Service as Software
For twenty-five years, we paid rent on tools we barely used. Call it what it was. SaaS sold you a seat and dared you to find value in it. The contract was simple and the contract was rigged: we give you the login, you go figure out how to turn a license into an outcome, and if you can't, that's a you problem. Renew anyway. Most of you did.
Nowhere was the rigging more obvious than higher education. I've spent long enough in this vertical to know the cadence by heart. A vendor lands a multi-year license. An implementation partner lands a bigger one next to it. Two years pass. The sponsor changes jobs. The steering committee quietly stops meeting. The tool goes live at 40% of what the pitch deck promised, and everyone agrees to call that success because the alternative is explaining the rest to a board.
The dirty secret of higher ed SaaS isn't that the tools are bad. Some of them are genuinely good. The dirty secret is that license pricing and outcome value were never on speaking terms. You paid for access. The vendor was not, contractually, on the hook for whether access produced anything. The implementation partner was happy to keep billing hours in the gap between them. And the institution, tuition-funded and already stretched thin, was left holding a very expensive flashlight and wondering why the room was still dark.
Look. I'm not mad at SaaS. SaaS was the right idea in 2001. Stop shipping CDs, stop installing things on servers in basements, move the software to the browser, charge monthly. It was a genuine leap. It funded a generation of companies that did real work. I built things inside that model and I'm not going to pretend otherwise.
But the model is now upside down. The unit of value is no longer access to a tool. It's a result that happened without a human dragging it across the finish line. Once you see that shift, the whole 25-year engine starts to look less like an industry and more like a toll booth on a road nobody needs to drive anymore.
Welcome to Service as Software.
The name is ugly. The idea is not. In SaaS, humans used the software to perform labor and paid for the right to use it. In Service as Software, agents perform the labor and humans pay for the outcome. Not the seat. Not the dashboard. Not the quarterly training. The finished thing. The transfer evaluation that's done. The re-engagement that landed. The degree map that was built and delivered while the student was still awake.
You already know what this looks like in your own life, even if you haven't named it. You don't pay a rideshare for access to a driver's time. You pay for the ride. You don't pay a delivery app for the right to talk to a restaurant. You pay for the food on your counter. The whole consumer internet learned this lesson a decade ago. Enterprise software is the last holdout, and higher ed is the last room in the last holdout, clutching its five-year license renewals and calling them strategy.
Here is the thing nobody wants to say out loud at the EDUCAUSE keynote: in an era of AI, data strategy is all you need. That's it. That's the sentence. Tattoo it on the inside of your IT steering committee's eyelids. Every other decision (which vendor, which platform, which implementation partner, which premium tier) is downstream of whether you own a clean, portable, living description of your institution and the people it serves. If you own that, you can make any tool serve you. If you don't, no tool on earth will save you, and the ones you already bought are quietly making it worse by fragmenting the one asset that actually matters.
The people who run this playbook well have figured out something that sounds like a technicality and is actually the whole game. They've separated execution from authentication. In the old world, those two things were fused. Your login was your keys. If you could authenticate, you could execute. Directly, against the database, through a UI, with whatever blast radius your role happened to allow. The entire security model of enterprise SaaS was built on the assumption that the human at the keyboard was the thing doing the work.
That assumption is no longer true, and pretending it is will get you breached, sued, or both.
In the new model, authentication is what proves who is asking. Execution is what happens after the ask is understood. Between them sits an agentic fabric. Think of it as the operating system for the work itself. The fabric spawns agents, scopes them, and holds them accountable. Every action is signed. Every data access is minimum-necessary. Every handoff between agents leaves a trail a regulator could read without a translator. The human logs in. The human states the intention. The fabric decides which agents wake up, what they're allowed to touch, and how the result gets delivered.
This is not a diagram on a whiteboard. This is how the work actually gets done now.
Let me put a number on it, because numbers are the only thing that cuts through 25 years of muscle memory. Yesterday we took a POC CRM, the one replacing a legacy Affiniquest-based system, from proof-of-concept to live production on AWS. Fully cloud-native. Fully serverless. End-to-end. React 19 + TypeScript + Vite on the frontend. CloudFront at the edge, rewriting paths through a five-line function into API Gateway. Fifty-two Lambda routes behind one HTTP API. Postgres in a private subnet. JWT auth with a bastion-only DB. Idle compute cost: zero. Total dev infra: under five dollars a month. One senior engineer. One day. Eighteen agents. Sixty-seven pull requests. Done.
