AI Credits Are Gas: The Monday.com Pivot Is a Tokenomics Transaction

Monday.com cut 620 people. That is 20% of the workforce. It booked a $45 to 55 million restructuring charge and wrapped a rebrand around it: Work OS becomes 'AI Work Platform.' The stock did the opposite of what fundamentals would suggest. Up 12.6%. From January highs it was down more than 50%. The market was not buying a balance sheet. It was buying a new narrative. Analysts called it a fresh start. I call it a pricing experiment dressed as a vision. Here is the technical core. Monday.com serves 225,000-plus enterprise clients. It defined the Work OS category - collaboration, project tracking, team workflows. In May 2026, it rewired the revenue model. Subscriptions remain. Seats remain. But the new layer is consumption-based. Every plan now carries a wallet of AI credits. Basic includes 1,000. Standard includes 2,000. Pro includes 3,000. Overage costs $0.01 to $0.0125 per credit. Monthly billing is priced 25% higher than annual. That small detail - the 25% premium - is the first clue. It is a prepayment discount. Monday.com wants cash upfront. That is what undercapitalized platforms do. But the larger signal is the architecture. The product is shifting from a system of record to a system of action. Old Monday.com stored what work happened. New Monday.com executes the work. Non-technical users configure agents through one-click connectors to Anthropic, OpenAI, and Microsoft. Risk moves from user error to agent error. UX priorities move from usability to controllability and explainability. This is where the industry underestimates the difficulty. An AI credit is not a marketing token. It is a metered unit backed by a real-time accounting engine. That engine must track every model inference, every tool call, every byte of data moved, and map them to billable credit units. I have audited this class of system. It is a lightweight cloud billing platform. AWS spent a decade building metering infrastructure. Monday.com is building one while shipping new AI features simultaneously. Building a metering and billing layer while maintaining an existing SaaS product is not a growth project. It is a rebuild. The 20% layoff was the organizational cost of that rebuild. Now the numbers that actually matter. Traditional SaaS gross margins run 75 to 85%. Marginal delivery cost is near zero. AI credits invert that. Every burned credit that routes through Anthropic or OpenAI carries a direct model cost. If inference consumes 30 to 50% of the credit price, the blended gross margin slides toward 60 to 65%. More AI revenue, lower aggregate margin. That is the hidden trade. Wall Street will ask for margin per credit. They will not get it in the first quarters. Code does not lie, but liquidity does. This is a familiar structure. DeFi protocols die this way. A lending app pays 15% yield in token emissions to attract liquidity, earns 8% in fees, and calls the gap growth. The ledger eventually calls it a drain. Monday.com is running the same experiment. Model API costs are the emissions. The question is binary: does a burned credit contribute more to margin than the cost of the model call behind it? If no, the product generates revenue and destroys value. The growth guidance of 19 to 20% says nothing about that math. Guidance is a promise. Margins are a fact. There is a deeper problem. Call it the AI efficiency paradox. Models improve. Better models complete the same task with less compute. That means fewer credits per completed workflow. In a subscription business, efficiency raises retention. In a metered business, efficiency lowers revenue. The meter spins slower. Crypto has seen this movie. It is the token velocity problem in a SaaS suit. When a token circulates too fast, the market needs less of it to clear the same level of activity. Price falls. When agents become more efficient, customers need fewer credits to produce the same outcome. Consumption falls. If your revenue is priced by input, your own product improvement becomes a headwind. The target user is the non-technical operator, not the developer. That is a deliberate choice. Monday.com is not building a developer API ecosystem first; it is building for the citizen developer. That widens the addressable market and lengthens the sales cycle. The buyer is not an engineer who loves tools. It is a marketing manager who fears the credit counter. That is a different customer from the one who built Monday.com's original growth. I spent 72 hours reverse-engineering the UST reserve in 2022. I watched a system whose 'stability' was a narrative, not a mechanism. The lesson that stuck: any unit of account whose supply is driven by consumption rather than by value is vulnerable to a spiral. AI credits are not stablecoins. But they share an ugly trait: liability opacity. When a customer pre-pays for 100,000 credits, Monday.com holds the cash and the customer holds a claim on future compute. When is that revenue recognized? On purchase or on burn? Unused credits are a liability on the balance sheet. If credits are counted as revenue before the underlying model call runs, that is unbacked issuance. The stablecoin playbook applies: count it when earned, not when minted. The growth shape changes too. Seats scale linearly. Credits scale along a power law. A small group of enterprise clients will burn most of the credits. In crypto we call those wallets whales. One whale churns, and the revenue line wobbles. If a handful of customers drive 20% of AI consumption, the credibility of the 19-20% growth guidance depends on procurement decisions inside five companies. That is concentration risk with a new name. The NRR question gets stranger. Subscription