AI Didn't Kill the Sales Engineer. It Made the Job Harder.

A conversation with Erik Scoralick, Director of Sales Engineering at Delinea

🎙️ This post is based on an episode of Building for Others, the elvex podcast where we talk to people who’ve built AI tools, workflows and products that real people use. No hype, no vaporware, just things that work.

Guest: Erik Scoralick, Director of Sales Engineering, West AMER & LATAM at Delinea

Company: Delinea, identity security for human and AI identities

LinkedIn: Erik Scoralick

Podcast: Building for Others by elvex, hosted by Sachin Kamdar and Doyle Irvin

The Sales Engineer Role Didn’t Die. It Got Harder.

When people started predicting that AI would kill the sales engineer, Erik Scoralick had spent 15 years in rooms with CISOs, turning technical detail into business decisions. He started in São Paulo, spent seven years at Forcepoint, and now runs a team of 10 sales engineers across West AMER and LATAM at Delinea.

He doesn’t think the role is going away. He thinks it’s getting more technical and more strategic, and harder to automate.

“AI is making things more technical,” Erik says. “It’s touching more resources. Reps can have more information, but there’s a depth of knowledge that’s necessary for when opportunities qualify.”

That’s the Jevons Paradox showing up in a sales team. When AI makes part of the job cheaper, people expect more from whoever does it. If a deck, a prospect brief or a demo script takes a fraction of the time it used to, the standard for a good one goes up by about the same amount.

Don’t Use AI As A Crutch

Erik is blunt about what AI can do to people early in their careers. “People starting their career need to find a balance between using AI and not using it, because otherwise you’re gonna be using AI as a crutch.”

A presentation used to take him hours. Now it takes 30 minutes. The catch is that you have to know what a good one looks like already, or you won’t notice when the AI wanders off.

“Sometimes I need to roll up my sleeves. It’s not building what I want. It’s going rogue. It’s using something that I don’t like.”

The sales engineers who do well know their domain well enough to use AI to get more done while still trusting their own judgment over it. They also know when to leave it closed.

Customers Arrive Pre-Briefed and Pre-Skeptical

One change Erik describes says a lot: customers now walk into meetings having already done their homework with AI.

“I had a customer that paused me and said, ‘Eric, can you stop presenting? I saw this presentation. Someone from your company recorded it, and I watched it. That’s why we’re here. Can you go straight to the demonstration?’”

That customer had used AI to go through Erik’s own material before the call. Buyers show up knowing the basics and want a demo built around their situation, not a slide deck.

So the SE’s job is now mostly translation in the moment. The customer already has the facts. What they need is someone who listens to their use case, adjusts on the spot and shows why the product matters for them. “Sometimes you need to whiteboard together, draw it out, work it out together,” Erik says.

The Moats That AI Can’t Cross

Erik names two things that keep sales engineers necessary. The first is translating between the technical and the business side. “Great sales engineers understand the technical bits and bytes and translate them to business challenges. You need a filter. You can’t say every single thing that goes on behind the scenes.” The second is building trust in person. “People buy from people. One in-person meeting is sometimes better than ten virtual meetings.”

AI can put together a demo, research a company and draft the follow-up email. It can’t sit across from a CISO, read the room, and decide how far into the technical detail to go or when to step back and talk about business results.

What Erik Looks for When Hiring

This has changed how Erik interviews. “If you’re someone more seasoned, you will need to know AI. That’s gonna be part of my hiring process. How are you using Claude? How are you using AI? What’s your second brain?” He holds early-career candidates to a different standard. “I would teach them and coach them.”

Knowing how to use AI is turning into a basic requirement. What separates people is judgment about when to use the tool, when to overrule it and when to set it aside.

What Erik Built for His Own Team

Erik isn’t only changing how he works himself. He’s building things his team uses. He standardized Delinea’s Proof-of-Value process globally across a spread-out SE organization, so every SE starts from a shared kit and adapts it to the prospect. His team runs an on-demand demo platform where prospects watch recorded use cases on their own time and pass them to other decision-makers. “We’ve gotten in contact with a CISO we didn’t have a connection with because someone shared that content.” He also uses AI to pull patterns out of 40 customers. “Where’s our winning ratio? Where we’re winning the most? How we’re losing? What do we need to improve?”

All of these were built for other people to use.

The Rest of the Discussion

AI Agents Are Identities. Treat Them Like One.

Erik’s main point on AI security is that an AI agent is an identity. It moves fast, it doesn’t have human judgment, and if you tell it “always allow,” it will do exactly that. He brings up the Hugging Face incident, where an AI agent with too many permissions began escalating its own privileges. “There were no boundaries. It started going around, finding privileges, self-escalating.”

Zero Standing Privilege for AI Agents

Delinea’s answer, which the StrongDM acquisition sped up, is to authorize at runtime and give agents no standing privileges. “You have an AI agent on your machine. It has no standing privilege to go to the database and query something. Policies tell the agent to only do the things that are permitted. Once you’re done, that privilege is gone.”

The Pool Model for AI Licensing

Sachin asked a question more companies running lots of agents are starting to face: if there are 3.5 agents for every human user, how do you pay for all of those identities? Erik suggests pooling. “Not every agent is gonna be running all the time. Some do a task once a day for a couple minutes and are done. You need a mindset of pool licensing. A hundred concurrently running agents, not a hundred thousand individual seats.”

Agentic vs. Autonomous AI

Erik separates agentic AI, where a person is involved (“I ask and prompt things”), from autonomous AI, which runs with no person in the loop. “Sometimes the agent can do both. It might start autonomous and then go agentic. The architecture needs to be ready for that.”

Closing Advice: Don’t Forget Security

His advice to builders: “There’s a lot of content out there. Start studying. But don’t forget the security part. You’re giving the keys of the kingdom of your whole social network.” He points to the flood of open-source AI harnesses hooked up to WhatsApp, Telegram and sensitive data with no limits on what they can do. His rule: “You’re not gonna drive a car without the seatbelt. Don’t start using AI without thinking about the security perspective.”

Bring your agent. Bring your team.