Over the past two years, almost everything we’ve seen of AI in enterprise applications looked the same: a chat window in a corner. Useful for answering questions, not very useful for getting work done. The next step is already here, and it has a somewhat dry technical name but a very clear idea behind it: agentic UI with Blazor, that is, interfaces where an AI agent doesn’t just respond, but executes multi-step tasks, proposes changes and collaborates with the user inside the application itself. Microsoft has published the new AI components for Blazor on the official .NET blog (devblogs.microsoft.com/dotnet/build-agentic-ui-blazor), and we believe it deserves a read from a business perspective.
From the chatbot that answers to the agent that acts
A chatbot returns text. An agent gathers context, invokes tools (checking stock, booking an appointment, generating a document), proposes a plan and moves forward step by step alongside the person.
That leap brings three design problems that anyone who has tried to put AI into a business management application will recognise instantly:
- The response arrives in pieces. The user sees text appearing and doesn’t know whether the system is thinking, waiting or has frozen.
- There’s a lack of visibility and control. If the agent is about to place an order or send an email to a customer, someone has to be able to approve or cancel it.
- State is shared. The document, the quote or the plan is touched at the same time by the person and the agent. Who’s in charge?
The usual answer — building all of that by hand with JavaScript and patches — is expensive to build and even more expensive to maintain. That’s where Microsoft’s proposal comes in.
What agentic UI with Blazor brings
The new components (package in preview, on .NET 11, according to the original article) turn the conversation with an agent into content blocks and observable state that the application can render and control like any other Blazor component. In plain terms: AI stops being a black box with text inside and becomes structured information that your application represents however it wants.
The pieces, without the jargon
- A complete chat component to get going in minutes, with a message list, text input, response status and retries.
- Typed blocks for each type of content: rich text, tool calls, pending approvals, actions within the interface itself.
- Low-level components for when you want your own screen rather than a chat: the AI feeds your design, not the other way around.
- Connection with remote agents via AG-UI, an open event protocol between agents and applications, together with the usual .NET ecosystem (ASP.NET Core, Microsoft Agent Framework, Aspire).
The practical consequence for a company: the agent can live on a server, the interface in the web application, and both can talk to each other using a standard instead of a handcrafted, one-off integration.
Five patterns that really do translate into value
The Microsoft article describes several example scenarios. These are the ones that, in our view, turn into real benefit fastest:
1. Structured responses, not a wall of text. The content as it arrives is transformed into well-formatted paragraphs, lists or tables. It may seem cosmetic; it isn’t. A user who understands at a glance what they’re looking at decides faster.
2. Tool results rendered as interface. If the agent checks a weather forecast, the availability of an item or a customer’s history, the application doesn’t show JSON: it shows a record, a card or a table with the house’s own design. And while it’s being resolved, it shows the "querying…" status.
3. Actions in the browser itself. The agent can ask the application to navigate to a screen, open a dialog or filter a list. The conversation and the application stop being two separate worlds.
4. Human-in-the-loop. For actions with consequences — confirming a booking, approving an order, sending a communication — the agent stops and waits for an Approve or Reject. This is probably the most important point for management evaluating AI: the agent proposes, the person decides, and everything is traced.
5. Shared state and changes under review. User and agent work on the same document. The agent can propose a new version that is shown as a draft compared with the original; if accepted, it’s consolidated; if not, it’s discarded. Nothing is modified behind anyone’s back.
A grounded example
Imagine a quote configurator at a distribution company. The salesperson writes: "prepare an offer for this customer with the material from last week’s job, applying their price list". The agent checks the history and the rates, proposes the lines, shows the draft alongside the current quote and highlights in yellow what has changed. The salesperson adjusts two prices, accepts, and only then is the document saved in the ERP.
That flow is exactly the shared-state pattern plus explicit approval. And it fits sectors where we already work: distribution, retail, construction, food industry, hospitality or fleet management.
Before diving in: four honest cautions
- It’s preview technology. The components are experimental and the API may change. It’s fine for prototypes and well-scoped pilots, not necessarily for the critical core of operations as of today.
- Without tidy data there’s no agent worth having. If the item master or the customer history is dirty, AI will amplify the mess. Data governance is the prior work, not an extra.
- Start with a measurable use case. Not "we want AI": we want to reduce the time spent preparing offers or the errors in order entry. With a metric before and after.
- Design the boundaries from the start. What the agent can do on its own, what requires approval and what is off limits. That conversation is a business one, not an IT one.
How we approach it at Tisa
We’ve been developing management software since 1987 and, as a Microsoft partner and ISV, we combine custom development, Power Platform and applied AI projects on a well-governed data foundation. Our approach to agentic UI is the same as always: identify a specific use case, validate that the data supports it and build a first version that can be measured.
Do you have a management application where an assistant that does things — and asks permission before the important ones — would save hours every week? Let’s talk with no obligation: (+34) 971 305 885 or info@grupotisa.com. We’ll help you separate what is already viable from what is still just a promise.
Source: .NET development blog, Microsoft — "Build Agentic UI with the new Blazor AI components" (devblogs.microsoft.com).