Most enterprise automation platforms still make you build one step at a time. Click, configure, save, click again. A single conversation can take hours before it does anything useful. Krista Studio just changed that math. Authors describe what they want in plain language, and the platform builds the conversation, asks the right follow-up questions, and hands back a fully editable, structured result. This release ships five AI authoring capabilities that cut manual configuration without taking control away from the author.
AI authoring in Krista Studio: the basics
AI authoring lets a person describe a business process in plain language and get a working conversation back. The platform reads intent, asks clarifying questions when it needs more detail, then generates a complete, editable step. Authors keep final say over logic and configuration on every generated step.
Building an enterprise brain has always meant translating business intent into technical configuration by hand. That translation step burns hours and introduces errors. Remove it, and the person who understands the process builds it directly.
AI Step Generation builds conversations from plain language
AI Step Generation reads a plain-language description, asks relevant clarifying questions, and generates a complete conversation step once the author confirms the details. It selects the right extension and the right request automatically, and every generated step stays fully editable afterward.
Consider the example in this release: an author needs a flow to trigger whenever a new email arrives. Instead of hunting through a list of extensions and requests, the author types the intent. Krista reads it, asks what it needs to know, and builds the trigger. The author never touches a configuration menu until they choose to fine-tune something.
The same pattern applies to Manage Information steps. An author asks to derive variables from a Mail Details entity. Krista does not generate blindly. It asks about field types first. Then it asks what logic should produce each value. Only after the author answers does it generate the fully configured step. If an author gets stuck on a term like field types, the built-in help chat answers the question without forcing a context switch.
Manage Info AI Assist derives and enriches data automatically
Manage Info AI Assist adds a panel inside the Manage Information step. Authors write instructions, select input values, and define outputs, all without manually mapping each field. The AI derives or enriches data based on those instructions, and a verify function tests the logic against sample inputs before anything reaches production.
The panel breaks into three parts. Instructions describe what the author wants calculated or derived, in plain language. Input values point to the data fields the AI should use. Outputs work in two modes: create adds a new field the AI will generate from scratch, and update modifies a field that already exists in the conversation. Authors can drag and drop these output rows to reorder logic as needed.
Before anything goes live, the verify button runs the prompt against sample inputs. That step matters. It lets an author catch a bad derivation before it reaches an end user instead of after.
The Solve a Problem model branches decisions by confidence score
The Solve a Problem model, found under the Ask Krista AI conversation step, lets authors define separate paths for high, moderate, and low confidence scores. Manage Info AI Assist takes a different approach: it throws an error automatically when confidence drops too low, keeping data accurate without extra logic.
Both features use AI to generate outputs, but they solve different problems. Manage Info AI Assist protects a single data field. Solve a Problem controls the branching logic of an entire decision. In the release example, an author configures Krista to rearrange emails by priority. The configure tab sets percentage thresholds for each confidence tier, and the author attaches different steps to each path. A high-confidence prediction moves forward automatically. A low-confidence one can route to a person instead. The author decides, case by case, whether the situation calls for flexible branching or a hard stop.
The Agentic AI step reasons, plans, and acts within guardrails
The Agentic AI step, currently in preview, lets an author describe an objective, a set of rules, and an expected outcome in plain language, then lets the AI decide which tools to invoke and when. It reasons, plans, and acts within guardrails the author sets, instead of following a fixed, predefined sequence.
This is the most significant shift in the release. Traditional conversation steps require an author to predefine every branch. The Agentic step inverts that. In the demonstrated example, an agent operates inside a support conversation with two tools. Authors equip it with a Jira Project Management request for creating tickets and an AI QA extension for answering questions from the knowledge base. The agent decides which tool the situation calls for based on what the user asks.
This is Krista as the fabric. One conversation coordinates a ticketing system, a knowledge base, and a human approver instead of forcing an author to stitch three separate tools together by hand.
