In March, at Nvidia's GTC conference, Jensen Huang said: "Every SaaS company will become a GaaS company." Since then, clients have been asking us two questions. What is GaaS? And should they still pay to build a SaaS product if it's about to be replaced?
We build SaaS for a living. Hisaabii, our invoicing and POS software for Saudi Arabia, is SaaS. So is School360. We also build AI agents, and our own website uses one to draft proposals. So we have a stake in both answers, and we'd rather give you an honest one than a dramatic one.
The short version: GaaS is real, but it won't replace SaaS the way streaming replaced DVDs. It will sit on top of it, change how it's priced, and wipe out some of the smaller tools. Here's the longer version.
What SaaS actually is
SaaS means software as a service. Instead of buying software and installing it on your own computers, you rent it and use it in a browser or an app. The company that makes it runs the servers, fixes the bugs and ships updates. You pay every month or every year.
You already use plenty of it: Gmail, Shopify, QuickBooks, Canva, Zoom, Salesforce. When a salon owner books appointments in a web app, or a shop issues invoices from a dashboard, that's SaaS.
The important thing about SaaS is who does the work. The software gives you the tools, and a person still uses them. Someone types the invoice, someone answers the support ticket, someone moves the lead to the next stage. That's why most SaaS is priced per seat: one login per person, so more staff means a bigger bill.
That model has been very good to software companies for twenty years. It's predictable, and it grows as the customer grows.
What GaaS means (and why the name is confusing)
Here's the awkward part. There's no agreed meaning for GaaS yet.
The letters have been used for "governance as a service" and "gaming as a service" for years. Since Huang's keynote, people use it for software that does the job itself through AI agents. You'll also see "generative agents as a service", "agents as a service" and "service as software". They all point at the same idea, so in this article we'll use GaaS for that.
The idea is simple. With SaaS you get a tool and do the work. With GaaS you describe the result you want, and an AI agent does the work using the tools.
A quick example. Say you have 40 unpaid invoices.
- With SaaS, your accountant opens the invoicing software, filters the overdue ones, writes 40 reminder emails, and checks back next week.
- With GaaS, an agent notices the invoices are overdue, sends a polite reminder in the customer's language, follows up on WhatsApp after a few days, marks the paid ones, and only brings a person in for the three customers who are disputing the amount.
The second one isn't science fiction. Every piece of it can be built today. What's new is that it's starting to be sold as a product, and priced by the result instead of by the login.
How the two compare
- Who does the work. SaaS: your staff, using the software. GaaS: an AI agent, with your staff checking and handling exceptions.
- What you pay for. SaaS: seats or plans, usually per month. GaaS: work done, such as a resolved support ticket, a booked meeting or a processed invoice.
- What you're buying. SaaS: access to a tool. GaaS: an outcome.
- What it competes with. SaaS: other software. GaaS: often the cost of hiring someone to do that job.
- Where it goes wrong. SaaS: nobody uses the tool properly. GaaS: the agent does the wrong thing confidently, and nobody notices for a week.
That fourth point is why investors are excited. A company might spend a few thousand dollars a year on support software, and far more on the people answering the tickets. Software that does the answering is chasing the much bigger budget.
Pricing is where you'll notice it first
You don't have to wait for this. It has already started.
Intercom charges for its Fin AI agent per outcome, $0.99 each, with no seat fees. An outcome is counted when a customer's problem is resolved or Fin finishes a task for them. Salesforce has been moving its Agentforce pricing towards usage and actions. Many AI-first tools now bill by usage, and research quoted by Deloitte found that 83% of AI-native SaaS companies offer usage-based pricing.
Gartner expects at least 40% of enterprise SaaS spending to move to usage, agent or outcome-based pricing by 2030. That's the clearest sign of where things are heading. When the software does the work, charging per person stops making sense, because there are fewer people logging in.
Outcome pricing sounds fair, but read the small print. Someone has to decide what counts as a "resolved" ticket or a "qualified" lead. If the vendor decides, and the vendor gets paid per resolution, they have a reason to count generously. Ask how an outcome is measured, whether you can see the record of each one, and what happens when the agent gets it wrong.
So will GaaS replace SaaS?
