I didn’t expect to agree with a vision document coming out of IES this much. When I read Reimagining the Institute of Education Sciences: A Strategy for Relevance and Renewal last week, I found myself surprised at how excited I was by the recommendations. To be fair, this is probably because the ideas at its center are part of a vision EA has been working toward for years, and who doesn’t like their own confirmation bias?
OK, so what? In many ways, the vision is the easy part. What happens next is what matters. Here are a few key ideas that I took from the report.
Education data should be a stream (not a snapshot)
For as long as I can remember, the education field has treated data like something you harvest once a year, package into a report, and send off to the people who are supposed to make decisions with it. The problem is that by the time those decisions get made, the data are already stale.
Reimagining IES calls for a move toward cloud-native, API-driven data infrastructure—in other words, treating education data as a continuous stream rather than a static dataset. I am thrilled to see that shift.
But here’s what concerns me: If this document gains the traction it deserves, many states and districts are about to receive guidance (and likely funding) to modernize their data infrastructure. If we aren’t careful, modernization will look like sleek slide decks explaining how data could be interoperable. The slides will reference data standards. The diagrams will be elegant. The arrows will connect. Everyone will nod.
Slide decks are not software, and diagrams are not infrastructure. Modernization only happens when states and districts adopt real, working software like Ed-Fi, EDU, OneRoster, CASE, CTDL, and others that securely move data continuously. We’ve learned from our work that these software decisions are also data decisions, process decisions, and people decisions—and if we don’t treat them that way, none of those decisions get made well.
The architecture decisions being made right now will shape what's possible in education research and analytics for the next decade. If we rebuild on proprietary, siloed foundations (even modern-looking ones), we'll have recreated the same structural constraints with prettier diagrams.
State Consortia are the model (not just a nice-to-have)
One of the clearest findings in the IES report is something we’ve seen repeatedly in practice: states don’t have to—and shouldn’t—do this work alone. The problems facing states’ data systems are largely the same across states; the data they need to collect, the federal reporting requirements they face, and the interoperability challenges with vendors are not unique. But historically, each state has largely built its own custom solution from scratch, paying separately for the same infrastructure decisions over and over again.
State consortia break that cycle. They allow shared investment in shared infrastructure, which creates the conditions for shared learning. We’ve seen this firsthand with the Ed-Tech Collaboratory (ETC), which brings together states to collaborate on common data models, interoperability challenges, and shared solutions.
The IES report is right to elevate state consortia. If we take that seriously, SLDS modernization evolves from fifty separate technology procurements into shared public software infrastructure.
We need to build software (not just standards)
This brings us to an argument that’s worth stating plainly: the field needs a reference software strategy for SLDS to work, not just a standards strategy.
Standards documentation and data models do not move data—software does.
The field has invested enormous effort into defining data standards and interoperability. That work is critical, because without shared definitions and shared identifiers, nothing else scales. Far less energy has gone into building and maintaining software that operationalizes those standards in product environments, which is what it takes to make standards impactful. The report highlights this need by pointing out the distinction between the CEDS data standards and the CEDS software, while making a claim that the software needs to be transferred and modernized. I agree with this recommendation and hope that efforts to accomplish this continue.
The combination of two prior ideas—state consortia and building software—is powerful. State-level consortia are best positioned to build shared, durable software that is both responsive to their needs and scalable across contexts. Building shared reference software is more difficult than procuring a single solution to meet a current state need; it requires ongoing engineering, shared funding models, and real governance. It is, however, much less difficult than building a single custom solution twice. If the field is serious about modernization, we have to be serious about building and maintaining the software that will actually make data standards work.
Educators need evidence that is useful today (not three years from now)
One of the report’s most pointed observations is that IES-funded research has too often produced rigorous findings that arrive too late to matter. A rigorous study published in an academic journal years after the data are collected doesn’t help a principal who needs to decide how to allocate intervention support next month.
The report’s push toward rapid-cycle, actionable analytics like early warning systems, real-time educator dashboards, and predictive tools for chronic absenteeism is a direct response to that failure. And it reflects something we’ve believed since EA’s early days: that the gap between what researchers know and what practitioners can actually use is not just a communications problem. It’s an infrastructure problem.
Building the tools that close that gap is genuinely difficult. It requires researchers and technologists to work together in ways that neither field finds entirely familiar. We’ve written about this tension before, and I don’t want to pretend we’ve solved it. But the federal shift toward funding rapid-cycle, practitioner-facing analytics is going to create real opportunities for organizations that have been building this capacity.
Education researchers need to speak data engineering (not just statistics)
This one is quieter in the report, but it struck me as important. Reimagining IES calls for redesigned master’s programs that train future education researchers to work with APIs, SQL, and open-source tools—not just statistical methods.
This has been a gap for too long. The researchers who are going to matter most in the next decade of public education are not the ones who can run the most sophisticated regression. They’re the ones who can sit with the data streams of a state longitudinal data system, understand their structures, extract meaningful signals from them in real time, and communicate what those signals mean to a superintendent who has 15 minutes tomorrow morning at 5:45 a. m. to make a consequential decision.
EA has been trying to build that kind of capacity internally, and it’s hard to hire for, because the training doesn’t really exist yet at scale. This is a key unlock that implies pretty substantial rethinking of our research training programs.
The AI paradox
Everything I’ve said above applies to AI as well as traditional research and analytics. Furthermore, AI also creates a paradox when we talk about software development and implementation.
AI makes writing software cheaper per line of code, but it does not reduce the complexity of data governance and semantic meaning that interoperability software infrastructure solves. The paradox is thus: It is now cheaper than ever to build software, but this means it is becoming more expensive than ever to coordinate software across governance jurisdictions (or even across applications) since it is so easy to “just build a thing.” This paradox may actually increase the total cost and time to deploy of effective software, frustrating everyone who expects software to be cheap and easy now.
Therefore, the more we can solve this interoperability problem in the same way across states, the cheaper software will become—maybe even to the point where building and maintaining software can be done in house by end users, instead of by software development and SaaS companies.
The hard work ahead
Reimagining IES is a vision document, not a policy mandate. Its authors have no authority to direct spending or restructure federal programs. What it represents is a signal—a serious one—that the framing of the problem is shifting at the federal level in ways that align with where the field needs to go. That’s progress in my view, but it doesn’t get the work done.
The work is hard in ways that aren’t always obvious from the outside. Building real software is harder than writing about it. Getting states to collaborate in ways that actually produce shared infrastructure and not just shared language requires more than agreement in principle. Data governance, privacy, political will, procurement cycles, workforce capacity: These are the real obstacles, and they are formidable. EA has been living inside these challenges for years, and we’ve made real progress toward making the vision in the report a reality, by learning hard lessons about what works and what fails. We regularly try to share those lessons out, so that others don’t have to learn them firsthand.
One of those lessons? This infrastructure can’t be built by any single organization focused on winning market share. It is infrastructure for the entire public education system, which means it only succeeds when others join in building and sustaining it. The vision in Reimagining IES only matters if the field treats it as a shared agenda.
If any of this resonates with the problems you’re working on, we’d love to talk.