The portability problem no one is talking about
Jason Lemkin, CEO of SaaStr, just published internal data that should worry every AI sales tool vendor: he copied a best-performing AI SDR prompt from one vendor, pasted it into a competitor, and got 80% of the value back in 20 minutes.
That is the prompt portability problem. And it is about to torch retention rates across the AI sales stack.
Lemkin runs 20+ AI agents at SaaStr. After watching churn patterns across his portfolio and testing switching costs internally, he has mapped four distinct levels of prompt portability. Your level determines whether you hold 95% gross retention or watch it decay to 80% over two years.
Level 1: Copy-paste portable (the danger zone)
AI SDR agents, outbound sequencers, email writers, meeting summarizers. Switching time: hours, maybe a day.
What transfers: everything. Tone of voice, ICP definition, objection handling, qualification criteria. You copy the prompt from Vendor A's dashboard, paste it into Vendor B, and it works on the first try. Not perfectly, but 80% of the way there.
The claimed moats (better UI, better analytics, fine-tuned models) do not stick. UI is nice but not sticky. Analytics are useful but replaceable. Fine-tuning advantages last six months before models converge.
The only real moat? Email deliverability and domain reputation. If your AI SDR vendor has spent months warming sending domains and managing IP reputation across millions of emails, that is hard to replicate. It is just not an AI moat. It is infrastructure.
Retention forecast: 80-85% gross. Expect annual bake-offs to become standard. Buyers will test 2-3 vendors every renewal cycle because switching costs are negligible.
Level 2: Prompt-plus-data portable (the uncomfortable middle)
AI customer support, sales coaching, content generators trained on brand voice, recruitment screeners. Switching time: 2-4 weeks.
Core prompts transfer easily. What does not transfer cleanly: thousands of resolved tickets, months of call recordings, hundreds of brand-approved content examples. This data has to be re-ingested, re-indexed, re-validated.
Vendors overestimate the moat. Buyers underestimate the switching cost. Reality sits in between: it is a real project, but a 2-4 week project, not a 6-month CRM migration.
The critical variable is data portability. If the vendor makes it easy to export training data and conversation history, they accelerate their own churn risk. If they make it hard, they buy time but breed resentment.
The winning play: make the data layer so valuable and continuously improving that switching costs are not about data transfer but data loss. If your model gets measurably better every month from accumulated interactions, and that improvement resets to zero with a new vendor, you have a retention story. But you have to prove it with metrics, not marketing.
Retention forecast: 85-90% gross. Better than Level 1, still vulnerable to annual competitive reviews. Vendors who publish clear "your model improved X% this quarter" reports will hold retention. Those who cannot quantify their data advantage will not.
Level 3: Workflow-embedded (the integration moat)
AI coding agents (Cursor, Windsurf, Copilot), agents embedded in CRM/ERP workflows, agents managing multi-step processes. Switching time: 1-3 months.
The prompt is almost irrelevant. Value is not in what you told the AI to do but how the AI is woven into existing tools, workflows, and daily habits.
Cursor is the example. The underlying LLM could be swapped. But value sits in IDE integration, codebase indexing, the way it understands your specific repo structure and coding patterns. Switching does not mean copying a prompt. It means relearning an entire development environment. That is a meaningful cost, measured in productivity loss.
Same with AI agents inside Salesforce, HubSpot, or your ERP. The prompt might be portable, but the 47 workflow automations, custom field mappings, approval chains, and reporting dashboards built on top are not.
What this means for sales teams
If you are buying AI SDR tools, AI email writers, or AI meeting assistants: assume zero switching costs. Test multiple vendors. Run annual bake-offs. Do not lock into multi-year contracts.
If you are buying AI tools embedded in your CRM or sales workflows: the switching cost is real but not infinite. Vendors will claim lock-in. They are right for 12-18 months. After that, competitors will have caught up on integrations.
If you are a sales leader at an AI sales tool vendor: Lemkin's data suggests your retention story needs to move beyond "our AI is better." Models converge. Prompts are portable. Your moat is either infrastructure (deliverability, integrations, compliance) or continuous data improvement you can prove with numbers.
The wave of AI agent churn is coming. The vendors who understand prompt portability will survive. The ones who ignore it will hit retention cliffs in 2026.