AIThis post was created with the assistance of artificial intelligence (AI).

TL;DR

OpenAI describes a cloud based system called the Performance Factory that routes performance alerts through Codex agents, engineers and deployment checks. The company says its first workflow tracked more than 1,700 investigations and identified root causes in over 600 cases in just over three months; it has not published independent measurements of how much faster ChatGPT became.

OpenAI says it has automated much of its performance regression process with a cloud based system called the Performance Factory, which uses Codex agents to investigate alerts, propose fixes for engineers to review and check whether performance recovers after deployment. The company reports that its first workflow tracked more than 1,700 investigations and traced the root causes of over 600 cases in just over three months, a process it says helps teams keep services such as ChatGPT responsive as development accelerates.

Buying for a business?Offer from Amazon

Get business pricing on tech for your team

  • Business-only prices and quantity discounts
  • Tax-exempt purchasing
  • Multiple users, one account, clear invoices
As an affiliate, we earn on qualifying purchases.

The first workflow, called the Regression Workflow, starts when a performance alert fires. An agent investigates the slowdown and records its findings in a task, then proposes a fix for an engineer to review. After a change is deployed, another agent returns to the task to check whether performance has recovered to baseline. OpenAI says this connects investigation, proposed remediation and post-deployment verification in one process.

OpenAI reports that the workflow tracked over 1,700 investigations and root caused more than 600 cases during a period of just over three months. The company characterizes that pace as 17 automated investigations per day. These are OpenAI’s reported operational figures; the account does not give a separate measurement of time saved per case or a quantified change in ChatGPT latency attributable to the system.

The system combines a task service with configurable workflows, a dispatch loop that assigns ready work to agents, cloud execution for Codex, and subscriptions that can resume tasks when events occur, such as a review comment or completed deployment. Engineers define task states, permitted transitions, and steps requiring human attention, with separate written prompts instructing agents at each stage. The article describes the system as handing work between agents and engineers while retaining task state and findings.

At a glance
reportWhen: Described in an OpenAI account; figures…
The developmentOpenAI has detailed how its Performance Factory uses Codex agents to investigate, propose fixes for and check recovery from service performance regressions.
OpenAI · Briefing

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.

1,700+
investigations tracked
600+
root causes identified
17
automated investigations per day, reported

From alert to recovery check

The Regression Workflow carries a performance task through investigation, human review and follow-up.

1
Investigate the alert

A Codex agent examines the slowdown and records findings in a task.

2
Review a proposed fix

An engineer reviews the agent’s suggested change.

3
Check after deployment

Another agent checks whether performance returns to baseline.

How the work changed

Earlier approach
  • Engineers inspected metrics, logs and traces.
  • They selected from more than 30 performance tools.
  • People guided investigations and follow-up.
Factory workflow
  • 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.

Operations

More alerts can be worked

A consistent workflow may help teams investigate slowdowns across services such as ChatGPT.

For users

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.
Read with careThese are OpenAI’s reported operational figures; the account provides no independent evaluation or quantified user impact.
The takeaway
The Performance Factory connects alerts to review and recovery checks, but investigation volume alone does not show that ChatGPT became faster.

How Regression Work Reaches Engineers

The reported workflow could reduce the amount of manual coordination required to move a performance alert toward a fix. OpenAI says engineers previously had to inspect metrics, logs and traces, select from a collection of more than 30 performance tools, and guide investigations with service-specific knowledge. Codex assistance made investigations easier to start, but people still steered the process and handled fixes and follow-up checks.

For users, the intended benefit is operational: teams can investigate more alerts and check deployed changes through a consistent process, potentially helping them respond to slowdowns in services such as ChatGPT. The figures show the system is being used at scale inside OpenAI, according to the company. They do not establish how often the investigations led to a deployed fix, how quickly recovery followed, or whether users experienced measurable improvements.

Products Worth Considering

DJI Mic 3 Wireless Microphone (2 Transmitters, 1 Receiver, Charging Case)

DJI Mic 3 Wireless Microphone (2 Transmitters, 1 Receiver, Charging Case)

As an affiliate, we earn on qualifying purchases.

