“An output of work.” How AI is changing software pricing

When Marc Andreessen famously proclaimed, "Software is eating the world," he didn't mean the world’s profits. Indeed, software has created more wealth faster than any technology since the wheel by improving productivity with automation and communication. However, the continued massive investments in AI, followed by the “low-cost” explosion of DeepSeek, has created shockwaves across the software investment community.

What is next with software and AI? A recent Andreessen Horowitz podcast describes the next era in software’s march as different because AI is converting software to labor. This means that AI is not just a new alternative in the niche enterprise software market; it is an alternative in the multi-trillion-dollar white-collar labor market. Taking advantage of this will be challenging, though, because new AI-based software is presenting price-to-value confusion for both buyers and sellers. The wrong pricing model can “eat customer value” just as fast as it can improve profits. Buyer beware: a software vendor with the wrong model can fail almost instantaneously, leaving the business’ operations high and dry. Just ask the sports fan customers of regional sports networks (Hey, Minnesota Twins fans!).

Technology forecasters are predicting “AI in everything” in 2025. But unlike their previous excitement over how ChatGPT can write a letter to grandma or recommend a hotel in any vacation spot around the world, pundits are now aware of how experienced software disrupters like Andreessen and Salesforce’s Marc Benioff are building a different kind of AI – AI that takes action. They are also leading the way by changing how they sell their AI software. AI is coming to the enterprise, but not with the usual software pricing.

Pricing productivity

For 25 years, software has delivered automation and productivity in pursuit of Andreessen’s prediction. Then AI enters the picture and muddies the economics. AI offerings, particularly generative AI like Microsoft Copilots and Salesforce AgentForce™, required tremendous investment. However, AI’s proximity to decision-making and human creativity confounds measuring productivity improvements that customers have not yet taken for granted. Early purveyors of GenAI are struggling to recover their investments within their traditional software pricing models. Microsoft recently put itself in the spotlight by using a “pay more for the next version” approach when it began charging Office365 customers for CoPilot whether they want it or not.

Per-seat software companies are heading toward a brick wall as they simultaneously tell the market that the value of their AI software is reducing the number of employees needed to run a business while charging per employee (seat). Their dilemma is exacerbated by customers struggling to see and measure the value of AI. How much should one pay for a more creative, CoPilot-enabled workforce?

Benioff sees this coming and is pivoting to a consumptive model for Salesforce's AgentForce™ offerings. Consumptive models have two advantages: they provide a clear definition of a unit of value, and they measure that value for both provider and consumer. Productivity is a measure of work output divided by work inputs – traditionally labor and materials. So, the key to consumptive pricing, Benioff says, is measuring how AI generates “an output of work” generally associated with labor. When the product is a CRM application, i.e., Salesforce, one example of an output-of-work for the CRM customer is a conversation with their customer. Benioff will charge $2 for every conversation completed by AgentForce™ with an end customer. It’s a win for Salesforce customers because every conversation made by the software saves time and money compared to the labor force making the same call. The model delivers economies of scale, and the cloud implementation makes scale unlimited. Salesforce’s stock price is up over 30% since the AgentForce™ announcement.

An-output-of-work in Equipment Finance - Using AI to drive action

In equipment finance, two obvious outputs of work are the size of a funded portfolio driving revenue and the profit from that revenue. However, these are high-level composite outputs, such as "customer satisfaction" for Salesforce. To follow Benioff’s and Andreessen’s insights on AI productivity, a more granular analysis of the outputs of work along the equipment finance workflow is necessary.

Example workflow with AI performing Outputs of Work

Following the software-becomes-labor analysis, two outputs of work generated during the workflow shown above are conversations with customers and decisions for action. Examples where software can deliver measurable outputs of work in the workflow include:

The value of these outputs of work varies significantly, and not all require AI for implementation. However, this analysis helps software developers like Tamarack identify and build the most valuable software products and helps both buyer and seller understand the impact the software can have on the performance of the enterprise. The existing SaaS and Per Seat pricing models can still apply to legacy software, but all software today should demonstrate economies of scale. If it does not, it will eat profit and equity. AI product pricing must move beyond the per-seat-of-labor model because AI is turning software into labor, breaking the standard productivity calculation of output per unit of labor. AI must be priced to clearly show the cost per unit output of work to prove that AI is a better economic choice.

AI agents are the right focus now because they are literally changing the composition of labor and the workforce, which is still the single most important input for businesses today. Not enough labor chokes growth. Too much or too high-cost labor undermines profitability. Productivity, the output of work per unit of input, is the key to business. In order for software to continue its march to change the world, AI software will have to deliver simple productivity economics.