Is vendor lock-in becoming a problem with AI agents?

Hi, I've been thinking about something that doesn't seem to get discussed enough when companies choose AI agents: vendor lock-in. Most comparisons focus on accuracy, resolution rate, pricing, integrations, etc. But what happens after you've spent 6–12 months building your support workflows around one AI provider? You've configured: Knowledge bases and instructions Custom actions and API integrations Refund and cancellation policies Escalation rules Conversation routing Reporting and evaluation Then another provider comes along with better models, lower pricing, or capabilities your current vendor doesn't support. How difficult is it actually to switch? I think there are a few different types of lock-in here. Platform lock-in: Your AI agent only works with one helpdesk or CRM. Workflow lock-in: Your business logic is built using proprietary tools that can't easily be migrated. Data lock-in: You can't easily export conversation history, configurations, or evaluation data. Model lock-in: You're dependent on whichever AI models your vendor decides to support. Of course, there's a tradeoff. A tightly integrated platform might work better out of the box and require less engineering. A more flexible architecture might give you more control but also require more setup and maintenance. I'm curious how people are thinking about this. For those already running AI agents in production: Have you ever tried switching AI providers? Was it relatively straightforward, or did you have to rebuild most of your workflows? And when choosing an AI agent platform, how much does portability actually influence your decision?