Nobody prices the exit. You evaluate the demo, the roadmap, the integration list. You negotiate the entrance fee down to the dollar. Then two years pass, the platform holds your briefs, your approvals, and your history, and the renewal is not a negotiation. Leaving would cost more than staying at any price, and the vendor knows it.
Agentic systems raise those stakes. A CRM holds records. An agentic system holds behavior. It accumulates how your brand decides, what it approves, what it rejects, and the reasoning behind both. If that accumulation lives in a format only one vendor can read, your switching cost compounds every day the system runs. The work you bought the system to do becomes the reason you cannot leave it.
Lock-in is set at signing, not at renewal.
Three dependencies do most of the damage. The model dependency: workflows wired to a single provider, so a price increase or a terms change lands directly on your operating cost. The data dependency: briefs, decisions, and outputs stored in proprietary structures that do not export cleanly, if they export at all. The evidence dependency: the approval records and run history that prove what happened, which you would have to rebuild from nothing on any other platform.
None of these show up in the demo. All of them are decided by the architecture you accept on day one. By renewal they are facts, not choices.
What portability requires.
We built the RDLB system against a simple test: every part of it should survive its own replacement. That produces four working rules. Models are routed, not married. The system sends each task to whichever model fits it, and swapping providers is a configuration change, not a rebuild. Records are exportable. Every run is written to audit-grade logs in plain, portable form, so the evidence of how work happened travels with the client. Briefs and brand rules live as documents you own, not as settings buried in a platform. And connectors stay read-only, so the system never becomes the only copy of anything it touches.
Routing is also an economics decision. Sending each task to the smallest model that clears the quality bar is how 44,000+ runs in 63 days cost under $50 in model spend. A system married to one provider cannot do that arithmetic. A portable one does it on every run.
Price the exit during the evaluation.
Three questions expose the architecture faster than any feature list. What leaves with us if we leave: which files, in which formats, containing which history. What breaks if the model provider changes terms: one workflow, or all of them. Who holds the approval record: you, or the platform. A vendor with good answers will put them in writing. A vendor without them will change the subject to the roadmap.
The approval gate matters here for the same reason. When a human signs off on every consequential output, your team keeps the judgment the system runs on. The rollout should install capability into your operation. It should never install dependence.
If you want to walk through what a portable system looks like from the inside, book the strategy blueprint call.
