What's New: AI Development Toolkit (2026 release wave 1)

Overview

The toolkit formerly known as the development tools for extending Copilot is now called the AI Development Toolkit. It covers using Azure Open AI from AL code and also designing and coding agents in Business Central. Agent development has been in public preview since February, with almost 1,500 agents built by the community.

The video walks through authentication changes for BC AI Resources, content filtering policies, prompt injection protection, and the renamed evaluation tool. Evaluations gain suite configurations, language categorization of test inputs, Copilot credit insights, and early agent evaluation support. Most of these were demoed.

Key points

• The toolkit was renamed AI Development Toolkit; the AI test tool is now called AI Development Toolkit evaluation.
• BC AI Resources no longer require your own partner subscription. The set managed resource authorization method has a new signature without subscription details; the old one is obsoleted but still works.
• Content filtering supports low and medium levels, with low as the default. High is not supported in managed resources and needs your own Azure Open AI subscription with approval. Levels are set through the Azure Open AI policy parameters codeunit.
• Prompt injection protection uses Azure AI content prompt shields. You must call the detection tags for untrusted prompts and enable XPI detection separately.
• Evaluation suite configurations define capability, test type and frequency (daily, weekly, manual). The frequency enum is extensible, and pipelines must adopt the configurations manually because they do not run evaluations automatically.
• Multi-language evaluations require name and language in data sets, and data set names must be common across languages. Copilot credit use shows per run, line and data set entry.
• Agent evaluation is in the early stages and lets you inspect agent tasks per run, line and entry. An agent library is announced as coming soon.

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 Open AI policy parameters” at 7:03
• codeunit “agent test test context code unit” at 16:37
• other “test context” at 16:49

Not found in BC28-30: codeunit “Azure Open AI policy parameters”, codeunit “agent test test context code unit”.

Quotes

• 0:24 “the toolkit now does not just support the ability to use azure open ai to use llms from al code”
• 0:55 “The latter has actually been in public preview since the beginning of February. We’ve already seen almost 1,500 agents being built by the community”
• 3:29 “we have now removed that requirement. So, when you authenticate or authorize in code using the set managed resource authorization we now have a”
• 5:38 “We have added support for only low and medium content filtering policies. Where low is the default and actually the the setup that will”
• 6:11 “We do not support the high setting, basically passing the most harmful content. That is something that you can have if you’re using your”
• 9:36 “the first thing you’ll probably notice is the terminology change from evaluation or from tests to evaluations. And this is also represented throughout the”
• 11:47 “you can set the run frequency to daily or weekly or manual. And this is an extensible enum, so you can extend with your”
• 13:20 “you can now see that they are categorized by languages”
• 14:34 “Copilot credit insights to help you understand AI consumptions for evaluations”
• 15:22 “This is still in the early stages, but now you can build and validate AI agents in BC”
• 15:47 “very soon we’ll have an agent library coming giving you even more capabilities to test agents”
• 17:39 “we now have changed the name of what was before the development tools for extending Copilot Business Central or just a Copilot tool kit”
• 17:54 “you no longer have to pass the your partner as your OpenAI subscription when you are using the built-in AI resources”
• 18:08 “flexibility you have for setting the Azure OpenAI policies specifically for content filtering on harmful content as well as um the X P I”

Disclaimers in the video

• 0:55 preview: The agent development capability has been in public preview since the beginning of February.
• 15:22 preview: agent evaluation support is still in the early stages
• 15:47 coming-later: agent library coming giving you even more capabilities to test agents
• 17:12 preview: agents that we just introduced in public preview

Presenters (as heard): Peter, Steen, Steve.