🔍 Read the full analysis: Why OpenAI Built A Performance Factory For ChatGPT on ThorstenMeyerAI.com
Get the latest gadgets delivered free with Prime
- Fast, free delivery on millions of items
- Prime Video, Amazon Music and more included
- Member-only deals all year
TL;DR
OpenAI says its cloud-based Performance Factory uses Codex agents to investigate performance alerts, propose fixes for engineers and check whether performance recovers after deployment. The company reports more than 1,700 investigations and over 600 identified root causes in just over three months, but has not published independent results or measured how much the system changed ChatGPT’s speed for users.
OpenAI says it built a cloud-based system called the Performance Factory to move performance alerts through investigation, engineer-reviewed fixes and post-deployment checks, aiming to help teams keep services such as ChatGPT responsive, as described in the original analysis. The company reports that its first workflow tracked more than 1,700 investigations and identified root causes in over 600 cases in just over three months, but it has not published a quantified effect on ChatGPT latency or user experience.
The system’s first workflow, called the Regression Workflow, begins when a performance alert fires. A Codex agent investigates the slowdown and records its findings in a task, then proposes a possible fix for an engineer to review. After a change is deployed, another agent returns to the task to check whether performance has recovered to its baseline.
OpenAI describes the factory as a combination of a task service and configurable workflows, a dispatch loop that assigns ready tasks to agents, and cloud execution for Codex. Subscriptions can resume work when an event occurs, such as a review comment or a completed deployment. Engineers set the task states, define which steps can follow others and mark where human review is required; agents receive written instructions for each stage.
The company characterizes the reported activity as 17 automated investigations per day. Those are OpenAI’s operational figures, not an independent assessment. Its account does not give a per-case time saving, show how many investigations resulted in deployed fixes or report how long it took for performance to recover.
How OpenAI’s Performance Factory keeps ChatGPT fast
OpenAI says its cloud system routes performance alerts through Codex agents, engineers and deployment checks. The reported activity shows internal scale, but not a measured improvement for ChatGPT users.
From alert to recovery check
The Regression Workflow carries a performance task through investigation, human review and follow-up.
A Codex agent examines the slowdown and records findings in a task.
An engineer reviews the agent’s suggested change.
Another agent checks whether performance returns to baseline.
How the work changed
- Engineers inspected metrics, logs and traces.
- They selected from more than 30 performance tools.
- People guided investigations and follow-up.
- Alerts can start cloud-based Codex investigations.
- Task states and findings carry across stages.
- Engineers review proposed fixes and deployment checks.
What the figures tell readers
OpenAI reports substantial internal investigation activity over just over three months.
More alerts can be worked
A consistent workflow may help teams investigate slowdowns across services such as ChatGPT.
Faster service is unproven
The account gives no measured latency or availability change attributable to the system.
What remains unmeasured
- How many investigations led to confirmed fixes.
- How many proposed changes engineers approved.
- How quickly cases recovered after alerts.
- Whether users experienced better latency or availability.
How Alert Handling Could Change
The factory addresses a practical operations problem: a performance alert does not improve a service by itself. Engineers must identify the affected requests, trace a slowdown to a likely cause, decide what to change and check whether that change worked. OpenAI says its previous process required people to inspect metrics, logs and traces, choose from more than 30 performance tools and apply service-specific knowledge.
Automating parts of that sequence could help engineers handle more investigations through a consistent process, while keeping people involved in reviewing proposed changes. For users, the potential benefit is indirect: if teams find and verify fixes more efficiently, slowdowns in services such as ChatGPT may be addressed sooner. OpenAI’s figures establish reported internal use, however, not a measured improvement in latency, availability or response times.
The distinction matters because investigation volume is not the same as successful remediation. Without data on approvals, deployments and recovery times, readers cannot tell how often agent findings led to a fix or whether the workflow made the service faster than it would otherwise have been.
Products Worth Considering

Amazon Kindle Paperwhite (2024), 16 GB, 7-inch Display, Black

Bose QuietComfort Ultra Wireless Noise-Cancelling Over-Ear Headphones with Spatial Audio, White Smoke

Amazon Kindle Scribe Colorsoft 64GB with Premium Pen
As an affiliate, we earn on qualifying purchases.
From Manual Debugging to Codex
OpenAI presents the Performance Factory as the latest step in a progression from manual investigations to more automated work. Before Codex, engineers compared metrics and traces, examined live systems and read code to locate causes. The company says recurring problems, including regressions in ChatGPT first-token latency, led its teams to develop specialized tools.
Codex later helped engineers search codebases, locate relevant signals and profile services when investigating a slowdown. OpenAI says the team gathered investigation steps in a shared skill called perf-forensics, covering questions such as when a slowdown began, which requests were affected and whether a deployment or configuration change could explain it. The company says it tested and refined that skill with evaluations based on past regressions.
Cloud-based investigations could then be started by alerts, but people still had to turn findings into fixes and track follow-up checks. OpenAI says the Performance Factory was built to carry tasks through those stages, retaining their state and findings as work passes between agents and engineers. The account describes an internal engineering system, not a new public ChatGPT feature.
What the Case Counts Do Not Show
OpenAI has not specified the calendar dates covered by the just-over-three-month period or published the underlying alert and case data. Its account does not define how it counts an investigation or a root cause, and it does not say what proportion of the 1,700-plus investigations produced a confirmed fix.
It is also unclear how many agent-proposed changes engineers approved, how long cases took from alert to recovery, or whether ChatGPT’s latency or availability measurably changed as a result. The report offers no external evaluation or comparison with performance work carried out without the factory. The reported counts show internal activity, but do not by themselves establish faster fixes or a better user experience.
OpenAI has not announced a public launch, a new feature for ChatGPT users or a timetable for expanding the system. The account also does not provide service-by-service results, so the extent of use beyond the described workflow remains uncertain.
Further Workflows and Outcome Data
OpenAI says the factory is designed to support recurring workflows across its services. It describes a possible chain in which a workflow that identifies an optimization could open a task in another workflow to test and implement it. That would extend the system beyond responding to regression alerts, but the company has not announced a rollout schedule or specified which additional workflows are in use.
The clearest next evidence would be reporting on deployment rates, recovery times and service-level outcomes, alongside definitions for how investigations and root causes are counted. Those measures could help establish whether the factory turns its reported case volume into quicker, verified fixes. For now, the system’s design and early activity figures are public; its measurable effect on ChatGPT performance remains unknown.
Key Questions
What is OpenAI’s Performance Factory?
It is an internal, cloud-based system that routes performance alerts through Codex-agent investigations, engineer review of proposed fixes and checks after deployment.
How many investigations has OpenAI reported?
OpenAI says its first workflow tracked more than 1,700 investigations and identified root causes in over 600 cases in just over three months. The company did not specify the calendar dates covered.
Does the system automatically deploy fixes?
OpenAI’s description says agents propose fixes for engineers to review. It does not say that the workflow automatically approves or deploys changes.
Has OpenAI shown that the factory makes ChatGPT faster?
No quantified change in ChatGPT latency or user experience is reported. OpenAI describes the system’s purpose and internal investigation counts, but does not publish a measured performance improvement.
Is the Performance Factory a public ChatGPT feature?
The account describes an internal engineering system. OpenAI has not announced a public ChatGPT feature or a timetable for a public release.
Primary source: OpenAI · via ThorstenMeyerAI.com
