Monday.com's AI Credit Meter Looks Like a Token Issue. The Burn Model Is the Real Defect.

On 2026-05, Monday.com deliberately removed the most valuable feature of SaaS pricing: predictability. It replaced pure seat-based subscriptions with a hybrid model. Basic plans include 1,000 AI credits. Standard includes 2,000. Pro includes 3,000. Overages cost $0.01 per credit on monthly billing, or $0.0125 on annual billing — a 25% premimum for flexibility. The market interpreted the move as a pivot. The stock, already down more than 50% since the start of the year, jumped 12.6% on the announcement. That jump is a category error. There is no new product substance here. There is a metering system wearing a token costume. Monday.com spent a decade selling a Work OS. It now brands itself an AI Work Platform. That is more than a label change. The product is moving from representing work to executing it. Native AI agents connect to Anthropic, OpenAI, and Microsoft in one click. Non-technical users can configure workflows without touching code. The company claims more than 250,000 enterprise customers. To pay for this transformation, it fired 20% of its workforce — 620 to 630 people — and reported a restructuring charge between $45 million and $55 million. Management reaffirmed 19-20% revenue growth. The CEO says the layoffs adapt the company to the new vision. On paper, this is coherent. In practice, the pivot's success is not determined by agent quality. It is determined by whether the credit ledger can be trusted. The AI credit is a token. It is not a blockchain token, but it is an economic token. It lives on a closed ledger owned by Monday.com. The company controls minting, pricing, expiry, and discounting. There is no external audit, no on-chain signature, no burn event. Enterprise customers receive a number in a database. That number is the unit of account for the entire AI platform. I have spent enough hours auditing smart-contract metering systems to know exactly where this breaks. The first failure is meter accuracy. A credit is not a single physical quantity. It is a composite claim on model inference, cache hits, compute time, tool invocations, and data throughput. The published pricing table states $0.01 per credit. It does not state the conversion function. That is the arithmetic rounding error of 2026. In my 2017 audit of a liquidity-pool contract, the failure was an apparently minor rounding error in a dynamic fee formula. Under high volatility, it drained 15% of early investor funds. The developers called it negligible. The market later proved otherwise. Credit systems fail the same way. If the conversion function between external model costs and internal credits changes after a customer signs, the customer has no anchor for unit economics. They are buying an abstraction with a floating exchange rate. The second defect is the gross margin bridge. Legacy SaaS runs at 75-85% gross margin because marginal service cost approaches zero. AI credits carry a direct marginal cost. If Monday.com is reselling inference from Anthropic or OpenAI, model cost may consume 30-60% of a credit's price. The blended gross margin drops toward 60-65%. Scale does not fix this. Traditional SaaS benefits from network effects that drive marginal cost toward zero. Every AI inference has a non-zero bill. A bigger customer base brings bulk discounts, but never a zero-MRB. Investors expecting historical SaaS margins in this model are not reading the cost curve. The third defect is the efficiency paradox. AI credits impose a consumption tax on work. In classic utility economics, the provider profits when consumption increases. In AI, every efficiency improvement means the same task consumes fewer credits. This is structurally deflationary. If the agents become 20% more efficient, the customer's prepaid credit balance lasts 20% longer. The next renewal is smaller. The revenue per workflow drops. That is the opposite of the seat model, where a user pays every month regardless of productivity. In the metered model, improvement in the product reduces the client's required spend. Unless the company raises the per-credit price faster than efficiency gains, revenue per enterprise user is a diminishing variable. This flaw is not visible in the 19-20% growth guidance. There is also an accounting ambiguity. Prepaid credits are not automatically recurrent revenue. If credits are recognized when purchased, the income statement looks strong, but the balance sheet hides an obligation. If they are recognized upon consumption, guidance becomes a burn-rate forecast. The market has not been told which one applies. That ambiguity is precisely the kind of unspecified conversion function that breaks enterprise-trust models. The 12.6% bounce suggests the stock market accepted a narrative without reading the meter. Trust the hash, not the hype. The bulls have a defensible answer. The shift from a system of record to a system of action is real. Agents configured inside a work platform generate a rich data flywheel. 250,000 customers produce workflow histories, agent failure patterns, and user corrections. That corpus is a durable asset. Monday.com does not need to own a foundation model. It needs to be the operational layer where models get injected into business process. Multi-model connectors reduce dependency on any one lab. The credit model also enables a stronger product-led motion: a free user receives 500 credits, not fourteen days of access to everything. Value gets demonstrated through execution. If the meter is honest and the agent economics are sane, this is a legitimate transition from subscription software to utility-as-a-service. The contrarian read is not all soft. Switching costs rise significantly. A client with thirty configured agents cannot depart without reimplementing every workflow. That lock-in is deeper than traditional data migration. But the same lock-in raises the stakes for trust. Almost no enterprise software relationship requires the customer to accept a proprietary credit token as the only measure of work completed. That is why the comparison to blockchain protocols is instructive. A token with an opaque mint function and no verifiable burn is not a utility. It is an IOU with brand equity. So the question is not whether Monday.com can build AI agents. It already built the interface. The question is whether its AI currency deserves institution-level trust. I want to see a public gross-margin bridge, a defined recognition policy for prepaid credits, and a response to the efficiency paradox. If an agent improves by 20%, does the enterprise pay less? If yes, revenue per customer is structurally deflationary. If no, the pricing is disconnected from the meter. Debug the intent, not just the code. The old Work OS was a system of record. The new AI Work Platform is a system of action. The only thing that matters is whether the action is measured honestly. Volatility is the tax on uncertainty. This transition will create volatility. The uncertainty is not about AI. It is about the ledger.