August 2026 Summaries

Usage-based pricing for AI agents charges customers according to consumption, such as tokens, tool calls, or completed outcomes, because agent costs can vary substantially between simple and complex tasks in ways flat seat-based pricing cannot reliably absorb. Tokens most closely reflect model-provider costs but are difficult for many buyers to understand, while calls are simpler but can obscure cost differences, and outcomes align most closely with customer value but require clear definitions and verification. Many products therefore use hybrid models combining subscriptions, included credits, and overages, tailored to different customer segments. Effective pricing depends on detailed per-run metering that attributes usage to customers, models, providers, tools, retries, and outcomes, particularly for multi-step agents that create many billable events. Companies should generally absorb costs from infrastructure-related failures and retries, account for provider pricing changes, and give customers near-real-time usage visibility to prevent unexpected invoices from becoming trust issues. The passage presents Lago’s Agent SDK as a tool for normalizing usage across AI providers and supporting flexible billing models without extensive engineering changes.

Lago’s MCP server connects Claude and other MCP-compatible AI clients to live billing data through natural-language requests, enabling users to inspect invoices, customers, payments, usage, and analytics without navigating dashboards or writing queries. Built in Rust on the open Model Context Protocol standard, it is primarily read-only, supporting data lookups, invoice simulations, usage analysis, overdue-invoice reporting, and analytics such as monthly recurring revenue, while its only permitted write action is retrying a failed payment. The open-source server can be run through Docker with a Lago API key and optionally connected to Lago’s analytics agent service, and it works with both self-hosted Lago and Lago Cloud, with self-hosting allowing organizations to keep billing data on their own infrastructure. Compared with other billing MCP offerings, Zoho provides broader write capabilities and Chargebee offers several specialized servers, while Lago emphasizes self-hostable access and narrowly limited actions.

Usage-based pricing depends primarily on selecting an appropriate billable metric rather than choosing a pricing structure first, because changing metrics later can disrupt customer workflows and contracts. Effective metrics should correlate with company costs, align with customer-recognized value, avoid discouraging beneficial product use through a “taximeter” effect, and remain predictable enough for customers to estimate bills in advance. Examples such as token-based AI pricing, SMS messages sent, and analytics events illustrate metrics tied to observable, customer-controlled activity, while per-page, per-email-open, or poorly chosen event charges can distort behavior and reduce product value. The text distinguishes billable metrics from plan-based usage restrictions, which can segment customers by scale without directly charging per unit, and notes that hybrid pricing may use multiple metrics if each meets the same criteria. It also emphasizes that billing systems must apply the correct aggregation method, deduplication, and transparent calculation rules to prevent inaccurate invoices and disputes.

Togai and Lago both convert product usage into metered, billable events, but they differ primarily in deployment, ownership, and extensibility. Togai is a managed, cloud-only, API-first platform with end-to-end metering, rating, entitlements, integrations, custom JavaScript rules, and a free starter tier, making it suitable for teams seeking rapid implementation without operating billing infrastructure. Lago is an AGPLv3-licensed open-source alternative that can be self-hosted or used through its cloud offering, allowing organizations to retain control of usage and financial data, inspect and modify billing logic, and avoid vendor lock-in. The comparison highlights Togai’s platform fees tied to event volume and invoice value versus Lago’s free self-hosted core and deal-based premium pricing, while noting that Lago may better serve regulated industries or companies with complex pricing and data-residency requirements.

AI can support revenue operations most effectively by reducing the manual work of combining CRM, billing, product-usage, and reconciliation data, particularly through natural-language cross-system queries, early anomaly detection, and drafting routine materials such as QBR preparation and renewal-risk summaries. These uses are well suited to RevOps because much of its data is structured, but human review remains important before outputs inform customer communications or leadership decisions. AI is less reliable for account-strategy judgments involving relationship context, competition, timing, or discounting, and it cannot resolve inconsistent definitions or poor data quality across source systems. Because many RevOps questions depend heavily on accurate billing, usage, invoicing, and revenue-recognition records, the usefulness of an AI layer is ultimately limited by the quality and alignment of the underlying data.

Agentic finance is presented as a gradual shift from static dashboards toward systems that can interpret questions, investigate live financial data, propose actions, and eventually automate bounded tasks with confirmation. Lago argues that the first practical finance agents should be read-only, allowing users to ask questions about billing metrics such as MRR changes, inspect underlying records, refine reports, and export results without altering invoices or subscriptions. Its Finance Assistant is separated from a Billing Assistant that can perform operational tasks, reflecting different risk levels and the need for explicit approval for consequential changes. Lago also supports an open approach through native agents and an MCP server that can connect compatible AI tools, including Claude and supported ChatGPT environments, to current billing data. The company notes that its beta Finance Assistant can make mistakes, produces point-in-time reports, and currently relies on API-key access rather than individual user permissions, emphasizing that transparency, provenance, and clearly defined permissions are essential as finance automation expands.

