to no” rule for new AI surface areaA practical operator guide to default to no rule for new AI surface area: what changes in real workflows, how
Organizational Design & Adoption Reality 13
- 01Why enterprise AI pilots stall before production — the unglamorous reasonsA practical operator guide to Why enterprise AI pilots stall before…: what changes in real workflows, how to d
- 02AI-native vs AI-enabled companies: why structure is differentA practical operator guide to AI-native vs AI-enabled companies: why…: what changes in real workflows, how to
- 03The entry-level door is closing: what the labor data actually showsA practical operator guide to entry-level door is closing: what the…: what changes in real workflows, how to d
- 04Rewiring the operating model for agents, not just toolsA practical operator guide to Rewiring the operating model for…: what changes in real workflows, how to design
- 05Who should own AI cost and AI value inside the enterpriseA practical operator guide to Who should own AI cost and AI value…: what changes in real workflows, how to des
- 06The skills that rise in value when routine cognitive work is automatedA practical operator guide to skills that rise in value when routine…: what changes in real workflows, how to
- 07Why “AI literacy” programs often fail and what works insteadA practical operator guide to Why AI literacy programs often fail and…: what changes in real workflows, how to
- 08Why AI centers of excellence often become bottlenecksA practical operator guide to Why AI centers of excellence often…: what changes in real workflows, how to desi
- 09The new division of labor between humans and agentsA practical operator guide to new division of labor between humans…: what changes in real workflows, how to de
- 10Funding models that match the way AI value actually appearsA practical operator guide to Funding models that match the way AI…: what changes in real workflows, how to de
- 11The career path problem created by task automationA practical operator guide to career path problem created by task…: what changes in real workflows, how to des
- 12Why “AI transformation” programs often produce tools without changing powerA practical operator guide to Why AI transformation programs often…: what changes in real workflows, how to de
- 13Performance management in an AI-augmented workforceA practical operator guide to Performance management in an…: what changes in real workflows, how to design for
Pricing & Monetization Models 4
- 01Token-based pricing is forcing software to look like electricityA practical operator guide to Token-based pricing is forcing software…: what changes in real workflows, how to
- 02Outcome-based pricing: when it works and when it is a trapA practical operator guide to Outcome-based pricing: when it works…: what changes in real workflows, how to de
- 03Why hybrid pricing is winning and pure consumption pricing often failsA practical operator guide to Why hybrid pricing is winning and pure…: what changes in real workflows, how to
- 04The pricing power of workflow embedding vs the pricing power of model qualityA practical operator guide to pricing power of workflow embedding vs…: what changes in real workflows, how to
Risk, Accounting & Structural Fragility 11
- 01What would actually falsify the AI bubble thesisA practical operator guide to What would actually falsify the AI…: what changes in real workflows, how to desi
- 02The bear case in one page: circularity, depreciation and the revenue gapA practical operator guide to bear case in one page: circularity,…: what changes in real workflows, how to des
- 03Historical infrastructure cycles: what railroads and fiber actually teachA practical operator guide to Historical infrastructure cycles: what…: what changes in real workflows, how to
- 04The difference between a technology cycle and a capital cycleA practical operator guide to difference between a technology cycle…: what changes in real workflows, how to d
- 05What “demand” means when much of it is intermediateA practical operator guide to What demand means when much of it is…: what changes in real workflows, how to de
- 06The accounting choice that can change reported AI profitability by tens of billionsA practical operator guide to accounting choice that can change…: what changes in real workflows, how to desig
- 07Correlated failure modes in the AI stackA practical operator guide to Correlated failure modes in the AI stack: what changes in real workflows, how to
- 08The difference between model risk and system riskA practical operator guide to difference between model risk and…: what changes in real workflows, how to desig
- 09The residual-value assumption that can make or break ownership economicsA practical operator guide to residual-value assumption that can make…: what changes in real workflows, how to
- 10How to tell intermediate demand from final demand in AI marketsA practical operator guide to tell intermediate demand from final…: what changes in real workflows, how to des
- 11Scenario planning for a sudden tightening of AI capitalA practical operator guide to Scenario planning for a sudden…: what changes in real workflows, how to design f
Substitution & Workforce Economics 4