From Codex Assistant to Workflow

OpenAI describes the Performance Factory as the latest stage in a progression. Before Codex, engineers manually compared metrics and traces, debugged live machines and read code to find causes. Repeated problems, including regressions in ChatGPT’s first token latency, led the team to build specialized tools; OpenAI says it accumulated more than 30.

With Codex, engineers could ask why a service had slowed and get help searching codebases, locating signals and profiling. The team then encoded investigation steps in a shared skill called perf-forensics, covering when a slowdown began, which requests were affected, and whether deployments or configuration changes could explain it. OpenAI says it tested and refined the skill using evaluations based on past regressions. The later move to cloud based automated investigations meant alerts could start Codex sessions, but people still had to turn findings into fixes and track follow-up. The Performance Factory was built to carry tasks across those stages.

What the Report Leaves Unmeasured

OpenAI does not specify the calendar dates covered by the three-month reporting period or provide the underlying alert and case data. The account also does not define how it counts an investigation or a root cause, or state what share of the 1,700 investigations produced confirmed fixes.

It remains unclear how many proposed changes engineers approved, how long cases took from alert to recovery, and whether the workflow measurably changed ChatGPT latency or availability for users. The company describes the system and its internal activity, but the account offers no external evaluation or comparison with performance work conducted without the factory.

More Workflows Across Services

OpenAI says the Performance Factory is designed to support recurring workflows across its services. The company describes a model in which a workflow that identifies a possible optimization could open a task in another workflow to test and implement it. That points to a broader use of the system beyond responding to regression alerts.

The account does not announce a release date, a new public feature or a timetable for expanding the factory. Further reporting from OpenAI on deployment rates, recovery times and service-level outcomes would help show whether the reported investigation volume translates into faster fixes and a more consistent experience for users.

Where I land

I read the Performance Factory as a practical attempt to address a real operations bottleneck: alerts are useful only if a team can investigate them, act on findings and verify the outcome. OpenAI’s reported volume suggests the workflow is doing substantial internal work, and carrying a task through review and deployment checks could make follow-up less dependent on an engineer remembering every step.

The strongest counterargument is that investigation counts do not show user benefit. A large number of cases can reflect workload without showing that fixes were correct, deployed quickly or improved service performance. I would be more confident in the claim that this keeps ChatGPT fast if OpenAI published comparable before-and-after recovery times, the share of agent-proposed fixes accepted, and measured latency or availability outcomes over a defined period.

Key Questions

What is OpenAI’s Performance Factory?

It is a cloud based system OpenAI says coordinates Codex agents and engineers through performance tasks, from alert investigation to post-deployment checks.

What does the Regression Workflow do?

When a performance alert fires, an agent investigates and records findings, then proposes a fix for an engineer to review. After deployment, another agent checks whether performance has returned to baseline.

How much activity did OpenAI report?

OpenAI says the workflow tracked more than 1,700 investigations and traced root causes in over 600 cases in just over three months. It describes this as 17 automated investigations per day.

Does this prove ChatGPT is faster?

No. The report describes an internal process and its investigation volume, but does not provide a measured change in ChatGPT latency or availability attributable to the Performance Factory.

Source: OpenAI

You May Also Like

Grok 4.7 Is Now In Viktor: What You Need To Know – BASENOR

A headline says Grok 4.7 is now in Viktor, but provides no details on availability, features, or what Viktor is.

Anthropic Changes Usage Policy To Ban Model Abuse And Election Interference – TechCrunch

Anthropic’s reported policy change targets model abuse and election interference, but the headline alone leaves the rule details and enforcement unclear.

New NVIDIA Cosmos 3 Open Models For Interactive World Simulation

NVIDIA adds four 16B-parameter open-weight Cosmos 3 Nano checkpoints for interactive driving and robotics simulation.

Using Thousands Of AI Agents To Uncover New Leads In Biological Data

OpenAI’s article headline describes using thousands of AI agents to find leads in biological data, but the available details do not establish the methods or results.