Here is what my product uses: [list every AI API, model, hosting provider, vector store, and any tool with AI inside it].<br/><br/>For each one, work out the chain underneath it as far as public information goes - who they run on, whose models they resell, whose silicon it ends on.<br/><br/>Then tell me where two or more of my supposedly separate dependencies converge on the same company. Those convergence points are my real single points of failure, not the vendors on my invoice list.<br/><br/>Where the chain is not public, say so and mark it unknown rather than guessing.
Here is the relevant section of my agreement with an AI provider: [paste terms].<br/><br/>Tell me, in plain language:<br/>- how much notice either side must give to end it<br/>- what happens to my data and my fine-tunes if it ends<br/>- whether they can deprecate the specific model I depend on, and with how much warning<br/>- whether price is fixed, indexed, or at their discretion<br/>- what I am committed to pay even if I stop using it<br/><br/>Then tell me which of those terms would hurt most if exercised tomorrow.
Assume my main model provider doubles its price, deprecates the model I use, or becomes unavailable in my region.<br/><br/>Walk me through what switching would actually cost: engineering time to re-integrate, prompt and evaluation rework, output quality regression, re-testing, and anything I would simply lose.<br/><br/>Give me a rough number of person-weeks and say which part of it is unavoidable versus which part I could pre-pay now by building differently.<br/><br/>MY SETUP:<br/>[architecture, how the model is called, how prompts are managed]
Here is how my application calls its model: [paste the integration code or describe it].<br/><br/>Identify everything that is specific to this provider and would break on another: SDK calls, response shapes, tool-calling format, token counting, streaming behaviour, system prompt conventions, model-specific prompt phrasing, safety-refusal handling.<br/><br/>Rank them by how much work each would be to abstract, and tell me which three are worth abstracting now and which are cheap to fix later.<br/><br/>Do not recommend a framework. Tell me what to change in what I already have.
Here is my dependency list and my business model: [paste].<br/><br/>For each dependency, tell me what happens to my product and my revenue if it disappears with 90 days notice. Sort into:<br/>1. Degraded but operating<br/>2. Broken until we rebuild<br/>3. Business over<br/><br/>Be blunt about category 3. Then tell me what the cheapest meaningful insurance is for each category 3 item - not the ideal fix, the cheapest one that moves it to category 2.<br/><br/>If nothing is in category 3, say so instead of inventing drama.
Argue both sides honestly for my situation.<br/><br/>FOR consolidating on one AI provider: simpler integration, volume pricing, less prompt maintenance, one relationship to manage, faster shipping.<br/>AGAINST: pricing power sits with them, deprecation risk, regional availability, and no leverage in any negotiation.<br/><br/>Then tell me which my situation actually supports given my size, margins and switching cost - and what I should do differently in either case.<br/><br/>SITUATION:<br/>[team size, revenue, margin, how central AI is to the product]