Contents
At a glance
- Category
- Business and buying
- Definition
- The accumulated cost and difficulty of moving away from a supplier
- Biggest line in automation
- Workflows, which do not port between platforms
- Cheapest mitigations
- Own the credentials, export workflow definitions, document logic outside the tool
- Model providers
- Less locking than they look - prompts port, APIs are similar
- Genuinely binding in AI
- Fine-tuned models, vendor-specific frameworks, large embedding corpora
What vendor lock-in actually is
Vendor lock-in is the accumulated cost of leaving. It is not a binary state, and complete avoidance is not a sensible goal - every choice creates some dependency, and refusing to depend on anything means building everything yourself, which is a far larger cost. The useful question is not whether you are locked in but how much it would cost to leave, and whether that is proportionate to what you are getting.
It matters because lock-in changes negotiating position. A supplier who knows you cannot leave prices accordingly, and a platform that repriced last year can reprice again.
Where it comes from in an automation stack
Five sources, ranked by how expensive they are to unwind. Workflows: they do not port between platforms, so a switch means rebuilding every one. This is usually the biggest line. Data: records held in a proprietary system, and the real question is whether a full export includes relationships and history or just flat lists.
Proprietary features: the more you use a platform's unique capability, the harder it is to replicate elsewhere. Integrations: connectors built and authenticated in one platform have to be redone. Skills and habits: your team knows this tool, and retraining is a real cost even when nothing technical stands in the way.
Practical ways to limit it
Six things, none of which require avoiding good products. Keep credentials in your own accounts rather than the platform's, so revoking a platform does not orphan your integrations. Export workflow definitions regularly, even if the format is only readable by that platform - it is still the specification.
Keep business logic documented outside the tool, in plain language, so a rebuild starts from a specification rather than from archaeology. Keep the source of truth in a system you control, treating derived stores as rebuildable caches. Prefer standards where they exist - MCP for tool integrations is a genuine example, since it makes the tool layer portable across AI clients. And test an export annually rather than assuming one works.
Lock-in in the AI layer specifically
Model providers are less locking than they appear at first glance, because prompts are portable and the APIs are similar enough that swapping is days rather than months. What does lock you in is anything provider-specific: a fine-tuned model on one provider's platform, an assistant or agent framework unique to one vendor, or a large corpus of embeddings that would need regenerating.
The reasonable position is to accept model-layer dependency - the switching cost is low and the alternative is a lowest-common-denominator abstraction that costs you capability - while being deliberate about the pieces that genuinely bind, and keeping your eval set so you can prove a replacement works before you commit.
What you should know
Price lock-in as part of total cost of ownership rather than treating it as a philosophical objection. A concrete question - "if we had to leave in two years, what would it cost and how long would it take" - produces a number, and that number is what you weigh against the product being better.
And be honest that the alternative has its own lock-in. A bespoke system built by one developer is locked to that developer's knowledge, which is often a tighter constraint than a commercial platform with documentation and a support contract.
Vendor lock-in FAQ
What is vendor lock-in?
The accumulated cost of leaving a supplier once you depend on them. It is not binary and avoiding it entirely is not sensible - the useful question is how much leaving would cost and whether that is proportionate to what the supplier gives you.
How locked in am I to my automation platform?
Mostly by your workflows, which do not transfer between platforms, so a switch means rebuilding them. Also by integrations you have authenticated there and by your team's familiarity. Your data is usually the easiest part to move if exports include relationships and history.
How do I reduce lock-in without avoiding good products?
Keep credentials in your own accounts, export workflow definitions regularly, document the business logic in plain language outside the tool, keep the source of truth in a system you control, prefer open standards like MCP for the tool layer, and test an export annually.
Am I locked into my AI model provider?
Less than it appears. Prompts are portable and the APIs are similar enough that a swap is days rather than months. What genuinely binds is provider-specific work: a fine-tuned model, a vendor-only agent framework, or a large corpus of embeddings that would need regenerating.
Does building it myself avoid lock-in?
It changes the lock rather than removing it. A bespoke system is locked to the knowledge of whoever built it, which is often tighter than a commercial platform with documentation, a support contract and a hiring pool. Be honest about that in the comparison.
How should I factor lock-in into a buying decision?
As a number in your total cost of ownership. Ask what leaving in two years would cost and how long it would take, get a concrete answer, and weigh it against how much better the product is. Treated as a philosophical objection it never gets resolved.
Primary source
Model Context Protocol - an open standard that reduces tool-layer lock-in
Checked August 2026. Anything priced or rate-limited moves, so verify against the source before you budget.
Want this built rather than explained?
Knowing what vendor lock-in means is one thing; having it running in your business is another. We scope one workflow at a time, quote a fixed price in writing after a free consult, build it, and hand it over with documentation you own.
