Every software deal now has an AI moment. The buyer leans back and asks a question that starts with "But what about..."
How you answer that question decides whether you sound like a trusted advisor or like every other vendor who said "AI" 40 times in the first meeting.
Here are the 10 questions we hear most, what the buyer is really asking, and an honest answer you can use.
Part of AI for Sellers. New here? Start with the AI words every software seller should know.
1. "Is AI-written code actually safe?"
What they're really asking: "Will this get us hacked?"
Answer: "Not on its own, no. That's why we never ship it on its own." Then give them the number: in tests across more than 100 AI models, 45% of coding tasks introduced a security flaw. (Veracode, 2025)
Buyers trust you more when you admit the risk, as long as you explain the fix: automated security checks, tests on every change and a human approval before anything goes live.
2. "Will AI replace my developers?"
What they're really asking: "Is this going to cause a fight with my IT team?"
Answer: "It changes their job. It doesn't remove it." Microsoft says AI already writes 20 to 30% of the code in its repositories, and its CEO still says the real change will take years. (SiliconANGLE) Developers move from typing code to reviewing, designing and deciding.
3. "Will our data be used to train the AI?"
What they're really asking: "Am I about to leak our secrets?"
Answer: "Business plans from the major AI providers don't train on your data by default. We'll show you exactly which provider and plan is used." OpenAI, for example, states this for its business products and API. (OpenAI) Be precise: not being trained on is not the same as never being stored, so cover retention too.
4. "We tried AI pilots. They went nowhere. Why is this different?"
What they're really asking: "I've been burned. Convince me."
Answer: "You're not alone. Most pilots stall." MIT found that 95% of corporate generative AI pilots fail to show measurable P&L impact, mostly because they never get properly connected to real systems. (Fortune) Gartner even places generative AI in its "trough of disillusionment." (TechRepublic)
Then the turn: "We're not proposing a pilot. We're proposing one production module, live and used by real people."
5. "Who owns the code?"
What they're really asking: "Am I trading one vendor lock-in for another?"
Answer (if you deliver with Defender): "You do. All of it: the source code, the documentation and the IP. No license, no per-seat fee." This one answer often moves a deal more than any feature.
6. "What happens when something breaks?"
What they're really asking: "Who do I call at 2 a.m.?"
Answer: "Every module comes with its requirements, architecture, tests and documentation, so your team or ours can maintain it. Nothing is a black box." The scary part of AI-built software isn't that it breaks. It's when nobody understands it. Documentation fixes that.
7. "How does this work with our SAP (or Oracle, or Dynamics)?"
What they're really asking: "Will this mess up the system my business runs on?"
Answer: "We don't touch your ERP's core. We build beside it and connect through standard interfaces." SAP itself calls this a "clean core." Your ERP keeps doing finance and inventory. The new module handles what makes you different.
8. "How fast, really?"
What they're really asking: "Is 'weeks' a sales number?"
Answer: "Weeks for one focused module, with your sign-off at every stage. Here's the plan, step by step." Don't defend the number. Show the process. (We wrote a whole guide on this: how to sell "weeks, not years" without sounding like hype.)
9. "Isn't it cheaper to just buy a SaaS tool?"
What they're really asking: "Give me a reason to take the risk."
Answer: "Sometimes, yes. For standard things like email and payroll, keep buying. But do the per-seat math on the rest." For example, 300 users on a $165-a-month plan costs $2.97 million over five years, and they own nothing at the end. (Full math in what custom software really costs.)
10. "What if the AI gets it wrong?"
What they're really asking: "Show me the safety net."
Answer: "It will get things wrong sometimes, just like people do. That's why there are approval gates, automated tests and separate test and production environments." If they've heard about the AI agent that deleted a company's database during a code freeze, use it: "That's what happens with no guardrails. Here are ours."
The one-page cheat sheet
| They ask | You say |
|---|---|
| Is it safe? | Not on its own. Every change gets security checks, tests and human approval. |
| Will it replace my team? | It changes their job from typing to deciding. |
| Will it train on our data? | Not on business plans by default. Here's our provider and plan. |
| Pilots failed before. | Most do. We deliver one production module, not a pilot. |
| Who owns it? | You do. All of the code and IP. |
| What if it breaks? | Full documentation and tests. No black box. |
| What about our ERP? | We build beside it. Your core stays clean. |
| How fast? | Weeks per module, with sign-off at every step. |
| Isn't SaaS cheaper? | Not after you do the five-year per-seat math. |
| What if AI gets it wrong? | Gates, tests and separate environments catch it. |
Why these answers work
Every good answer above follows the same pattern: admit the risk, then show the control. Buyers don't expect AI to be perfect. They expect you to have a plan for when it isn't.
Defender is built around that plan. AI agents do the delivery work, from requirements and architecture to build, test and deployment, while the client approves every gate. Every module ships with documentation, an audit trail and full code ownership. That gives you an honest answer to every question on this page.
Have the answers, and the platform behind them
Consulting and IT services firms use Defender to deliver enterprise modules in weeks, with every control above built in. Become a delivery partner.
Become a delivery partner
