Architecting AI-Native Platforms in the Real World — Amar Akshat on Agentic Systems, Cells, and Prompts-as-Code

AI is already in the loop for writing code, reviewing changes, and even sketching architecture diagrams—but turning those capabilities into resilient, auditable, production-grade systems in regulated domains is a different problem. In payments and financial services especially, architects have to reconcile non-deterministic models with deterministic guarantees around correctness, security, compliance, and cost.

In this conversation, we speak with Amar Akshat—SVP of Architecture at Paysafe Group and author of the forthcoming Decode the Compiler (Packt, 2026). At Paysafe, Amar has led large-scale modernization and AI-native transformation across payments, wallets, and compliance platforms; earlier at Apple, he helped shape the architectural foundations of Apple Pay and contributed to wallet and tokenization frameworks. His current work focuses on making architecture itself intelligent—through agentic systems like MCPX and ArchX, “cell” architectures that keep decision paths safely bounded, and treating prompts, guardrails, and evals as first-class architectural assets.

Over the course of the interview, Amar explains when to keep workflows purely deterministic versus putting an AI in the path, how to structure data planes, guardrails, and system prompts as design primitives, and how to choose between modular monoliths and microservices for AI-heavy workloads. He shares concrete practices around confidence-based routing and trust deltas, prompts-as-code and AI Behavior Reviews, prompt manifests as “Dockerfiles for AI,” cost control with “cache, batch, distill,” and vendor-neutral orchestration via protocols like chat completions and MCP. We close with how compiler-level thinking—and understanding what actually happens to our code—can sharpen the way we design AI-driven systems at scale.

What you’ll learn

• How to think about agentic architecture: MCPX, ArchX, and systems that can reason about their own ADRs, diagrams, and deployments

• A practical lens for “AI in the loop vs. deterministic only” based on pattern recognition vs. regulatory and financial certainty

• How to design “cells” for wallets, payments, and ledgers so that critical analysis never leaves safe boundaries

• Ways to model data, guardrails, and system prompt packages as first-class architecture: data planes, prompt manifests, and governance

• Confidence-based routing, trust deltas, and treating prompts as code with CI/CD, eval pipelines, and AI Behavior Reviews

• Patterns for privacy and governance in regulated environments: middleware, hybrid RAG, SAML-aware access, and auditable “architectural replays”

• Strategies for cost and vendor control: “cache, batch, distill,” multi-provider orchestration with chat-completions APIs and MCP, and AI gateways

Who should listen:
Enterprise and platform architects, AI and payments leaders, SRE and ops teams responsible for AI reliability, security and compliance owners in regulated environments, and senior engineers designing AI-heavy workflows who want systems that are intelligent and explainable.

Prefer reading? You can find the complete Q&A article here:

For distilled insight read Deep Engineering #27:

To receive newsletter issues featuring interviews like this subscribe to Deep Engineering.