On May 2026, Monday.com stopped selling seats. It began selling trust in a meter—and then it cut 20% of its workforce to prove it was serious. The market bounce of 12.6% after a 50% drawdown was read as bad news priced in, new story confirmed. The ledger remembers what the narrative forgets: a pricing-model change is not a feature launch. It is a re-architecture of the entire revenue engine, whether the chart says so or not.
Monday.com’s move from Work OS to AI Work Platform is the kind of rebranding that produces press releases and confident board decks. Internally, it is something else. The platform now connects natively to Anthropic, OpenAI, and Microsoft. Non-technical teams can configure AI agents with one-click connectors. The per-seat subscription tiers have been overlaid with a metered allotment: 1,000 credits on Basic, 2,000 on Standard, 3,000 on Pro, with overages at $0.01–$0.0125 per credit. Monthly billing carries a 25% premium over annual prepayment. Translation: prepay, or pay more. That is a cash-flow mechanism dressed as a customer discount.
To understand what this means, reconstruct the protocol from first principles. The old Monday.com revenue equation was simple: users × price − near-zero server cost = margin. Software is expensive to build but cheap to run. That is why SaaS companies trade like annuity contracts. The new equation is different. Users get the right to invoke agents, and every invocation has a real marginal cost: token inference, external API calls, tool execution, data transfer, and the orchestration layer in between. Monday.com is not selling a product anymore. It is reselling compute wrapped in workflow logic. That puts the company closer to a utility than a software vendor. Utilities do not get SaaS multiples unless they can prove the meter is honest.
The technical complexity of that meter is the part that sells for free. Based on my audit experience with ledger infrastructure and metering systems, the hard problem is not the AI model. It is provenance. Every credit has to be assigned to a tenant, an agent, a workflow, a model vendor, and a timestamp. It has to survive billing reconciliation, refund disputes, fraud attempts, and the inevitable customer question: why did my credits drain overnight? Without a precise metering engine, the AI Work Platform is not a platform; it is an accounting accident waiting to be audited.
This is why Monday.com will spend the next twelve months building infrastructure that investors cannot see. The platform connects to Anthropic, OpenAI, and Microsoft, so it needs a multi-model abstraction layer. That abstraction has to be invisible to users, unified across model APIs, and semantically precise enough to process the same instruction through different providers without breaking state. Behind that layer sits an agent runtime with workflow orchestration, task state, retry policy, and permission checks. And on top of that sits the credit engine: a metering and billing system that maps model calls, tool invocations, and data throughput into a single unit of value. This is not a project management tool. It is a mini cloud-billing platform with a UI skin.
The real unit economics are invisible in the announcement. Traditional SaaS gross margins sit in the 75–85% range. Metered AI credits have a direct cost attached to every consumed unit. If model inference and tool calls consume 30–50% of the credit price, a reasonable assumption for a reseller of OpenAI, Anthropic, or Microsoft APIs, then blended margins fall to 60–65%. The more successful the AI platform becomes, the more low-margin revenue dilutes the company-level gross margin. The market narrative treats AI as a premium add-on. The ledger treats it as a pass-through risk. Stability is not a feature; it is a discipline. The credit meter must be designed with the same rigor as a consensus protocol, because a small rounding error in virtual price calculations, the kind I spent weeks chasing in 2020, will not surface during the demo. It will surface in customer billing disputes at scale.
There is also the question of revenue recognition. Are prepaid credits counted as ARR before they are consumed? The 25% discount for annual prepayment suggests the company wants cash in advance. That shifts the risk of non-consumption from Monday.com to the customer, and it creates an incentive to book consumption as quickly as possible. If the company books the credit sale on delivery, then ARR is inflated by stored value. If it books on consumption, ARR is deferred and the reported number will look weaker for a few quarters. The market hates ambiguity. A serious AI platform should disclose both: subscription revenue and credit consumption revenue, with a liability line for unearned credits. Protecting the user means asking for that disclosure before the next earnings call.