Stripe’s January 2026 acquisition of Metronome has prompted questions about payment-processor dependence in usage-based billing, despite Metronome retaining strengths in enterprise rate cards, contracts, spend alerts, and revenue-recognition workflows. The comparison positions Lago as an open-source, self-hostable and payment-provider-neutral alternative that supports custom metering, prepaid credits, multiple processors, and complex usage models, while acknowledging Metronome’s greater maturity in large enterprise quote-to-cash implementations. Other options include Adyen-owned Orb, which emphasizes pricing simulations and rapid alerting; Chargebee and Sequence, which focus more on finance, invoicing, contracts, and revenue recognition; Kong-owned OpenMeter, which specializes in real-time entitlement enforcement; Flexprice, a newer open-core developer-focused platform; and Stripe Billing, which suits simpler Stripe-centric billing needs. The central evaluation criteria are deployment control, processor neutrality, metering complexity, pricing transparency, finance functionality, and the potential effects of vendor acquisitions on long-term product roadmaps.

An AI finance assistant enables company finance and RevOps teams to ask plain-language questions about structured billing and financial records, such as invoices, subscriptions, revenue, churn, and usage, without needing dashboards, reports, or data queries. Unlike many search results focused on consumer budgeting tools, this type of assistant is designed for organizational financial operations and should provide transparent, inspectable evidence for its answers while generally remaining read-only. The text distinguishes assistants, which answer and explain questions for human review, from AI agents, which pursue multi-step goals and can act autonomously, creating greater potential risk when reasoning or data is incorrect. Because billing data is structured and well-defined, it is especially suitable for assistant capabilities, and teams are advised to establish trust in reliable, read-only answers before expanding toward agent functions such as automated reconciliation or anomaly detection.

AI gateways are increasingly important control points between applications or agents and the models, tools, APIs, or resources that execute requests, but the term covers several distinct functions rather than a single product category. Recent reported acquisitions and product launches, including Stripe’s potential purchase of OpenRouter and Palo Alto Networks’ acquisition of Portkey, reflect competition to control this middle layer, where providers can influence routing, policy, security, observability, cost, and access decisions. The text identifies four overlapping gateway roles: model routing across providers, traffic control and technical governance, agent access management for tools and credentials, and commercial control over what customers may consume under payment or entitlement rules. Although products in these categories are converging in features, they differ in buyers, distribution channels, business incentives, and the customer relationships they own. Lago argues that commercial control is an underdeveloped function requiring a separate layer that evaluates contracts, commitments, credits, pricing, and balances in real time, while execution gateways continue to provide usage and cost data; its integrations with services such as Cloudflare AI Gateway are intended to apply customer-specific billing policies without replacing the underlying execution infrastructure.

Generative AI in finance is most useful as a human-reviewed tool for drafting routine communications, summarizing existing information, and enabling natural-language queries over clean, structured billing and revenue data, rather than as an autonomous system for reconciliation or financial decision-making. The piece distinguishes generative AI, which produces content, from AI agents, which can take multi-step actions, and argues that marketing often blurs this difference. It identifies confident but ungrounded number generation as the most serious risk, alongside inappropriate use for context-dependent judgment calls and poor underlying data quality, which AI can amplify rather than correct. Organizations considering these tools should verify that outputs and figures can be traced to source records, assess how the system handles uncertainty, require enforced human approval for customer-facing actions, and ask vendors for examples of tasks the tool flags or refuses.

AI agents in finance are most usefully defined by their ability to autonomously complete multiple bounded steps, rather than merely provide fluent chatbot responses or display dashboard data. Current practical applications in billing and revenue operations include natural-language querying of structured revenue data, proactive anomaly detection with possible explanations, drafting customer communications such as dunning notices, and rules-based reconciliation and close checks. These tools can reduce the time required to investigate discrepancies such as invoice prorations, but they remain vulnerable to ambiguous data, false anomaly alerts, and confidently incorrect answers. Human review remains important, particularly where errors could affect customers or where senior finance judgment is needed to interpret unusual but legitimate cases. Billing data is considered a strong foundation for these systems because subscriptions, invoices, and usage events are structured and have defined relationships, enabling finance and RevOps teams to access insights without first building reports.

AI pricing requires real-time operational controls because agentic workloads can change costs rapidly through model routing, context length, tool use, retries, and execution time, making monthly invoices too late to prevent overspending. Pricing now has both a commercial role, defining subscriptions, usage, outcomes, or hybrids, and an operational role, determining whether a customer’s next request should proceed based on entitlements, budgets, forecasts, and contract terms. Products should estimate costs before runs, meter usage during execution, and apply policies near limits such as seeking approval, switching to cheaper models, using overage pools, or stopping activity. Companies across payments, developer tools, cloud data, spend management, and API gateways are competing to control this live request path, where they can influence access, cost, model selection, and payment. Usage and outcome-based pricing redistribute financial risk between vendors and customers rather than eliminating it, while hybrid models combine revenue protection with variable charges tied to value or service costs. Predictable pricing therefore does not necessarily mean fixed pricing; it depends on customers receiving understandable, actionable visibility into projected costs, consumption drivers, thresholds, and the consequences of continued usage before charges accumulate.