By Antanas Bakšys, CEO & Co-Founder of Ace Waves
There’s a debate right now about how customer service AI should be built – self-service vibe coding or delivered by forward-deployed engineers (FDEs).
Some believe the winning approach will be self-service platforms, where CS teams configure and iterate themselves through tooling and natural language, while questioning whether the forward-deployed model can scale.
But after working with 30+ large consumer companies, I learned that this framing assumes something that is rarely true in practice.
Most customer service AI projects don't fail at the application or model layer.
They fail because most CS operations are not sufficiently structured to be executed by AI agents. When policies, exceptions, and decision logic are not clearly defined, the agent – human or AI – will always be guessing.
And this is the reality for many companies: policies drift, exceptions pile up, half of the rules live in scattered docs, the other half in people’s heads, resulting in teams operating on contradictory procedures.
Introduce an AI agent into that environment and expect reliable execution – it just won’t happen. CSAT will drop. Hallucinations will lose your trust.
The real problem: Customer Service Debt
Let’s take a basic case like a physical product refund. Whether it will happen depends on the refund window, shipment status, damage, and retention strategies. The logic sits in notes, Slack threads, and human agents’ judgment – not neatly documented workflows.
We call this Customer Service Debt (CSD).
I see big consumer brands covering simple Q&A and basic tasks with self-service automation like n8n or quickly vibe coded flows. But these solutions sit only on a surface layer.
To automate 80%+ of customer service in a reliable way, companies need real AI engineering depth, a strong understanding of agentic AI foundations, and powerful tooling and infrastructure. This is what ensures the right action happens in the right system at the right moment.
The role of Forward Deployed Engineering
The Forward Deployed Engineering approach is essential here. It brings AI engineering directly into operational workflows, legacy tech-stack, translating ambiguous human work into deterministic logic that AI agents can run at scale. That work sits below the product layer, unrelated to the interface improvements.
Natural language interfaces, or simply put – prompting, and better tooling help teams move faster. But you don’t get reliable execution through configuration alone. You get there by working through the actual logic, defining edge cases, aligning systems, and making decisions explicit.
What’s more, FDE work is a continuous process. AI agents need to be integrated with real systems, tested and iterated in production, and handle edge cases to make sure the system consistently delivers the desired outcomes. When done properly, this allows easier scale, maintenance over time, and actually moves business metrics in production.
The real challenge of AI in customer service
Customer service leaders are not looking for another interface or another tool. They are looking for the right action to happen in the right system under the right conditions.
AI is already good enough to understand customers. The real challenge now is making sure it can execute the work correctly. And that starts with addressing Customer Service Debt through the deep FDE-led implementation work that turns operations into scalable systems and real business results, not just improves the product layer.
At Ace Waves, we provide exactly this through an experienced forward-deployed squad of AI engineers and product managers, paired with our product, powered by Agent Procedures.
Together, these deliver fast, safe and large-scale customer service transformation that actually moves the numbers in the P&L.
The Ace Waves Insights Newsletter
Quality insights on AI and customer service. Thoughtful, useful, never promotional.