For decades, buying software meant buying permission. You paid a vendor for the right to use code they wrote, on terms they set, and you shaped your operation around whatever that code could do.
That model is on its way out. Software matters more than ever, but the code itself is no longer the scarce part. What businesses pay for now is a set of services: model inference, compute, data, and the expertise to orchestrate them into something useful.
At NPG we have always leaned toward building custom tools over licensing off-the-shelf products. For years that was a position we had to defend in every sales conversation. Lately it is just how the math works out. Here is what changed, and what it means for your next renewal.
First we bought it. Then we rented it.
The original deal was the perpetual license. You paid once, installed the software on your own machines, and paid an annual maintenance fee for updates and support. The vendor's product was the code.
Then SaaS changed the delivery model. Vendors moved the software to their own servers and charged a subscription, usually per user per month. Operationally it was a real improvement. No servers, no installs, no upgrade weekends.
The commercial logic barely moved, though. You were still paying for access to a fixed set of features that someone else designed for their average customer. The unit of value was the seat: a person logging in to do the work. More people meant more seats, whether or not the software did anything more for you.
That is still software licensing. We just stopped calling it that.
Held hostage by our own billing software
Here is what renting looks like when the landlord changes.
NPG runs its time tracking and invoicing on Harvest. In 2025 Harvest was acquired by Bending Spoons, the company that also owns Evernote and WeTransfer (and as of yesterday, MailChimp). Then the new pricing arrived. Customers started reporting renewals at 10 to 15 times what they had been paying. The BBC reported on one UK consultancy whose monthly bill went from $130 to $2,110. We were one of those customers.
Harvest's explanation is that older plans were outdated and the product does more now. From the customer's side, the software was the same. What changed was who owned it.
This is the part of the SaaS bargain nobody reads closely. The vendor's leverage is not the code. It is everything you have put into it: years of time entries, invoices, client records, and the habits of your whole team. Leaving is painful, the vendor knows it, and the renewal notice gives you 30 days. I call that being held hostage.
Notice how the increase arrived, too. Harvest moved from a flat per-seat price to a base fee plus usage charges. A metered model does not protect you by itself. If your data and workflows live inside someone else's product, someone else sets the terms. In our case, they wouldn't even tell us the usage formula but did indicate it was based on our total invoice amounts. Which, according to how their software works, counted some of our billings twice!
A few years ago we would have had two choices: pay up, or spend months moving to another vendor who could do the same thing to us later. This time we had more. With agentic coding tools, our options included rebuilding the entire platform, or at least every part of it we touch, in a weekend.
That option changes the negotiation even if you never use it. A vendor can only hold you hostage when leaving costs more than staying. In our case, the only cost became how we could get the data out. Clearly, others are leaving the platform too, because when we first started looking into this a month or two ago, there was a global export - now Harvest moved it to multiple places to make it harder!
The code stopped being the scarce part
Software licensing rested on one assumption: writing good software is slow and expensive, so it makes sense to pay someone who already did it. Spread across thousands of customers, the vendor's cost per customer drops and everyone wins.
AI coding tools broke that assumption. Work that used to fill a quarter for a development team can now be done by a much smaller team in a fraction of the time. The custom build that used to lose every cost comparison against a subscription now wins a lot of them.
The data is catching up with what builders already see. In McKinsey's 2026 State of AI survey, 32 percent of respondents said their organization had decided against buying at least one software product or feature because it could be built in-house with agentic coding tools. Among the companies getting the most financial value from AI, it was nearly half.
Retool's 2026 Build vs. Buy report points the same way. In a survey of 817 builders, including its own customers, 35 percent had already replaced at least one SaaS tool with a custom build, and 78 percent expected to build more this year. That sample leans toward people who like to build, so treat it as a signal and not a census.
Investors have noticed. By May, Betterment's investment team reported that Adobe and Salesforce were trading at roughly half their early-2025 price-to-earnings multiples. The worry behind that repricing is simple: if customers can build their own tools, what happens to the subscription?