expansion comes from headcount growth. Consumption expansion comes from agent usage growth. But there is a countercurrent. Customers will optimize. They will watch the credit counter. They will route prompts to cheaper models. They will tune agents to burn less. The psychology shifts from how many seats do we need to how much can we avoid spending. Utility pricing forces vendors to help customers consume less. That is not a natural growth loop. It is a managed decline loop unless the vendor moves to outcome pricing. The agent runtime is the part most people skip. A serious AI Work Platform needs an abstraction layer across multiple models, state management for long-running agents, tool-call permissions, and observability into every step. This is not a CRM with a chatbot bolted on. It is orchestration infrastructure. The hardest part is cost isolation: knowing which tenant consumed which credits, at what time, through which model. Cloud providers solved this after years of enforcement. Monday.com has to ship it while also explaining it to non-technical buyers. The meter has to be both accurate and legible. Most metering systems fail at one of the two. The multi-model connector strategy deserves scrutiny. Monday.com connects to Anthropic, OpenAI, and Microsoft. That makes it a model-neutral layer. Neutral layers have no pricing power. The COGS is set by the same suppliers who can become competitors. Microsoft already competes with Monday.com in collaboration software. It also supplies the models. If any of the three model labs ships an enterprise agent orchestration product, the middle layer gets squeezed from above. This is the Layer2 dilemma from crypto: rent security from the base layer, then hope the base layer never expands into your lane. Hope is not a mechanism. The data flywheel is the last pillar. 225,000-plus customers generate workflow data: which automations succeed, which agents fail, how users correct errors. That is high-value training signal. Traditional SaaS competitors do not have it. But the compliance boundary is real. Enterprise clients did not sign up to be training data. If consent walls rise, the flywheel stalls. An un-harvestable moat is a monument, not a moat. There is another adoption damper. Enterprise clients are afraid of sending workflow data to external models. The fear does not have to be justified to be costly. It only has to be real. If risk teams restrict agents to low-value, low-risk tasks, the credits do not burn fast. The product performs, and the revenue line stays flat. The revenue potential of the AI platform is capped by an invisible limit: how much trust the customer has in third-party model endpoints. Zero-retention agreements and private model options are not a feature add; they are a revenue prerequisite. The credit wallet creates a new internal economy. Enterprise IT must decide which department gets which budget. That is FinOps logic applied to collaboration software. Someone inside each company becomes the credit allocator. Every agent deployment is now a budget question. This slows procurement. Sales cycles lengthen. Selling a seat count is simple. Explaining a hundred work tasks times an average credit burn is not. The customer education burden is real, and the layoffs cut people from the function that does that education. Crypto projects have tried this shape before. Compute credits, API credits, usage tokens. Most die in the metering complexity or collapse under whale demand. The ones that survive share one trait: the price of the credit is anchored to a real cost, and the supply is burned on use, not reissued on narrative. Monday.com is not issuing a token. But the accounting discipline it needs is identical. Burn must equal service cost. If it does not, the credit is just a coupon with a marketing layer. Now the contrarian angle. The market's 12.6% pop was a narrative trade, not a ledger trade. The AI Work Platform story offers a new valuation frame: AI infrastructure multiples instead of SaaS multiples. That can hold - for a while. But the highest-risk assumption is the one everyone repeats: that Monday.com's 225,000 customers are an unassailable moat. The installed base is also the installed inertia. Existing customers were sold a collaboration tool. Now they are being sold an execution platform with a consumption meter and a 25% prepayment penalty. Migration resistance is not a moat; it is a churn risk. And the layoffs cut customer success headcount at the exact moment customers need education. The companies most likely to churn are the ones least likely to raise their hands. Two quarters from now, watch the per-customer credit consumption curve. If it bends downward as models improve, investors will finally ask the question nobody asks: what happens when technology gets so good that customers need almost no credits? The financial model collapses back to a subscription floor - minus the laid-off staff. The 50% drawdown priced the old model poorly. The 12.6% bounce prices the new model generously. Both are guesses dressed as analysis. None of this is a verdict on Monday.com's survival. It has a real install base, real cash flow, and a real first-mover claim on the AI platform category. But the ledger will answer three questions: margin per credit, burn linearity, and whether pricing migrates from inputs to outcomes. If Monday.com moves to outcome-based pricing, it validates the only token design that ever worked: pricing results, not resources. If it stays input-metered, its own AI improvements will shrink its revenue. Trust the math, ignore the memes. The moon is a myth; the ledger is the only truth. Survival is the first profit metric. Watch the burn, not the bounce.