[KongchangAI]
· 2 min read· 1,230 words

OpenAI to Shut Down Agent Builder and Two Other Services on November 30, Developers Must Migrate Urgently

OpenAI to Shut Down Agent Builder and Two Other Services on November 30, Developers Must Migrate Urgently

OpenAI will shut down Agent Builder, evals, and reusable prompts on November 30, and developers need to export and migrate soon.

OpenAI announced it will shut down three developer features on November 30: Agent Builder, the evals dashboard, and reusable prompts. evals has already entered read-only mode, allowing only historical data to be viewed while preventing new or modified evaluations. The official migration path is: move agents to the Agent SDK, move evaluations to the third-party tool PromptFu, and manage prompts back in developers' own codebases. This combination indicates OpenAI prefers developers to build agents with code rather than through the platform interface. The article recommends developers complete inventory and export of three types of assets before November 29 and reserve time for converting graphical orchestration into SDK calls and recalibrating evaluation baselines and prompt assets. It also notes that third-party platform assets should be treated as having expiration dates, requiring regular backup habits.

OpenAI is scaling back its developer toolchain. According to a short video report from the YouTube channel HopCommit, OpenAI will shut down three features that developers actively use on November 30: Agent Builder, the evals dashboard, and reusable prompts. Among them, evals has already entered a read-only state, meaning the window for teams to export their data is running short.

Agent Builder, evals dashboard, and reusable prompts

What Are the Three Features Being Shut Down

Agent Builder is OpenAI's visual/low-code entry point for building agents, where developers use it to assemble tool calls, orchestrate flows, and define prompt logic; the evals dashboard is used to manage evaluation sets, run evaluations, and view results; reusable prompts is a feature that turns commonly used prompts into shareable assets. Together these three capabilities cover the common stages of agent development—from building and evaluation to prompt management—so the shutdown directly affects teams that have already embedded these features into production workflows.

One detail worth noting: evals didn't stop serving only on November 30—it has already switched to read-only mode. According to the same source, this means you can still view historical evaluation data but can no longer create or modify evaluation content. For teams that rely on evaluation results for version comparisons, exporting should start as early as possible.

evals is already in read-only mode

The Difference Between Agent Builder and Agent SDK

Agent Builder is a visual orchestration layer where developers define an agent's tool-call order, conditional branches, and prompt templates by dragging nodes and filling out forms, with the output hosted on the OpenAI platform. The Agent SDK, by contrast, is a code-level development library where developers use Python or TypeScript to directly write an agent's control flow, invoke tools, and manage session state, with the output being a program that runs in their own environment. The core difference is that the former stores orchestration logic on the platform side, while the latter stores logic in the code repository, and thus naturally has version control, code review, and cross-environment deployment capabilities. The essence of migration is not switching to a different interface but rewriting "configuration" as "code."

What Are evals (Evaluations)

evals (evaluations) refers to the systematic assessment of the output quality of large model applications: first prepare a set of test samples with expected answers or scoring criteria (an evaluation set), then have the model or agent run through them one by one, and finally score and retain results according to metrics such as accuracy, relevance, and format compliance. Evaluation sets typically expand continuously as the business iterates, and historical results are used to compare the effectiveness of different prompt versions or model versions. OpenAI's evals dashboard productizes this process, providing evaluation set management, batch runs, and result visualization, so once it enters read-only mode, teams can no longer add new test cases or rerun baselines.

OpenAI's Suggested Migration Path

OpenAI's official recommendations are quite clear, which also indirectly shows that these features will not be retained as-is. According to HopCommit's citation, the migration direction falls into three parts: migrate agents to the Agent SDK; migrate evaluations to PromptFu; and manage prompts back in the developer's own codebase.

OpenAI's official migration recommendations

The signal this combination sends is that OpenAI prefers developers to build agents with code (the SDK) rather than through the platform interface, hands off the evaluation stage to third-party tools, and returns prompt management to the code repository. For small teams that have long relied on the visual platform for prototyping, the migration cost mainly falls in two places—translating graphical orchestration into SDK calls, and rebuilding evaluation sets and prompt assets after exporting them from the platform.

Migration direction: Agent SDK, PromptFu, and your own code

What Is PromptFu

PromptFu is a third-party prompt and evaluation management tool positioned to take over the evaluation workflows originally completed in OpenAI evals: it supports creating evaluation sets, defining scoring dimensions, running batches, and comparing results across different prompt versions. Treating it as the official recommended destination means OpenAI will no longer operate evaluation as a built-in platform capability but will hand it to third-party products focused on this area. For teams, migration is not just importing data; it also requires reconfiguring scoring criteria and baseline thresholds, and the comparison dimensions previously built up in evals may not map one-to-one.

Actions Developers Should Complete This Week

Time is the most concrete pressure. HopCommit's advice is direct: export everything this week, don't wait until November 29. Specifically, you need to inventory three types of assets: the agent logic already configured in Agent Builder, the evaluation sets and historical results in evals, and the prompt text in the reusable prompt library. In read-only mode, the exportable scope of evaluation data may be limited, so the earlier you act, the safer you are.

The rebuilding work after export also needs to be scheduled. Migrating to the Agent SDK is not simple copy-and-paste; it involves re-expressing the original orchestration logic. After evaluations move to PromptFu, scoring criteria and baselines also need to be recalibrated. Handling these two things in the same sprint is more manageable than scrambling after the services are shut down.

What This Shutdown Says

The simultaneous retirement of three features reflects a trade-off between platform-style features and developers' own workflows. Visual building lowers the barrier to entry, but when developers need stronger version control, auditability, and portability, the combination of code and third-party tools is often more suitable. By pushing agent development toward the SDK and handing evaluations to external tools, OpenAI is effectively giving the parts that caused "platform lock-in" back to developers.

For developers, the lesson from this kind of shutdown event is universal: any prompts, evaluation sets, and orchestration configurations stored on third-party platforms should be treated as assets with an expiration date. Building the habit of regular exports and versioned backups is more cost-effective than rushing to migrate before a deadline.

Platform Lock-in and Portability

Vendor lock-in refers to a state in which business capabilities are deeply dependent on a vendor's proprietary features, resulting in high migration costs and limited bargaining power. Visual building tools are most likely to cause lock-in: orchestration logic is stored in a platform's proprietary format and cannot be directly exported as runnable code, so once the feature is discontinued, teams can only rewrite it manually. Portability, by contrast, requires assets to exist in general-purpose forms—prompts stored as text files in a code repository, evaluation sets stored as structured data, and orchestration logic written as SDK calls. This shutdown is precisely pushing developers from the former toward the latter, and it also reminds teams to prioritize evaluating "whether assets can be fully exported" when choosing tools.

FAQ

Can Agent Builder still be used after the shutdown? According to available information, the feature will no longer be available after November 30, and the official recommendation is to migrate to the Agent SDK.

Can I still create new evaluations in evals now? No, evals is already in read-only mode, and you can only view historical data.

Where should prompts be placed? OpenAI's recommendation is to manage them back in the developer's own code rather than continuing to rely on the platform's reusable prompts feature.

One-line Summary

OpenAI will shut down Agent Builder, evals, and reusable prompts on November 30, and developers need to export and migrate as soon as possible.

#OpenAI#Agent Builder#evals#PromptFu#Agent SDK#developer migration
Share:

Related articles