What you are actually paying for now
Follow the money in a modern AI-powered application and very little of it goes to a license.
There is a thin layer of code on top, and you can own that. Underneath it, everything is a metered service. You pay a model provider per token for the reasoning. You pay a cloud provider by the hour or the request for compute and storage. You pay for data, search, and specialized APIs by the call. Nobody hands you software. They do work for you and bill you for how much work they did.
The incumbents are moving in the same direction. Salesforce lists Agentforce at 10 cents per action or $2 per conversation, next to its traditional per-user plans. Zendesk announced outcome-based pricing for its AI agents back in 2024.
The analysts expect this to spread. Gartner predicts that by 2030 at least 40 percent of enterprise SaaS spend will shift toward usage-, agent-, or outcome-based pricing. It also estimates that up to $234 billion in application software spending is at risk by 2030 as agents start doing work that people used to do through a screen.
Read those price lists closely. A charge per action or per resolved ticket is a fee for work performed. It looks more like an invoice from a staffing firm than a software license. That is what I mean when I say we are all licensing services now.
The same shift applies to the people involved. The valuable skill is no longer writing every line of code. It is orchestration: choosing which services to combine, governing how they are used, and owning the result.
Own the thin layer. Rent everything underneath.
If the services are rented either way, the strategic question is what you should own. My answer: the layer that makes your business different.
Your workflows and business rules. This is the part off-the-shelf software forces you to bend around. When you own it, the software fits the operation.
Your prompts and context. The instructions, examples, and reference data that make a model useful for your work are intellectual property. Keep them in your own system, versioned and auditable.
Your data, and what the system learns from it. Every correction and exception your team feeds an AI system is knowledge. Gartner's George Brocklehurst argues that who owns that learning will be the most important clause in the next generation of software contracts.
Your ability to switch. Put your own interface between your application and the model providers. Models and prices change every few months. If you can route work to a different model without rewriting the application, you keep your negotiating position.
This is how we approach client builds at NPG: a governed layer the client owns, sitting on model and cloud services that can be swapped.
One warning. Rented services come with a meter, and meters run. In the same McKinsey survey, about one in five respondents said AI operating costs, including tokens, had constrained their use of AI. Budget for consumption the way you budget for payroll, with a forecast, a cap, and a person accountable for the number.
Where a license still makes sense
A headline like this one needs a caveat. Licensing is not disappearing everywhere, and building is not always the right call.
Buy when the software is a commodity and the vendor carries a burden you do not want. Payroll, accounting, email, identity, and security tools fall here. The vendor absorbs regulatory change and liability, and nothing about your payroll run sets you apart.
Buy when you cannot commit to maintaining what you build. Custom software has an owner or it has a problem. In Retool's survey, 60 percent of builders had created tools outside IT oversight in the past year. That is a lot of software with nobody responsible for patching it.
Be realistic about returns. McKinsey found that 37 percent of respondents attribute any EBIT impact to AI, a share that has not moved in a year. Building got cheaper. Building well still takes discipline.
As Gartner's Brocklehurst put it, this is "less an apocalypse and more of a metamorphosis." I agree. What is dead is the license as the default. The starting assumption used to be buy first and build only if you must. For the parts of your business that set you apart, that has flipped.
Five questions for your next renewal
Before you sign another multi-year agreement, ask:
What unit are we paying for? If it is seats, does the number of people logging in still track the value we get?
Which parts do we use? List the features your team touches every week. That list is the real scope of a replacement.
Can our agents use it? Check that the API covers what a person can do on screen, and that the contract permits automated use. Gartner is advising CIOs to negotiate this now.
Who owns what it learns from us? Your corrections and exceptions are improving someone's system. Make sure it is yours.
What happens to the bill at ten times the usage? Ask for caps, alerts, and a rate card you can model.
You may still renew. Plenty of software earns its place. But renew because the service is worth the price, and not because building is out of reach. That reason is gone.
If you are weighing a renewal against a build, we are glad to walk through the numbers with you.