Salesforce is changing the conversation about AI productivity with its new 'Agentic Work Units' (AWUs).
For a while, the tech world has been obsessed with a concept called 'tokenmaxxing'—the idea that using more AI tokens equals more work done. This is like measuring a writer's productivity by the number of words they type, not the quality of the story they write. Boards, CFOs, and regulators are growing tired of these vanity metrics. They want to see a real return on their significant AI investments, not just impressive usage charts. The pressure is mounting from all sides: financial leaders demand clear ROI, and new regulations like the EU AI Act require transparent, auditable outcomes.
This is where AWUs come in, aiming to shift the focus from cost to value. The logic behind this move is threefold.
First, it reframes the entire ROI discussion. Instead of asking "How many tokens did we use?", the question becomes "What tasks did the AI complete?". Salesforce itself reported processing approximately 19 trillion tokens while delivering 2.4 billion AWUs, perfectly illustrating the difference between raw input and valuable output. AWUs measure concrete actions—like resolving a customer service ticket or generating a sales report. This provides a clear operational KPI that finance teams can actually understand and track against business goals.
Second, it directly addresses the cost problem. "Tokenmaxxing" can lead to runaway cloud spending without a corresponding increase in business value. By treating tokens as an input cost and AWUs as the valuable output, companies are incentivized to achieve more with less. It's a strategic move to align engineering efforts with financial discipline, a core principle of FinOps that is now being applied to AI.
Third, this signals a potential evolution in how enterprise software is priced and sold. By highlighting the strong growth of its Agentforce platform alongside the AWU metric, Salesforce is positioning this as the key signal for customer adoption. If the industry adopts outcome-based metrics like AWUs, we could see a shift away from traditional per-seat licenses towards more flexible consumption or hybrid models that tie software costs directly to the value created.
Salesforce's introduction of AWUs is a deliberate attempt to set a new standard for measuring AI's contribution. The ultimate success of this initiative will depend on its governance. If AWUs can be clearly defined, benchmarked across the industry, and independently verified, they could become the essential tool for proving AI's value. If not, they risk becoming just another buzzword.
- Tokenmaxxing: The practice of maximizing the number of tokens processed by an AI model, often as a superficial measure of productivity or usage, without necessarily creating proportional value.
- Agentic Work Units (AWUs): A metric developed by Salesforce to measure the completion of a specific, outcome-oriented task by an AI agent, shifting focus from resource consumption (tokens) to value delivered.
- Vanity Metrics: Statistics that look impressive on the surface but don't actually correlate with business success or meaningful outcomes.