What's New: Business Central AI Resources (2025 release wave 1)

Overview

The video explains Business Central AI Resources, Microsoft-managed Azure OpenAI resources that let publishers use large language models without procuring and managing Azure OpenAI subscriptions for each customer. In this release they move from private preview (invitation only) to public preview. Consumption-based direct billing of customers is being added, though billing details were not ready at recording time.

It then covers how developers use them. The set managed resource authorization function takes an account name, API key and deployment name. Bring-your-own subscription is meant for development, testing and special cases. A conditional code pattern switches between the two by environment. The video also covers content safety, data residency, model deprecation telemetry, and current limits such as no embeddings, no data sources and no non-OpenAI models.

Key points

• AI Resources are Microsoft-managed Azure OpenAI resources, available in SaaS production environments only, with GPT-4o and GPT-4o mini at recording time.
• The release moves AI Resources from private preview (invitation only) to public preview and adds consumption-based direct billing; billing details were not final at recording.
• Partners still use their own Azure OpenAI subscription for prototyping, development, testing and troubleshooting in non-production and non-SaaS environments, and must hold an active subscription for compliance verification even though its resources are not consumed.
• Use the set managed resource authorization function (account name, API key, deployment name) for AI Resources; use the set authorization function for bring-your-own. Use the environment information codeunit to switch between them.
• Content safety adds a blocking policy (hate, violence, sexual content, self-harm, jailbreak, cross-prompt injection) and system prompts, which can change model behavior slightly versus on-premises testing. Test in a sandbox before customer deployment.
• Telemetry signals warn of model deprecation. Deadlines are stricter than AL deprecation, and an unmigrated feature may be moved to a new model and behave differently.
• Not supported now: embeddings (semantic search planned in roughly half a year to a year), data source capabilities (no retrieval augmented generation), and non-OpenAI models such as DeepSeek.

Features

AL objects mentioned

As heard in the captions. A name that matches one object page by exact type and name links to it; the others stay as named.

• codeunit “Azure AOI deployments” at 9:18
• codeunit 457 “Environment Information” at 10:44

Not found in BC28-30: codeunit “Azure AOI deployments”.

Quotes

• 0:39 “in this release, we’re moving BC or Business Central AI resources from private preview, that means by invitation only, uh to a public preview”
• 0:56 “we are adding consumptionbased uh direct billing of customers uh AI resource use as well”
• 1:35 “the business central ai resources will be the default and recommended approach but some publishers or customers might have special needs”
• 1:52 “partners still need to use their own subscription when they are prototyping, developing and testing and troubleshooting in non-production uh and nonsas environments for”
• 7:10 “we’re not using data to train models uh we’re not using data for example um to like improve Microsoft services and such uh we”
• 8:43 “we still require you as a partner to have an active Azure OpenAI subscription and this is mostly to make sure that you are”
• 11:10 “whenever possible, yes. So we as Microsoft of course commit to data privacy. So your data will stay within geography if there is the”
• 13:19 “different models might behave differently. So your feature might just not work as well with the new one”
• 13:32 “that is currently not possible neither by using the ai resources from the toolkit nor by bringing your own azure”
• 14:12 “we only expose to your partners capabilities that we as Microsoft use. And that’s because of course we want to keep them tested and”
• 17:06 “uh approach where the customers will build directly from microsoft”
• 17:19 “However, the details of that uh is not ready at the time of this recording. Um but keep an eye out for details after”
• 17:42 “when you use the business central AI resources Microsoft manages those resources but it also means that it’s available in SAS only and production”
• 18:56 “we do not support uh embeddings although we have plans in probably next release uh in a you know half a year to a”
• 20:45 “customers will be able to track usage search, set limits, and get a single detail invoice across all of the ISVS that are using”

Disclaimers in the video

• 0:39 preview: moving BC or Business Central AI resources from private preview, that means by invitation only, uh to a public preview available for all of you
• 1:52 not-in-this-release: partners still need to use their own subscription when they are prototyping, developing and testing and troubleshooting in non-production
• 6:07 coming-later: we have uh in the plans to have some even better protection
• 13:32 not-in-this-release: data source capabilities that Azure OpenAI exposes from our toolkit. That is currently not possible
• 14:12 not-in-this-release: we can expose non open AI models in our toolkit for example deepseek or other models that have come out recently. And the answer is we don’t support that currently
• 17:19 subject-to-change: the details of that uh is not ready at the time of this recording. Um but keep an eye out for details after launch to learn more about how the billing works.
• 18:56 coming-later: we have plans in probably next release uh in a you know half a year to a year’s time frame to provide some semantic search capabilities

Presenters (as heard): Peter Bo, Enrico Chibitan.