Do You Need Forward Deployed Engineers? 5 Signals from New Relic Now

AI agents are writing code and acting in production now. In customer conversations I have, this question often comes up: who’s making sure these agents deliver?

Increasingly, the answer is a forward deployed engineer, or FDE. Gartner^® predicts that by 2028, more than half of AI tech and service providers will use FDEs to speed up AI adoption. In March 2026, that figure was 10%. ^*

At our most recent New Relic Now, Camden Swita, who leads AI product management for New Relic, sat down with Kevin Bai from Anthropic’s applied AI team, who helped build the FDE function at Rippling before joining Anthropic. Five lessons below from that conversation are very relevant for enterprise engineering and operations leaders.

1. FDEs shine where deep technology meets deep business expertise

FDEs create the most value when a powerful, highly technical platform meets a customer whose expertise lies in running their business. Those teams know their use cases better than anyone. Their engineers are often committed to other priorities, and new platform capabilities arrive quickly.  An FDE pairs deep platform knowledge with the customer’s business expertise. The customer gets full value from the platform, and the FDE carries much of the complexity. 

We’ve lived this at New Relic. We recently formalized our own FDE function, specifically for customers adopting our newer AI and AI observability capabilities.

2. Fit the technology to the business, and outcomes follow

In enterprise, the technology must fit around the customer’s business. The more advanced it gets, the more time someone needs to spend in the field finding where it really applies. That fit is what turns a powerful platform into real value for the customer’s business.

As stated by Gartner in the 2026 The Forward Deployed Engineer Handbook, "Forward deployed engineers bridge technical execution and business operations by embedding in customer teams to solve challenging technical problems and translate product capabilities into measurable client outcomes."^**

3. Give agents context the way you’d onboard a new hire

Models keep getting more powerful. Even so, they still need specific knowledge of how your company works, and the good news is that providing it can be simpler than many teams expect. You’d never leave a sharp new hire to figure everything out alone. You’d tell them what the job is, how to do it and where to find what they need.

Connectors are a big part of that. Anthropic introduced the Model Context Protocol (MCP), an open standard for connecting models to systems of record. A good rule of thumb: if a human teammate would need access to your telemetry to finish a task, the agent needs it too. An IT operations leader at a large benefits administration company described this issue when we spoke recently. Early bots they built, he said, read the IT tickets and missed the telemetry. A ticket only captures what someone reported, while the telemetry in your observability system of record shows what happened across your systems.

Our customer Turo shared an example at the same event. When an alert fires, their agent checks New Relic and has a root cause analysis waiting before engineers are even online. In one incident at midnight, it found the root cause within a minute.

4. Build evals from the agent’s goal, and keep running them

As Kevin put it, “You cannot understand what you cannot measure, and you definitely can’t improve something that you can’t benchmark.” 

First, figure out what the agent is really for. A support agent’s job is to resolve the customer’s issue in a way that works for the customer and the business. Then work backward to the things you can measure along the way, like tone, timeliness and the resources it shares. Finally, keep the evals running. Applications change. Connectors change. Models get swapped out.

Gartner also sees growth of evals use. In its Top Strategic Technology Trends in Software Engineering for 2026, Gartner^® predicts: “By 2028, 70% of software engineering teams will adopt an eval-driven development approach to build responsible and trustworthy AI, up from less than 20% in 2026.” ^*

5. Treat AI as an evolution

In their conversation Kevin and Camden made a parallel to the early days of mobile. Back then, companies had to decide whether to stay on the web, go mobile or do both, and the right call depended on the business. There’s no universal answer this time either, but there is a universal question: how do people use your platform today, and how should they use it to be more successful tomorrow?

The companies best positioned for this shift share a pattern. They’re intentional about their AI strategy across people, processes and agents.

Earning trust in your agents 

So, are your AI agents doing the right thing for the business and can you trust them?

Across New Relic 2026 State of AI Coding and New Relic State of Observability Forecast reports, 67% of leaders say AI now writes most of their code in a typical week, while 78% report more production incidents. One in four organizations runs AI agents in production with no observability at all. And the organizations that actively observe their agents are twice as likely to see a 3x return on their observability investment.

At New Relic Now, our Chief Product Officer, Brian Emerson, described reliability as a spectrum that runs from agentic coding through performance health, user experience, business impact, and cost and ROI. An agent can return a perfect 200 and still hand a customer the wrong answer. In the keynote demo, a refund agent made up its own refund policy. The team traced it to a single prompt change, then built a custom evaluator from production data so the next regression pages someone first.

Much of what we launched at New Relic Now meets customers where they are, whether they’re still experimenting, keeping a human in the loop or running autonomous agents.  AI evaluations, prompt ops, experiments and topic discovery bring evals into the same data model as the rest of your telemetry, with OpenTelemetry as a first-class citizen. New Relic Ground Truth takes the New Relic intelligence observability into wherever our customers are from coding tools and the agents you build yourself, with a focus on answers that are accurate, fast and token-efficient. New Relic Autopilot, our set of SRE agents, investigates issues using that unified context. And across the platform, teams can keep a human in the loop until they trust automation to act.

Whether your FDEs come from a vendor or from your own team, what they build must keep running, keep improving and keep proving its value after the engagement ends. That takes visibility into every agent and the systems it touches, every prompt it runs and every token you spend.

Want to hear the full conversation between Kevin and Camden? Watch the full session on demand at New Relic Now.

See what your agents are doing

Try New Relic and turn on AI observability today or explore New Relic Ground Truth to bring insights to wherever your agents work.

*Gartner, Top Strategic Technology Trends in Software Engineering for 2026, Joachim Herschmann, Mark O'Neill, 5 March 2026

**Gartner, The Forward Deployed Engineer Handbook, Peter Hyde, 23 September 2026

GARTNER is a trademark of Gartner, Inc. and/or its affiliates.