The Walled Garden

Last October I published a piece about the problem with app-centric software. I have been going back through the record I put out while we were building BeMo, checking each claim against what building it actually taught. This edition is different from the earlier ones in one way: the industry just showed up at the same diagnosis, from the other side, and watching it arrive is what exposed the two things my essay got wrong.

What I claimed

The diagnosis, first: "Most software still treats apps as the unit of work, each with its own data, interface, and memory. Even AI tools, for all their intelligence, tend to reproduce that logic: one chatbot per product, one conversation per silo." And the cost: "Users become the connectors, copying content between systems, re-explaining goals, re-uploading files, and re-establishing context again and again."

Then the prescription, and here is the sentence to watch: "Instead of separate products linked by integrations, our four apps operate inside a single platform, one ecosystem, one tab, one shared memory."

I diagnosed a fragmented world and prescribed a walled garden. Everything connected, inside the wall. Integrations dismissed in a subordinate clause as the old way.

The industry arrives

Last week Stacey Segal, COO of BrightVine Solutions and a longtime Blackbaud-ecosystem practitioner, published a piece for nonprofits on "headless" technology. Her definitions are clean: "The 'head' is the user interface. The screen. The portal. The app. The dashboard. The form. The place where someone clicks, searches, reviews, or takes action. The 'body' is the underlying system. The data, rules, workflows, security, and logic that make everything work." Separate them, and data stops being trapped on a single screen; it shows up in Teams, in Slack, on a phone, in an AI agent.

That is my October diagnosis, arriving from the data side a year later. The person as the connector, named as the cost of fused systems, by someone who spent a career inside the biggest fused systems in the sector.

She writes for the enterprise tier, and her prescription fits it: a lakehouse foundation, integration tooling, a governance process. Organizations that can staff that should read her and do it. The organizations I build for, a staff of three with no data engineer, will never convene that committee, which is why nobody had been writing this for them. But the part of her piece worth the most is not the prescription. It is the warning: "Headless should not mean directionless." And then six questions: "Which system owns donor identity? Which system owns giving history? Which system owns communication preferences? Which system owns finance reconciliation? Which version of engagement should AI use when recommending next steps? Which data is ready for automation, and which still needs human review?"

Those questions are the wall I hit in 2026, published by someone else.

The two things my essay got wrong

The first I have scored before, from the inside: the suite framing died on contact with the beta, and I wrote that edition already. The walled garden is the outside view of the same miss. My essay solved the person-as-connector problem for work that happens inside our wall, and a nonprofit's working life does not happen inside anyone's wall. The donor record, the ledger, the inbox: the connector I described is still carrying context between those systems, because my prescription only helped once she was already home.

The second is the one I could not see until Segal's questions put words on it. My essay framed the win as continuity: the conversation that follows you, the context that never resets. That is a user-experience claim, and it held as far as it went. What it underplayed is that the hard part was never continuity. The hard part is authority over the data. Who may read this? Who may act on it? Which system's version is true? When the AI answers, on whose record does it stand? I spent a year of building learning that every capability question eventually becomes one of her six governance questions, and my October essay contains no answer to any of them, because in October I had not hit the wall yet.

What building forced

The answers we ended up with are structural, and I hold them as rules rather than claims of finished glory. The organization's knowledge lives in one record that belongs to the organization: readable by a person, cited when the AI answers from it, corrected by a human when it is wrong, portable when they leave. Every answer carries "how do I know this." The AI proposes and a person confirms, so nothing in the confirmed record is something the software invented. Those behaviors are live today, and they are the small-organization answer to "which version is true" for what the organization knows about itself: the version it confirmed, with the receipt attached. For the donor record and the ledger, her answer stands. The system that manages each one owns it, and our job is to read it, never to replace it.

What I have not shipped: the reads into the systems where the rest of the data lives. That is the direction the walled-garden confession forces, the work is in front of us and not behind us, and I would rather say so than imply a finished bridge. The essay's mistake was not that the wall was too small. It was thinking the wall was the point.

The sector-scale version

The 2026 Nonprofit AI Adoption Report (Virtuous and Fundraising.AI, 346 organizations) says 92% of nonprofits now use AI, and 7% report major improvements in organizational capability. The number in between explains the gap: 81% use it individually, without shared workflows. Adoption without a shared record does not move capacity; it gives each person a faster way to produce work the organization still cannot hold. That is my October diagnosis and Segal's headless argument at population scale, and it is the finding I would have dismissed as marketing in 2025 and now recognize as the whole problem.

In one sentence, the delta: I wrote that the fix for the person-as-connector was one platform with everything inside the wall; building taught us the fix is one record with the authority questions answered, and the wall was never the point.

Then: one ecosystem, one tab, one shared memory. Now: one record, its receipts, and a person in charge of it, reaching toward systems we do not own and never needed to.