Earlier this year a wave of AI-agent announcements helped trigger a sharp selloff in public software companies. The shorthand that stuck was "SaaSpocalypse." The argument behind it is simple. If anyone can build software now, why pay a vendor for it?
It's a fair question, and the people asking it have real evidence. Retool's 2026 Build vs. Buy report found that 35% of organizations had already replaced at least one major SaaS tool with something they built in-house, and 78% expected to build more this year. Sixty percent of those builds happened outside formal IT, put together by operations leads and finance teams who stopped waiting for a vendor roadmap and started shipping. Klarna became the example everyone cites, talking publicly about replacing licensed tools with systems it built itself. The thesis on this side is that per-seat software was always a tax on companies that couldn't build for themselves, and AI just removed the reason they couldn't.
The other camp says something more boring and, I think, more correct. Companies still need a place where the data and the core business processes live. A system of record. What changes is not whether that substrate exists. What changes is where the value sits on top of it.
I'll tell you what we see, because we live inside this every day.
Building the first version of something is now genuinely easy. An ops lead can have a working prototype in a day. The part that hasn't gotten easier is everything that comes after the prototype. Production software has to be secured and kept compliant, and it has to keep evolving as the business around it changes. The gap between "works in a demo" and "works every day for three years" is exactly the engineering AI doesn't remove. A revenue system isn't a weekend project you finish. It's a thing you maintain, and most of the cost of owning it shows up long after you've built it.
Then there's the cost nobody modeled at the pilot stage. Some people call it token maxing: defaulting to the most capable and most expensive AI for every task, with no routing and no cost visibility. The per-token price has actually been falling. Consumption has exploded anyway, because many agentic workflows repeatedly resend large amounts of context across multi-step tasks, so costs scale faster than teams expect. Deloitte published a CFO guide on this in the spring, a topic that didn't exist on a finance team's radar eighteen months ago. Uber reportedly burned through its entire 2026 AI coding budget by April.
Here's the part that matters for the build-it-yourself argument. The runaway can be worse in self-built systems, because the teams building them often skip the routing and caching discipline that a mature platform bakes in. The freedom to build everything yourself also includes the risk of owning a system whose operating cost nobody fully modeled.
So the macro picture, as I read it: the substrate survives, and the value moves up. Now here's what we actually see in the work.
Salesforce's move toward Agentforce, Data 360, and more composable front-end patterns is directionally right. I say that as someone whose firm makes its living on this platform. The bet is that the system of record stays valuable precisely because it's wired into everything else the enterprise runs, and that the experience and intelligence layers are where the next decade of value gets created. That matches what our clients ask us for. Two patterns, specifically.
First, clients want far better user experience than the platform gives them out of the box. We've been building it. We built a quoting and deal-room experience layer on top of Revenue Cloud, NovaPact, because the native quoting flow, powerful as it is, asks reps to work the way the platform wants instead of the way they actually sell. The goal was not to replace Revenue Cloud. It was to preserve the Revenue Cloud data model and governance layer while giving sellers a faster, cleaner front end for complex deal work. The data stays in Salesforce. The experience gets rebuilt to fit the human doing the work.
Second, and this is the one I find most interesting, clients are asking us to build solutions that lift the dependencies and constraints of Salesforce out of specific workflows, while keeping Salesforce as the system of record and making the platform stronger rather than weaker. That sounds like a contradiction. It isn't. You free a workflow to where it can move fast and feel right, and you keep it wired into the platform so the data and the downstream compliance flows stay intact. Strengthening the platform and selectively freeing a workflow from it turn out to be the same motion when you do it well.
That's the part the macro takes miss. The choice isn't build-your-own against stay-on-the-platform. The work we're being paid for is both at once.
And we use AI to do all of it. Not as a talking point. Our delivery now compresses assessment and design work that used to take weeks into days, which is what lets a focused senior team take on work that used to need a much larger one. The same shift making companies braver about building is making us faster at building the right things on a foundation that holds.
For the leaders making these calls, the practical question isn't "should we build or buy." It's which workflows should stay standardized, which ones are worth differentiating, and how you keep both governed by the same system of record.
Where does this end up? I don't know, and anyone who tells you they're certain is selling something. The build-your-own wave is real, and some of it will stick. But from where we sit, inside dozens of these systems, the shape that's emerging is a durable substrate with the value migrating upward into experience and agentic workflows that hold up under real security and compliance demands, and that prove far more durable than what comes out of a one-week prototype.
That's not a prediction. It's just what we're seeing right now.