From Owned to Rented to Used to Done: How Software Licensing Actually Works Now

For most of computing history, buying software meant buying software. You paid once, you owned a copy, and it sat on your machine until it broke or went out of date. That model made sense when software was a box on a shelf.

Then everything changed. Not once, but three times in quick succession. Each shift changed not just how you paid, but what you were actually paying for.

We are now in the middle of a fourth shift. And this one is the most significant yet.

The four stages in plain English

Before going further, here is the full arc in one place:

1. Perpetual licensing: you pay once and own the software forever
2. SaaS subscriptions: you pay monthly to access software someone else hosts and maintains
3. Usage-based pricing: you pay for what you actually consume, measured in API calls, messages, compute hours, or transactions
4. Outcome-based pricing: you pay for results delivered, not for access or consumption at all

Each stage is a response to the limitations of the one before it. Each one shifts more risk onto the vendor and more convenience onto the buyer. And each one changes the relationship between software and the work it is supposed to do.

A short history of software licensing (optional, click to expand)

The perpetual licence model dates back to the early days of commercial software in the 1970s and 1980s. Microsoft built its empire on it. So did Oracle, Adobe, and Autodesk. You bought a version of a product, received a licence key (often a physical certificate in the box), and that was the transaction complete. Upgrades were a separate purchase.

This model had a logic to it. Software was expensive to build and cheap to copy, so vendors needed to recoup development costs upfront. Buyers, in turn, got something permanent. You owned your copy of Word 97. You could use it indefinitely. Microsoft could not switch it off.

The cracks appeared in the late 1990s and early 2000s. Software was getting more complex. Maintenance and upgrades were becoming as important as the initial product. And the internet made distribution essentially free, removing the need for physical media entirely.

Salesforce was founded in 1999 and launched its subscription CRM in 2000 with a deliberate provocation: "No Software." Their CRM was delivered entirely over the web. No installation, no IT involvement, no licence key. You subscribed monthly and logged in. The software vendor handled everything else.

By the mid-2000s, this model was spreading. By the early 2010s, even the holdouts had capitulated. Adobe launched Creative Cloud in 2012 alongside Creative Suite 6, then ended perpetual sales of Photoshop and Illustrator in 2013. Microsoft launched Office 365 in 2011. The industry had decided: subscriptions were the future.

Usage-based pricing emerged in parallel, primarily from cloud infrastructure providers. AWS, launched in 2006, charged by the hour for compute and by the gigabyte for storage. You paid for what you used. This suited developers building products with unpredictable traffic patterns far better than fixed subscriptions. The model spread upwards into application software. Twilio, founded in 2008, charged per SMS and per minute of voice call. Stripe charged a percentage of each transaction. Snowflake charged for compute consumed. The pattern was consistent: align the cost with the value delivered, moment by moment.

What the subscription model actually solved

SaaS was not just a pricing change. It was a relationship change.

With a perpetual licence, the vendor's incentive was to sell you the next version. Your incentive was to stay on the version you had as long as possible. The interests were misaligned, which is why enterprise software from that era was so notoriously painful: clunky, expensive to upgrade, and maintained just well enough to keep you from switching.

A subscription flipped this. If you could cancel at any time, the vendor now had to keep earning your business every month. Updates became continuous. Customer success teams appeared. The product had to improve or customers would leave.

For buyers, the upfront cost disappeared. Instead of a large capital expenditure, software became an operational cost. Easier to approve, easier to justify, easier to cancel.

The downside nobody talked about at the time was accumulation. Individual subscriptions were cheap. Dozens of them were not. By the late 2010s, "subscription fatigue" had become a genuine concern for finance teams trying to understand what their organisation was actually spending on software each month.

Why usage-based pricing felt like progress

The subscription model solved the upfront cost problem but created a new one: you paid the same whether you used the product heavily or barely at all.