Partly. Here's what the people who track this for a living say:
- Gartner predicts that by 2030, 35% of point-product SaaS tools will be replaced by AI agents or absorbed into the agent platforms of bigger software companies.
- Deloitte expects some enterprise applications to be replaced by agents eventually, but not in 2026. It puts that at five years or more, with customer service going first and bigger systems like ERP and CRM after.
So roughly a third of small single-job tools may go. The other two thirds stay, and change.
From what we see in our own projects, that sounds right, and here's why the replacement is only partial.
An agent needs somewhere to work. It can't send an invoice reminder if there's no record of the invoice. The ledger, the customer list, the stock count, the patient file: that's the system of record, and it's still SaaS. In Saudi Arabia, every tax invoice has to be issued, stored and reported to ZATCA in a set format. An agent can chase payment on those invoices, but the invoicing system underneath has to exist and has to be right. The agent is a new worker in the building, not a new building.
Some jobs need a person to be accountable. Signing off a payroll run, approving a refund over a limit, deciding to fire a supplier. Agents can prepare all of that. Most businesses, and most regulators, will still want a named person to press the final button.
What disappears is the thin stuff. A tool that only schedules social posts, only transcribes meetings, or only turns a form into a PDF is in trouble. Those jobs are easy for an agent inside a bigger system to do on the side. If your SaaS idea is one feature with a login page, that's the risk.
What this means if you run a business
You don't need to rip anything out. A sensible order:
- Find the repetitive work your team hates. Chasing payments, answering the same 20 questions, copying data between two systems, writing first drafts of quotes. That's where agents earn their keep first.
- Keep your system of record clean. Agents are only as good as the data they work from. Messy customer records make a confident, wrong agent.
- Start with a person checking. Let the agent draft and a person approve, then loosen it once you trust it. That's how we run our own.
- Ask vendors how they price agents. If your support software starts charging per outcome, compare it to what that work costs you now, and check how "resolved" is defined.
- Keep your data portable. If an agent platform holds all your history and you can't export it, you've swapped one kind of lock-in for a worse one.
What this means if you're building SaaS
We get this question most from founders, and our answer is: still build it, but build it differently than you would have three years ago.
- Make it work for agents, not only for people. A clean API, clear permissions and good webhooks mean other companies' agents can use your product. Software that only works through its screens will get worked around.
- Put an agent inside it. Users will expect the product to do the obvious next step, not just show a dashboard. Hisaabii can already share an invoice straight to a customer's phone. The natural next step is for it to chase the late ones by itself.
- Think about pricing early. Seats may still be fine for your first customers. But leave room to charge for work done, or for usage, without rebuilding billing later.
- Own something an agent can't fake. Regulatory approval, deep industry rules, a trusted record of transactions, data that builds up over years. That's what makes a product hard to replace.
How we're using it ourselves
We'd be a bit suspicious of anyone writing about this who doesn't use it, so here's what we do.
On our own menu page, when someone picks services and asks for a proposal, an AI model writes a first draft: what each item covers, a timeline, the next steps. The customer reads it, changes their order if they want, and sends it to us. Then a person on our team reads it, calls them, and gives a fixed price. The agent saves everyone the first hour. The person still makes the promise.
Our website chat works the same way. It answers the common questions and links people to the right page, and hands over to a person on WhatsApp when it should.
That's GaaS at a small scale. Neither replaced our software. Both sit on top of it and do some of the work.
The bottom line
SaaS is software you use. GaaS is software that does the work. The second is growing fast, and over the next five years it will change how software is priced and kill off a lot of single-feature tools. It won't remove the need for good systems underneath, because agents need something solid to work on, and businesses need someone accountable at the end.
If you're planning a SaaS product and wondering whether it'll be out of date by launch, or you want an agent to take a boring job off your team, tell us what you have in mind. We'll tell you honestly which parts are worth automating and which aren't.
Sources
- Nvidia CEO says the next software battle is over outcomes, PYMNTS, March 2026 (Jensen Huang at GTC 2026)
- SaaS meets AI agents, Deloitte TMT Predictions 2026, including the Gartner forecasts for 2030
- Fin AI agent pricing, Intercom