Objection Handling Playbook
Version: 1.0
Last Updated: 2026-06-18
Refs: #4200
How to Use This Playbook
Every objection is a request for more information. Your job is not to argue — it's to identify the real concern beneath the objection and address it with a proof point.
Framework for every objection:
- Validate — Show you heard them. Never jump straight to rebuttal.
- Clarify — Ask what's behind the objection if it's vague.
- Reframe — Shift the frame from "either/or" to "both/and" where possible.
- Prove — Drop a specific number, URL, or customer name.
- Close — End with a question that moves the conversation forward.
Objection 1 — "We already use Supabase / Firebase"
Why They Say It
They've invested time and money in a stack. They're worried about migration cost, re-training developers, or just don't see the gap between what they have and what they need.
Response Framework
Validate:
"Totally makes sense — Supabase is solid, especially if you started building before agent memory was a concern."
Reframe:
"The question isn't whether to replace Supabase. It's whether Supabase solves the problem your agents have: they don't remember anything between sessions."
Proof point:
"pgvector is a column in a table. ZeroDB is a memory system designed for the agent context window — episodic memory, semantic consolidation, retrieval ranked by relevance to the current task. Supabase doesn't have that. Our benchmark hit 96.1% LLM judge score on the LoCoMo memory dataset — that's the best published score."
Close:
"Would it help to see ZeroDB running alongside your existing Supabase setup? It's one SDK install — you don't touch your current DB."
Objection 2 — "Why not just use OpenAI directly?"
Why They Say It
They're using ChatGPT or the OpenAI API and it's working. They don't see why they'd pay for an intermediary.
Response Framework
Validate:
"OpenAI is great — we route to them as part of our model pool. This isn't about replacing OpenAI."
Reframe:
"The problem with going direct to OpenAI is threefold: you're locked to their pricing, their rate limits, and their availability. When they have an outage — and they do — your agent goes down with them. We route across 100+ providers automatically."
Proof point:
"We reduced P99 latency from 38.8 seconds to 1.6 seconds by routing to the right model for each request. That's not something you get going direct to OpenAI. And when Cerebras is 10x cheaper for a specific task, our Inference Router uses Cerebras — you don't think about it."
Close:
"What's your current monthly OpenAI spend? I can show you what the same workload costs with automatic routing."
Objection 3 — "We built our own memory system"
Why They Say It
Engineering pride. They've invested in a custom solution and don't want to admit it might be suboptimal. Or they genuinely think their use case is too unique.
Response Framework
Validate:
"That's impressive — most teams don't even get that far. What does it look like today?"
Clarify:
"Does it handle cross-session persistence? Episodic-to-semantic consolidation? Context window ranking? Multi-agent memory sharing?"
Reframe:
"The question isn't whether your memory system works — it's whether you want your engineering team spending cycles maintaining it instead of shipping product. ZeroDB is what your memory system is trying to become."
Proof point:
"We have 51,000+ memories in our own production ZeroDB instance. We run 13 autonomous agents on it. The consolidation loop, the ranking system, the RLHF feedback — it took us months to build. You can skip that and plug in ZeroDB Local in an afternoon."
Close:
"Would you be open to benchmarking ZeroDB against your system on the LoCoMo dataset? We scored 96.1% — what does yours score?"
Objection 4 — "Too expensive for a startup"
Why They Say It
Budget constraints are real. They may be on a tight runway or comparing our Pro price ($49/mo) to "free" alternatives they're stitching together.
Response Framework
Validate:
"Totally fair — every dollar matters when you're early."
Reframe:
"What's your current AI infrastructure stack costing you? Add up your vector DB, your model API, your deployment platform, your memory layer. We replace all four for $49/mo."
Proof point:
"Our Free tier includes the ZeroDB community pool, access to our model router, and MCP server connectivity. Zero dollars. You only upgrade when you need more tokens. Pro is $49/mo — that's less than one AWS t3.medium instance."
Close:
"Start on Free. When you hit the limits, you'll know it's worth $49. Want me to spin up your account right now?"
Objection 5 — "We don't need MCP servers"
Why They Say It
They may not know what MCP is, or they're already building direct API integrations and don't see the value of standardized tool connectivity.
Response Framework
Validate:
"Fair — if you're only connecting to one or two tools, direct API calls make sense."
Clarify:
"How many tools does your agent currently need to talk to? And how do you handle adding new ones?"
Reframe:
"MCP is the USB-C of AI tools. Every tool your agent might ever need — Slack, GitHub, Salesforce, your own database — can be connected through one standard interface. Without MCP, every new tool is a custom integration your engineers have to build and maintain."
Proof point:
"Our MCP server gives agents access to 76+ tools out of the box. When a customer needed their agent to query their internal CRM and send a Slack notification in the same workflow, the MCP setup was 10 lines of JSON. The alternative was two separate API integrations, auth flows, and error handling for each."
Close:
"What's the next tool your agent needs to connect to? Let me show you how fast MCP gets you there."
Objection 6 — "How is this different from LangChain?"
Why They Say It
LangChain is the most common framework they've heard of. They're conflating "AI infrastructure" with "AI framework."
Response Framework
Validate:
"Great question — LangChain is everywhere and the confusion is totally understandable."
Reframe:
"LangChain is code you write. AINative is infrastructure you run. They're different layers. LangChain helps you define what your agent does. AINative provides the memory it needs, routes it to the right model, connects it to tools, and deploys it to production."