Usage-based pricing addressed this directly. Pay for what you actually consume. If your usage drops, your bill drops. If you scale up, your bill scales up with you.

This suited a certain kind of buyer very well, particularly startups and businesses with variable workloads. It also suited API-first products where usage was genuinely measurable and meaningful.

The catch was unpredictability. A fixed subscription is easy to budget for. A usage-based bill can spiral if a campaign goes viral, a bug causes a retry loop, or usage grows faster than expected. Cloud bills became legendary for their capacity to surprise. The AWS bill that doubled overnight because someone left a development environment running is almost a rite of passage now.

Both models, subscriptions and usage-based, share one thing in common: you are paying for access. Access to the tool, access to the infrastructure. What you do with that access is still entirely your problem.

The shift that changes everything: paying for outcomes

This is where AI agents enter the picture, and where the licensing conversation gets genuinely interesting.

Early AI tools were billed on usage. ChatGPT's API charges per token. Image generation APIs charge per image. You pay for the compute, not the result. The same model that covered cloud infrastructure was just applied to a new kind of workload.

But AI agents are starting to change this. An agent does not just process a request and return an output. It pursues a goal. It takes actions, makes decisions, and delivers something at the end. The logical question is: why pay per action when you could pay per outcome?

This is not purely hypothetical. Salesforce's Agentforce launched in 2024 at $2 per conversation, with pricing that has continued to evolve since. You pay per conversation rather than for the compute underneath, though the exact model is still shifting. Intercom has moved in the same direction with its AI support agent, Fin, which is priced per outcome at $0.99 (where an outcome is a resolved conversation or a completed workflow handoff) rather than per seat or per message. The outcome is the unit of value, so the outcome becomes the unit of pricing.

For buyers, this is appealing in a way that previous models were not. If the agent resolves the problem, you pay. If it does not, the risk stays with the vendor. The misalignment between buyer and seller that plagued perpetual licensing, and that subscriptions only partially fixed, disappears. You are buying a result.

What this means if you run a business

The practical implication of outcome-based pricing is that it changes how you should evaluate AI tools.

With a subscription, the question is: does this tool make my team more productive? With usage-based pricing: does the value I get exceed what I am spending? With outcome-based pricing, the question simplifies: did it deliver the result, and is that result worth what I paid?

That is a much more honest conversation. And it is one that most AI vendors are not quite ready to have at scale yet, because it requires them to stand behind the quality of their results, not just the availability of their platform.

We are in an awkward middle period right now. Most AI tools are still priced on usage or seats, even when they are clearly capable of being evaluated on outcomes. That will shift. The vendors who figure out how to price on outcomes credibly will have a significant commercial advantage over those still charging by the token.

The thread that runs through all of it

Look at the four stages together and a clear pattern emerges. Each shift moves the risk from the buyer to the vendor, and moves the value proposition closer to the actual result the buyer cares about.

With a perpetual licence, all the risk is yours. You bought it, you own it, whether it works well for you or not. With a subscription, the vendor needs to keep you happy or you cancel. With usage-based pricing, you only spend when you are actually using the product. With outcome-based pricing, you only pay when the work is done.

The direction is consistent and the logic is sound. Software pricing is gradually converging on the thing buyers actually want, which is not access to a tool but a specific result in the world.

AI agents are accelerating this because they are the first category of software that is capable of genuinely owning an outcome end to end. A spreadsheet cannot be held accountable for the quality of your financial model. An AI agent that handles customer enquiries can, in principle, be measured on resolution rate, satisfaction score, or cost per resolved case.

That is a fundamentally different relationship with software. And it is coming faster than most businesses are prepared for.

How AI with Ash can support you

If you are trying to work out where AI agents fit in your business, or how to evaluate them properly when the pricing models are still shifting, that is exactly the kind of question worth talking through. Not in the abstract, but for your specific operation, your specific costs, and the outcomes that actually matter to you.

Book a call and we can work out what this shift means in practice for your business.