The pricing model creates a second-order problem for customers: budget forecasting. A seat-based contract takes one variable: how many people. A credit-based contract requires the customer to predict agent behavior, task volume, and inefficiency. That is not a trivial calculation. The sales cycle will lengthen because procurement needs a usage model, not a user count. The 20% layoff makes this worse. The company cut 620–630 people at the same moment it introduced a product that requires more customer education, not less. The remaining customer success team must morph from software trainers into AI workflow consultants. If they cannot, adoption stalls and the credit burn stays low.
The sales complexity also transforms the growth engine. Monday.com was historically a product-led growth company: try it, invite your team, expand by seats. AI credits add a value-selling layer. A free user can now receive credits to experience an agent completing real work. That can be a stronger funnel than a two-week time-limited trial. But the funnel only works if the credit math is calibrated so generously that users feel value without supporting runaway cost. It is a tightrope. Too generous, and the unit economics break. Too stingy, and the trial becomes a bill instead of a demo.
There is a structural paradox hidden in the model. AI efficiency is supposed to improve over time. As models get better and agents get optimized, the same task will consume fewer credits. That is great for the customer and bad for a consumption-based revenue model. In traditional SaaS, efficiency improves retention. In metered AI, efficiency reduces the quantity consumed. This is the same paradox that clouds the entire usage-based AI industry: the better the product works, the less often you pay for it. Monday.com needs to anchor value to business outcomes, not to model calls. If it prices by outcomes, it can absorb efficiency gains. If it prices by credits, it will face a slow, structural leak in expansion revenue.
The lock-in story has changed too. In the Work OS era, switching costs were moderate: export your columns, reassign stakeholders, redraw your workflows. In the AI Work Platform era, switching costs are much higher. A configured agent contains decision logic, prompt chains, tool permissions, and data flows that were tuned against Monday.com’s native runtime. Moving to a competitor means re-debugging every agent. That is a deep moat, but it is also a new liability in the sales process. Enterprise buyers know that lock-in is a one-way door. They will ask for configuration exportability, data portability, and zero-retention guarantees from the external model providers before they let their procurement teams sign.
The market frames the competitive battle as Monday.com against Asana, Notion, ClickUp, or the rest of the Work OS crowd. That is the wrong frame. The structural threat is upstream. Monday.com is a model-neutral connector: it connects to Anthropic, OpenAI, and Microsoft. That is a comfortable spot until one of those model providers decides to absorb the workflow layer. Microsoft is already both partner and competitor. OpenAI and Anthropic are building enterprise agent orchestration features. If they pull that layer into their own product, Monday.com becomes a thin coordination layer inside someone else’s stack. The multi-model connector strategy manages short-term capability gaps but does not guarantee long-term strategic independence.
The data side is the quiet killer. Enterprise workflow data contains customer records, pricing, headcount decisions, and internal strategy. Sending that data to third-party model APIs is not a technical decision; it is a governance decision. Without zero-retention agreements, private model options, or regional deployment, enterprise customers will only automate low-risk tasks with AI agents. That is a direct constraint on credit burn. The revenue engine depends on trust. The coding is easy compared to the policy problem.
What should the market watch next? The 12.6% bounce was narrative relief. The real test is in the 10-Q and in the product telemetry. Does Monday.com disclose credit revenue separately? Does it break down the gross margin on AI credit consumption? Does it report active agent retention, average credits consumed per active agent, and the ratio of prepaid but unconsumed credits to total billings? If those numbers are transparent, the market can price the transition. If they are not, the stock is trading on a rebrand. The ledger remembers what the narrative forgets.
AI Work Platform is a genuine strategic pivot, but the execution risk is not in the model quality. It is in the discipline of metering, the quality of revenue recognition, and the willingness to build trust with enterprise customers before every task is automated. The question for Monday.com is not whether the AI works. It is whether the meter can be audited. I want to see the meter before I believe the 12.6%.