Martin Spier, who leads ChatGPT Performance at OpenAI, framed the core problem at QCon AI as explosive user growth crashing into agentic development at scale. ChatGPT hit one million users in five days from launch and the engineering teams serving it never fully caught up. Agentic development made that catch-up structural.

The speed problem is not primarily GPU-bound. A single user request passes through client processing, conversation loading, context assembly, tokenization, routing, inference, streaming, and observability before a token reaches the user. Any step can bottleneck. If a task requires 30 model requests, an extra second per request compounds. Reducing repeated work throughout the system matters as much as making the model faster in isolation. Shaving one second per request off the orchestration layer matters as much as raw model throughput.

Per-request inference pipeline: every stage before and after the model can become a bottleneck in agentic workloads.
FIG. 02 Per-request inference pipeline: every stage before and after the model can become a bottleneck in agentic workloads. — OpenAI / QCon AI — Martin Spier

OpenAI's response: a Rust-based agentic harness connecting models, tools, and the user environment. The key design choice for cache efficiency treats all model-visible history as append-only. New messages, tool results, and environment state append at the end rather than insert earlier. Tools surface in deterministic order. Approval policies and runtime settings apply at execution time, not embedded in definitions. The prefix seen by the model stays stable across turns, boosting prompt-cache hit rates. GPT-5.2 API pricing embeds the incentive: 90% discount on cached input tokens ($1.75/1M standard) gives that design discipline financial weight.

Token typePrice per 1M tokensDiscount vs. standard
Standard input$1.75
Cached input$0.17590%
FIG. 03 GPT-5.2 API input token pricing: financial incentive built into the cache-efficiency design — OpenAI GPT-5.2 pricing page

OpenAI spent $5.02 billion on Azure inference alone in the first half of 2025. In June 2026, engineers developed a software-only optimization—no new hardware, no architectural overhaul—that cut inference costs by more than 50%. Applied to ChatGPT's logged-out visitor traffic, it reduced the Nvidia GPU count serving that segment to roughly a couple hundred. The technique remains undisclosed; analysis points to KV cache reuse, quantization, and smarter request batching as likely contributors. Whether gains transfer to paid API tenants running agentic workloads—the most compute-intensive traffic—remains open.

MetricValue / Detail
Azure inference spend (H1 2025)$5.02 billion
Cost reduction achieved>50%
Optimisation approachSoftware-only — no new hardware or architectural overhaul
GPU count serving logged-out traffic (after)~a few hundred Nvidia GPUs
Likely techniques (analyst assessment)KV cache reuse, quantization, smarter request batching
FIG. 04 OpenAI inference cost optimisation — key metrics from June 2026 software-only effort — OpenAI engineering / TechTimes / AI Weekly

Agentic coding adds second-order pressure: regressions land faster. Teams shipping with coding agents sit one abstraction layer removed from understanding what they pushed. Humans no longer know all details of a change before it goes live. That compresses the detection window before regression compounds. OpenAI's answer: meet automation with automation. Deploy always-on AI agents for profiling, regression detection, and continuous optimization. The performance team is building telemetry and tooling that agents can read directly, rather than waiting for humans to analyze flame graphs.

Standard benchmarks—fixed 1k-in/8k-out or 8k-in/1k-out token pairs—don't capture the multi-turn, heavy-tailed, tool-interleaved shape of agentic traces. The metrics that matter shift: end-to-end trace latency, time-to-first-answer-token across a full agent session, cache hit rate across turns, and scheduler behavior under high concurrency of short output bursts. KV cache pressure from long-running sessions and scheduler pressure from high output request volumes are the load characteristics that matter; single-turn benchmarks miss both.

DimensionStandard BenchmarksAgentic Workloads
Token profileFixed pairs (e.g. 1k-in/8k-out or 8k-in/1k-out)Multi-turn, heavy-tailed, tool-interleaved
Primary latency metricSingle-request latencyEnd-to-end trace latency
Time-to-first-token scopePer single requestPer full agent session
Cache measurementNot capturedCache hit rate across turns
Concurrency patternNot capturedScheduler under high concurrency of short output bursts
KV cache pressureNot capturedLong-running sessions
Scheduler pressureNot capturedHigh output request volumes
FIG. 05 Standard inference benchmarks vs. agentic workload metrics — what each captures and misses — OpenAI / QCon AI — Martin Spier

Inference serving for agentic workloads is a systems design problem with multipliers. Fix the harness before you tune the model.