Multi-Agent LLM Serving Is a Full-Stack Problem: A Small Step Towards Algorithm–System Co-Design

Large language models are increasingly deployed as multi-agent systems (MAS) that solve a task by coordinating many LLM invocations through structured control flow such as branching, debate, and iterative refinement. This capability imposes a serving cost that current infrastructure is ill-equipped to absorb: algorithm researchers iterate on MAS designs rapidly and optimize for task accuracy, whereas serving systems evolve slowly and cannot be specialized to each workflow, so MAS run on general-purpose engines blind to workflow structure. We introduce a unified graph representation $G=(N,E,Op)$ that casts existing and emerging MAS algorithms as typed graphs, turning opaque multi-agent executions into analyzable system workloads. On this representation we develop two families of optimization: static patterns, recurring motifs that a runtime accelerates without changing the algorithm, and dynamic strategies, small structural changes that trade accuracy for large efficiency gains. On real GPUs shared-prefix reuse cuts prefill latency by $1.07$–$2.42\times$, confidence-gated early exit reaches $1.7$–$5\times$ speedup at $54$–$89\%$ fewer tokens, and sparse expert dispatch lowers cost roughly $6\times$ for $2.8$ points of pass@1. Crucially, the prefill win does not survive end to end: decode is $95.6$–$99.8\%$ of latency here, so end-to-end speedup stays at parity ($0.94$–$1.06\times$, every $95\%$ CI covering $1.0$), making phase-awareness a prerequisite for co-design. We intend this work as a guide for designing hardware-friendly MAS algorithms and a step toward algorithm–system co-design for LLM serving.