Cloud providers and AI: avoiding a dependency you did not choose
How AI features arriving inside the cloud and software platforms you already pay for create dependencies nobody decided on, and how to keep the choice open.
The short answer
A main AI dependency arrives without being chosen. It will arrive as features inside the platforms they already pay for: the office suite, the CRM, the accounting system, the support desk, the cloud provider. Each feature is convenient, needs no new vendor and uses data already in place, so it gets adopted without a decision. The dependency that accumulates is not on the model, which is replaceable, but on the data and the workflow: company knowledge indexed only inside one vendor’s assistant, prompts and rules configured in their interface, automations built with their proprietary tools, analysis that exists only in their dashboards. Once those are in place, leaving means rebuilding the AI layer, which is why the price increase or the changed feature gets accepted. The way to keep the choice open is not to avoid platform AI, which is often the right first option, but to keep the assets portable: your source of truth for documents and knowledge in a place you control, your prompts and rules in your own repository, and an annual review of what leaving would cost.
Where platform AI is a good idea and where it binds
| Situation | Use the platform’s AI? | Caution |
|---|---|---|
| Summarising documents already in that platform | Yes | Keep originals in your own store too |
| Drafting inside the tool people work in | Yes | Prompts and style rules also kept in your repository |
| Search across the platform’s own content | Yes | Know whether it can index content from elsewhere |
| Company-wide knowledge base | Careful | Keep the source documents outside; the index is the vendor’s |
| Business logic and rules | No | Rules encoded only in a vendor interface are not portable |
| Automations across systems | Careful | Proprietary workflow tools are the hardest thing to leave |
| Analysis and reporting | Careful | Keep the definitions and the underlying data extractable |
| Customer-facing assistants | Careful | Conversation logs, prompts and grounding data should be exportable |
Keeping the choice open
- Inventory the AI features you already pay for or have switched on, including per-seat add-ons.
- Read the data terms for each feature specifically; they sometimes differ from the base product.
- Keep the source of truth outside: documents, knowledge and customer data in stores you control, with the platform indexing rather than owning.
- Keep prompts, rules and definitions in your repository, even when they are also configured in a vendor interface.
- Prefer open connectors over proprietary workflow tools for anything that crosses systems.
- Know the exit cost for each dependency: what would have to be rebuilt.
- Review annually at renewal, when the cost of AI add-ons becomes visible.
The reasonable position
Use platform AI where it is genuinely good and the data already lives there, which is often. Do not let it become the place your business logic lives. The distinction is between using a vendor’s AI to work with your data and letting a vendor’s AI become the system of record for how your business works. The first is convenience; the second is a dependency you did not choose, and it is worth ten minutes of thought before each new feature is switched on.
What this means for you
AI features inside the platforms you already use are convenient and worth using, and they create dependencies nobody decides on. Keep the source of truth for documents and knowledge in stores you control, keep prompts and rules in your own repository, prefer open connectors for cross-system automation, know the exit cost of each dependency, and review it annually at renewal. Use the features; do not let them become where your business logic lives.
Frequently asked questions
Is it wrong to use the AI features in software we already pay for?
Not at all, and often it is the sensible first option: the data is already there, the terms are already agreed, and there is no new vendor. The caution is about accumulation. When your documents, your knowledge base, your prompts, your automations and your analysis all live inside one platform's AI layer, that platform's pricing and roadmap decisions become yours, and the exit cost grows quietly. Use the features; keep the assets portable.
What does the dependency actually look like?
Your company knowledge indexed only inside one vendor's assistant. Prompts and rules configured in their interface rather than stored in your repository. Automations built with their proprietary workflow tools. Analysis that only exists in their dashboards. None of it is exportable in a useful form, so a change of platform means rebuilding the AI layer from scratch, which is why the price rise or the feature change is accepted instead.
How do we use these features safely?
Know what each costs, including per-seat AI add-ons that appear on renewal. Read what the vendor may do with your data in that feature specifically, which is sometimes different from the base product. Keep the source of truth for documents and knowledge in a place you control. Keep your prompts and rules in your own repository even when they are pasted into a vendor interface. And review annually what would be lost if you left.