Read that again. POC to production, in a single working day, by one human who never touched a keyboard except to review diffs. The agents read the repo, wrote the code, ran the tests, opened the PRs, and answered each other's comments. The human was the author of intention and the arbiter of taste. Everything else was fabric.
Sixty-seven PRs. Eighteen agents. One dev. Humans manage intent. Agents deliver it.
Stop and feel that in your body for a second. Because every CIO I know is still budgeting for a world where the delivery rate is one PR per developer per day and the overhead is meetings. They are about to be out-shipped by institutions that figured out the new ratio and stopped paying for the old one.
Here's how it lands in higher ed specifically. A dean wants to launch an accelerated BSN pathway for working LPNs in three counties. In the old world, that's a slide deck, a cross-functional steering committee, a six-month feasibility study, a SaaS vendor pitch, an implementation plan, a soft launch in fall of the year after next, and a line item in a budget that dies in committee. In the new world, the dean describes the intention. One paragraph. Maybe three. "We want a stackable LPN-to-BSN path for the following counties, with these clinical partners, these start dates, these credit transfer assumptions." The fabric reads the institutional markdown. It already knows the accreditation constraints, the advising capacity, the clinical seat availability. It spawns agents that draft the curriculum, map the skills, generate the marketing page, wire the financial aid logic, and open (wait for it) a pull request against the institution itself. A diff. Here is what your university would look like if you approved this intention. Merge, and it's real. Don't merge, and you cost yourself nothing.
The intention is the input. The PR is the output. The human judges. The agents execute. And the vendor who used to sell you a $1.2M "program launch platform" with a nine-month implementation is suddenly competing with a workflow that runs in an afternoon and bills nothing per seat because there are no seats.
I want to be clear about the pricing shift, because this is where the SaaS industry is going to fight the hardest and lose the ugliest. License-based pricing was never aligned with the customer. The vendor got paid the same whether you got value or not. Every CFO knows this and every CFO has been quietly rationalizing it for two decades because the alternative, tying price to outcomes, would have required the vendor to define outcomes, measure outcomes, and accept liability for outcomes. None of which the license model was designed to do.
Service as Software cannot dodge that definition. It's the whole pitch. You don't get paid unless the work is done. A transfer credit evaluation that completes end-to-end. A re-engagement SMS that produces a reply. A degree map that gets clicked. A student who enrolls. The invoice follows the outcome, not the calendar. Which means the vendor has to care about whether the outcome happens, because if it doesn't, the vendor doesn't eat. Incentives finally align. A miracle disguised as a billing change.
The SaaS CEOs reading this, and I know some of you are, should feel something cold in your stomach right about now. Not because your product is bad. Because the container you sold it in is about to be repriced by the market. The smart ones will pivot. They'll expose their capabilities as things the fabric can call. They'll get paid per outcome and make more money than they did on seats because the outcomes will scale past any org chart. The ones who don't pivot will spend 2027 explaining to their boards why renewals are soft and then spend 2028 not explaining anything because there won't be a board meeting to attend.
Higher ed, the warning is sharper. You are the most over-licensed vertical in enterprise software. I've audited stacks that carried four student engagement tools, three CRMs, two LMSs running in parallel, a data warehouse nobody could query without a ticket, and an AI "pilot" that was a ChatGPT wrapper with a logo on it. The annual spend was a rounding error away from a full-time faculty line. The value a student could actually feel was approximately zero.
You do not need more tools. You need a described institution, a described learner, an agentic fabric to govern the work, and the discipline to pay only for outcomes. Everything else is noise, and most of it is noise you're currently paying monthly for.
The elixir is this: in an era of agents, your moat is not your software. It's your description of the world and your willingness to pay for finished work. Own the description. Rent the capabilities. Pay for results. Let the humans author intent and let the agents carry it across the line. That's the whole job now.
A license was a promise you'd figure it out yourself. An outcome is a promise someone else already did. The industry that made itself rich on the first promise has thirty-six months, maybe less, to make peace with the second one.
I'm not here to mourn SaaS. It did its work. It paid for a lot of good engineers and a lot of reasonably decent tools. But the road is empty now, and the toll booth is still charging, and somewhere a dean is describing an intention into a fabric that's already opening the pull request.
The seat was never the product. The outcome was. We just finally built the system that could tell the difference, and the bill that follows it.
If your institution is still buying seats in a world that's learning to buy results, ring Phil and he'll show you what your stack looks like when every line item has to prove it earned the work.