Originally published on Start at Zero on Substack.
My verdict, up front: Oracle’s new Fusion Setup and Configuration Registry is a real shift in how ERP setup work gets done, and it will make configuration faster and cheaper, but it does not make your implementation cheap, because the expensive part of most public sector projects was never typing settings into a system. It was getting the organization to decide what those settings should be, and nothing in this week’s announcements touches that.
Here is what Oracle actually said. At AI World on October 26, Oracle announced the Fusion Setup and Configuration Registry (FSCR), a public Git repository that captures Fusion setup and configuration knowledge as machine-readable skills that AI agents can use. According to Oracle, agents can take business requirements, process documents, policies and existing configurations, turn them into implementation runbooks, execute the setup directly in Fusion, and then test and validate the results, after which expert reviewers examine the work, ask for changes, and approve it for deployment. Systems integrators can extend the registry with their own methods and accelerators. In a companion announcement, Oracle put a set of agents for discovery, design, configuration, testing, integration and adoption on top of FSCR inside its AI Success Navigator platform.
What Oracle did not publish matters just as much to a buyer. Oracle describes the registry as automating “much of” the manual setup work, so not every setup task is covered, and I have not seen published module coverage, pricing, or savings data that would let anyone put a number on what this does to a project budget. So I’m not going to put one on it either.
With that on the table, these are the questions I’ve been turning over since I read the releases, and the ones I’d want a finance or IT leader to ask before signing anything.
What does the integrator do now?
For a long time, a large share of what you paid an integrator for was people who knew where the settings lived, what order to configure them in, and which dependencies would break if you got it wrong. Oracle is now saying that knowledge belongs in a repository it maintains, kept current alongside the product, and that agents can act on it. If that holds up, the integrator’s value has to move somewhere else, and Oracle’s own framing is that it moves toward process redesign and business transformation.
Some firms are good at that work, but process redesign is much harder to scope, staff and evaluate than configuration, and it’s also the area where buyers have historically had the least ability to tell a strong team from a weak one. So the question for your next selection isn’t whether the integrator has used FSCR. It’s what, specifically, they will do for you that the registry and the agents won’t, and who on their team does it.
Will integrators adapt quickly when this eats into time and materials?
Every major integrator quoted in the release sounded enthusiastic, which is what you’d expect at a vendor conference. But a firm whose revenue comes from billable hours on configuration is being asked to embrace a tool that reduces billable hours on configuration, and I’ve never seen an industry give up a profitable habit faster than its clients forced it to.
My guess, and it’s only a guess, is that you’ll see two things at once for a while: firms marketing their AI-accelerated delivery while still pricing projects on staffing plans that look a lot like the old ones. The way you speed up the adjustment is by asking, in writing, how the use of FSCR changes their effort estimate and their rate structure, and treating a vague answer as an answer.
Speed to value goes up, but what about quality?
Oracle’s pitch is speed and consistency, and I can believe that an agent following a maintained, product-aligned set of skills will make fewer of the careless configuration errors that humans make at two in the morning before a deadline.
But a configuration can be technically valid and still be wrong for you. If the requirements going into the agent are incomplete or reflect the way you used to work rather than the way you mean to work, you’ll get a faster, cleaner build of the wrong thing. Automated testing and validation can confirm that the system does what it was told. It can’t tell you whether you told it the right thing.
Who gatekeeps the agent’s decisions, and whose experts are they?
Oracle says expert reviewers examine and approve completed configurations before deployment. I think that’s the right design, and I’d want to know exactly who those reviewers are on my project. Are they the integrator’s people, whose incentive is to keep the project moving and get to the next milestone? Are they yours, and if so, do you actually have staff with the time and the depth to review agent-generated configuration in finance, HR and procurement? The releases don’t say, and on a public project that question deserves a named answer in the contract.
There’s a bigger issue underneath this one, and it’s the thing I see most often in practice. Projects in cities, counties, states and universities rarely run late because configuration takes too long. They run late because the organization can’t make decisions: about the chart of accounts, about approval chains, about which departmental exception is worth preserving and which one is just habit. Being able to configure something in a day doesn’t help much if it takes three months to decide what to configure, and I’d argue that faster setup actually raises the stakes, because now the decision is the only thing standing between you and a working system. If your governance structure can’t decide quickly, this tool will make that very visible.
Are managed services dead?
If agents can make configuration updates, test them and validate them, it’s fair to ask whether you still need a managed services contract with a team of people billing monthly to keep the system running. I think it’s plausible that one or two sharp internal people, with good tools and real authority, could run and keep improving a Fusion environment that used to need a much larger support arrangement.
I’d stop short of declaring managed services dead, though, because the hard part of post-implementation support has usually been knowing what to change and why, handling the quarterly updates without breaking anything that matters, and keeping institutional knowledge from walking out the door when someone retires. Agents may help with the mechanics of all of that. They don’t replace the person who knows why your agency set something up the way it did. What I’d expect is that support contracts get smaller and more bespoke, and that agencies which invest in a couple of capable internal owners will get a lot more out of this than agencies that outsource everything by default.
Do tokens replace hours, and is that worse?
Somebody pays to run these agents. Oracle’s releases don’t say how that’s priced, and I don’t want to guess. But it’s worth thinking about now, because if integrators move from billing hours to passing through or marking up AI usage, you’re trading one cost you understand for one you may not.
Hours are expensive, but at least they’re legible. You can see a staffing plan, challenge it, and cap it. Usage-based spend on agents can be harder to forecast, depends on how many times something gets rerun or redone, and isn’t something most public budget processes are built to handle. I’m not saying it will be worse. I’m saying you should ask how agent costs are billed, who bears the cost of reruns when requirements change, and whether there’s a ceiling, before you find out from an invoice.
What I’d do this week
If you’re early in a Fusion evaluation or already mid-project, I wouldn’t change course because of an announcement, and I wouldn’t ignore it either. I’d ask your integrator, in writing, how FSCR changes their scope, their effort estimate and their pricing, who the expert reviewers will be and whose payroll they’re on, and how agent usage shows up on your bill. And I’d take an honest look at how long your own organization takes to make a decision, because that’s now the part of the project you control most directly.
If you’re in the middle of an ERP or AI decision and want a second set of eyes before you commit, book a free 30-minute decision review with me at startatzero.averoadvisors.com
Planning your next move? Write to us.