Proof point:
"LangChain memory is in-process — when your Python process dies, the memory is gone. ZeroDB persists across sessions, across agents, and consolidates short-term memories into long-term knowledge automatically. That's not something LangChain provides."
Close:
"Are you using LangChain today? If so, we have a ZeroDB integration that plugs into LangChain's memory interface. You don't have to choose."
Objection 7 — "We're not ready for AI infrastructure"
Why They Say It
They're still in the "experimenting with prompts" phase. They think AI infrastructure is a problem for later. Usually this means they haven't shipped an agent to production yet.
Response Framework
Validate:
"That's honest — and it's better to know that than to over-buy."
Clarify:
"What does 'ready' look like for you? When you get there, what would be the first AI thing you'd ship?"
Reframe:
"The teams we see fail at 'getting ready' are the ones who build a prototype on ad-hoc infrastructure and then have to tear it down when they go to production. Starting with ZeroDB Free today means your prototype is already production-ready."
Proof point:
"95% of AI pilots fail. The number-one reason is infrastructure — not model quality, not prompts. Teams that start with production-grade infrastructure from day one are in the 5% that ship. We've seen this pattern with our 7,270 current users."
Close:
"What's the first agent you'd want to prototype? I can have you running on ZeroDB Free in 10 minutes — no commitment."
Objection 8 — "Can't we just use vector databases directly?"
Why They Say It
They've heard of Pinecone or Weaviate. They think "vector database" = "agent memory" and don't understand the gap.
Response Framework
Validate:
"Vector databases are a great starting point — and you'll probably use one at some layer of your stack."
Reframe:
"A vector database answers: 'what's similar to this embedding?' Agent memory answers: 'what does this agent need to know right now, given its current task, its history, and its context window?' Those are different questions. Vector DB is a component; ZeroDB is the memory system that uses it."
Proof point:
"ZeroDB handles embedding, storage, retrieval ranking, context window packaging, episodic-to-semantic consolidation, and RLHF feedback scoring. A raw vector database gives you similarity search. You'd spend 2–3 months building everything on top of it — or you can use ZeroDB today."
Close:
"Have you tried implementing memory management on top of Pinecone or Weaviate? I'd love to hear what you built — and then show you what ZeroDB handles automatically."
Objection 9 — "What about vendor lock-in?"
Why They Say It
Legitimate concern. They've been burned before — or their engineering leadership has — by proprietary APIs that changed or companies that got acquired.
Response Framework
Validate:
"That's a completely legitimate concern and one we take seriously."
Reframe:
"We have two answers: open standards and open source. First, we're built on MCP — an open protocol backed by Anthropic, not proprietary to us. Second, ZeroDB Local is a PyPI package you can run yourself.
pip install zerodb-localand your memory layer runs on your hardware."
Proof point:
"ZeroDB Local v0.3.0 is live on PyPI right now. The data format is standard — Parquet + SQLite. If you ever leave AINative, you take your data and your memory with you. We also publish our API reference publicly at docs.ainative.studio."
Close:
"Would it help to walk through our data export and portability guarantees? We can put that in your enterprise contract too."
Objection 10 — "How do you handle data privacy?"
Why They Say It
They're handling PII, proprietary business data, or regulated data (HIPAA, GDPR, SOC 2). They need assurance before any data touches your infrastructure.
Response Framework
Validate:
"Data privacy is a hard requirement, not a nice-to-have. Let me be specific about what we do."
Reframe:
"Our architecture is privacy-first by design: data is encrypted at rest and in transit, we have RLS (Row-Level Security) on every table, and we're on our SOC 2 Type II roadmap at 97% readiness."
Proof point:
"We have 20+ compliance documents published: security policy, privacy policy, data retention policy, risk register, vendor DPAs. We've signed DPAs with 12 vendors and are in process with Anthropic and OpenAI. For regulated industries, ZeroDB Local gives you fully on-premise memory with no data leaving your infrastructure."
Close:
"Can I send you our security documentation? And if you have specific requirements — HIPAA, SOC 2, GDPR — tell me which and I'll walk you through exactly how we address them in the enterprise contract."
Quick Reference
| Objection | Core Reframe | Key Proof Point |
|---|---|---|
| Already use Supabase/Firebase | Complements, doesn't replace — memory layer is different | 96.1% LoCoMo benchmark |
| Just use OpenAI directly | Multi-provider routing, no single-vendor risk | P99 38.8s → 1.6s |
| Built our own memory | Maintenance burden vs. shipping product | 51K+ memories, 13 agents |
| Too expensive | Replaces 4 vendors; Free tier available | $49/mo vs. $200+/mo cobbled stack |
| Don't need MCP | MCP is USB-C for tools | 76+ tools, 10 lines of JSON |
| Same as LangChain | Framework vs. infrastructure — different layers | Persistent cross-session memory |
| Not ready for AI infra | Start Free, avoid the 95% failure rate | 7,270 users, zero marketing spend |
| Just use vector DBs | Vector DB is a component; ZeroDB is the system | 3 months of custom work vs. one SDK |
| Vendor lock-in | Open source + open standard | pip install zerodb-local, MCP protocol |
| Data privacy | SOC 2 in progress, RLS, on-premise option | 97% SOC 2 readiness, 20+ compliance docs |