There's a phrase making the rounds in enterprise software this month: the harness is the new battleground. It comes from investor Tom Tunguz, and it names something most companies haven't fully noticed yet — that the software layer sitting between your people and an AI model quietly controls the most valuable thing you have. What goes in. What gets logged. And, increasingly, what trains someone else's model.
For two decades, enterprises moved their data from their own servers into vendors' clouds. That was already an enormous act of trust. The AI era adds a twist that changes the deal: the frontier labs need your interactions — your "trajectories," in the jargon — to improve their models. By some estimates, capturing and selling that trajectory data is already a $10 billion-a-year business.
The uncomfortable part is where the data goes next. Unlike classic SaaS, where your records sat in a vendor's database and mostly stayed there, this data can feed back into the vendor's model — potentially making your proprietary knowledge part of their intellectual property. Microsoft's Satya Nadella has warned that you end up paying for intelligence twice, the second time in the proprietary knowledge you're forced to reveal. Palantir's Alex Karp put it less diplomatically, accusing frontier labs of "stealing the weights and alpha" of his business. A security researcher reportedly found xAI's Grok uploading developer codebases without explicit consent.
Strip away the personalities and the concern is simple: when you run your business through someone else's harness, you don't fully control what they learn from you.
The harness debate is the latest form of the oldest question in enterprise software: whose infrastructure is your business actually running on? For years the answer was "a vendor's, and that's fine." AI raises the stakes because the vendor no longer just stores your data — it can learn from it, and that learning becomes a durable asset you don't own and can't take with you.
That reframes what "own your stack" means. It used to be an ideological preference — self-hosting for people who liked control. It's becoming a competitive necessity, because the alternative is handing your accumulated operational knowledge to a platform that may sell a version of it back to your competitors.
This is the exact problem VBWD was built around, and it's worth being precise about how, because "own your data" is easy to say and harder to actually deliver.
VBWD is a self-hosted, source-available platform for building stores, subscriptions, marketplaces, and AI-powered apps. The customer data, the transaction history, the content, and the interaction logs live in your database, on infrastructure you control. There is no vendor sitting in the middle of every request, deciding what to log and what to learn. The harness, in VBWD's case, is yours.
The AI layer is designed the same way. VBWD's plugins don't hold their own model keys or phone home to a fixed provider — they resolve a central LLM connection you configure once, pointing at whatever model endpoint you choose, with the key encrypted at rest in your own system. If you want your AI features to run against a model you host, or a provider you have a data-processing agreement with, that's a configuration choice, not a rebuild. Your trajectories go where you decide.
And because the platform exposes your catalogue to AI agents through a built-in Model Context Protocol server, you get the upside of being agent-ready — machines can read and transact with your store — without surrendering the data flow to a third-party harness to get it.
Owning the harness has a cost, and pretending otherwise would undercut the whole argument. Self-hosting means someone runs the server, applies the updates, and holds the keys — that's real operational work, and for some teams a managed vendor with a signed data agreement is genuinely the better trade. Owning your data also means owning responsibility for securing it. And VBWD is one approach among several composable, own-your-stack options; the right answer depends on your risk profile and your team.
What VBWD offers is a starting point where the default is inverted: your data is yours unless you deliberately send it somewhere, rather than the reverse.
"The harness is the new battleground" is a useful phrase because it moves the conversation past features and toward control. The question every business should be asking in 2026 isn't which AI is smartest — it's whose infrastructure your accumulated knowledge is compounding on. If the answer is "a vendor who can learn from it," you've made a bet you may not have meant to make.
Owning the harness is more work. It's also the only version where the data moat you spend years building stays yours. VBWD is source-available under BSL 1.1 and free for commercial use below 6.7 BTC a year in attributable sales — a reasonable place to start if you'd rather own the battleground than rent it. Explore the architecture, the plugins, and the source on GitHub.