- 01The productivity paradox is back: firm-level gains, macro silenceA practical operator guide to productivity paradox is back:…: what changes in real workflows, how to design fo
- 02Seniorisation of entry-level work: the real early labor-market effectA practical operator guide to Seniorisation of entry-level work: the…: what changes in real workflows, how to
- 03The tasks that are being automated vs the jobs that are being redesignedA practical operator guide to tasks that are being automated vs the…: what changes in real workflows, how to d
- 04Where human comparative advantage is actually risingA practical operator guide to Where human comparative advantage is…: what changes in real workflows, how to de
Unit Economics & Cost Architecture 25
- 01Cost-per-completed-task is the only AI metric that belongs on a P&LA practical operator guide to Cost-per-completed-task is the only AI…: what changes in real workflows, how to
- 02The agent multiplier: why one user request can cost 15–30× a simple chatA practical operator guide to agent multiplier: why one user request…: what changes in real workflows, how to
- 03Utilization is the only number that decides buy vs rent for GPUsA practical operator guide to Utilization is the only number that…: what changes in real workflows, how to des
- 04Model routing as a real-time procurement decisionA practical operator guide to Model routing as a real-time…: what changes in real workflows, how to design for
- 05The six hidden adders that turn list price into the real billA practical operator guide to six hidden adders that turn list price…: what changes in real workflows, how to
- 06Why inference is now 80%+ of enterprise AI spend — and what that changesA practical operator guide to Why inference is now 80%+ of enterprise…: what changes in real workflows, how to
- 07Jevons Paradox quantified: why falling token prices raise total spendA practical operator guide to Jevons Paradox quantified: why falling…: what changes in real workflows, how to
- 08The break-even volume where fine-tuning or LoRA beats heavy prompting and RAGA practical operator guide to break-even volume where fine-tuning or…: what changes in real workflows, how to
- 09Why cheaper intelligence does not automatically expand what the firm can doA practical operator guide to Why cheaper intelligence does not…: what changes in real workflows, how to desig
- 10Internal chargeback for intelligence: solving the tragedy of the commonsA practical operator guide to Internal chargeback for intelligence:…: what changes in real workflows, how to d
- 11Why token budgets must be owned by the business, not just ITA practical operator guide to Why token budgets must be owned by the…: what changes in real workflows, how to
- 12The difference between average and marginal cost in AI systemsA practical operator guide to difference between average and marginal…: what changes in real workflows, how to
- 13Caching, batching and speculative decoding as real cost leversA practical operator guide to Caching, batching and speculative…: what changes in real workflows, how to desig
- 14Why average cost per task falls and then rises again with autonomyA practical operator guide to Why average cost per task falls and…: what changes in real workflows, how to des
- 15The real cost of context: why longer windows are not freeA practical operator guide to real cost of context: why longer…: what changes in real workflows, how to design
- 16Why “cost per token” is a vendor metric and “cost per decision” is an operator metricA practical operator guide to Why cost per token is a vendor metric…: what changes in real workflows, how to d
- 17The point at which more context starts destroying unit economicsA practical operator guide to point at which more context starts…: what changes in real workflows, how to desi
- 18Batch vs real-time inference: the economic trade-off most teams ignoreA practical operator guide to Batch vs real-time inference: the…: what changes in real workflows, how to desig
- 19The hidden cost of human-in-the-loop at scaleA practical operator guide to hidden cost of human-in-the-loop at scale: what changes in real workflows, how t
- 20Why self-hosting economics improve non-linearly with utilization and specializationA practical operator guide to Why self-hosting economics improve…: what changes in real workflows, how to desi
- 21The economic case for aggressive prompt cachingA practical operator guide to economic case for aggressive prompt…: what changes in real workflows, how to des
- 22Why “cheaper model” is the wrong primary optimisationA practical operator guide to Why cheaper model is the wrong primary…: what changes in real workflows, how to
- 23The cost of evaluation and red-teaming as a permanent line itemA practical operator guide to cost of evaluation and red-teaming as a…: what changes in real workflows, how to
- 24When to pay for frontier reasoning and when to refuseA practical operator guide to When to pay for frontier reasoning and…: what changes in real workflows, how to
- 25The economics of speculative decoding and other inference optimisationsA practical operator guide to economics of speculative decoding and…: what changes in real workflows, how to d
Want this applied to your stack?
Fixed-scope pilots for AI agents and automations. Map first. Ship one real workflow. Then run it.