Authors still hold the reins. The Process Controller lets an author set checkpoints that pause execution for approval, additional input, or human intervention before the flow continues. Results and Data Extraction capture outputs, like a Jira ticket ID or a knowledge base answer, and pass them into downstream steps. Execution Rules cap tool invocations, conversation calls, and request timeouts, so the agent operates inside boundaries the author defines rather than wherever the model decides to wander.
GenAI generates text. Krista generates outcomes, and the Agentic step is the clearest evidence of that distinction yet. An agent that reasons and acts still needs a record of what it did and why, plus a clear point where a person steps in. That is governance, not friction.
Automatic title, description, and icon generation
Krista generates conversation titles, descriptions, step titles, and custom icons automatically by reading the conversation’s steps and flow. An author clicks a single generate icon next to each field, and the AI produces output based on what the conversation actually does.
No manual naming. No searching an icon library for something that vaguely represents a workflow. The AI already understands what the conversation accomplishes, because it built it. Generating a name and an icon from that understanding takes one click, and it removes one more piece of busywork from the author’s day.
What’s next: the Conversation Builder
Krista is building a Conversation Builder that will let authors create entire conversations by explaining a use case in natural language, with no steps to configure and no canvas to build by hand. The goal driving this release stays the same: describe the outcome, and let the platform handle construction.
Call this tension the Guardrail Trap. Enterprises adopting agentic AI keep choosing between two bad options. Ship an agent with no guardrails, and the AI makes decisions nobody can audit. Wrap every action in manual review, and the process loses the speed that justified automation in the first place. Krista Studio’s AI authoring tools close that trap. The author sets the objective, the rules, and the checkpoints once. The agent works inside those boundaries every time after.
Build the enterprise brain faster. Keep it accountable at every step. That is the trade Krista Studio just made obsolete.
Transcription
Hello and welcome. Building your enterprise brain is made simpler with AI authoring in Krista.
In this video, we’ll walk you through all AI features, each one designed to reduce the manual effort of authoring, so you can focus on what your brain needs to do, not how to build it.
Here’s what we’ll be covering. AI Step Generation, enhanced and more powerful than ever. Manage Info AI Assist, to derive and enrich data with AI. Solve a Problem, AI-powered decisions with confidence paths. The Agentic AI step, AI that reasons, plans, and acts. And title, description, and icon generation: auto-named, auto-described, automatically.
Let’s get started with AI Step Generation. It isn’t new to Krista Studio, but we’ve significantly enhanced it, and it remains one of the most powerful ways to start building a conversation.
Instead of manually building out each step of a conversation, you just describe what you want in plain language, and the AI takes care of the rest.
Here we are asking to trigger a flow whenever a new email is received. The AI reads your intent and asks you relevant questions. Once you confirm everything and ask to proceed, it builds out a complete, structured conversation step.
You can see how the AI picked the right extension and selected the right request. And the best part: you’re not locked in. Every step stays fully editable, so you can make any further changes you need.
Let’s look at another example of AI Step Generation. Here we want to add a Manage Information step where we need to derive a few variables from the Mail Details entity. We simply describe what we want in plain language.
Once the prompt is submitted, notice what happens next. The AI doesn’t just generate immediately. It asks a set of relevant questions. First, it asks about the field types for each variable. Second, it asks what logic should be used to derive those values. It wants those details before it proceeds.
If you’re new to Krista Studio and don’t know something, like what Krista field types are, use the help chat right here to ask about them. Then come back to this screen and keep going.
Once you provide all the relevant information and ask the AI to proceed, it generates the step, fully configured and ready to go.
And here you can see something else worth noting: the Manage Info AI Assist feature.
Manage Info AI Assist is the second important feature we’re introducing in this release.
Let’s take a closer look. When you add a Manage Information step and open the AI Assist tab, you’ll see a clean panel with three things to fill in: your instructions, your input values, and your outputs.
Start by writing your instructions in plain language. Describe what you want the AI to derive or calculate. Then select your input values, the data fields the AI will use to produce the output.
In the outputs section, clicking create adds a new output row, where you define the instructions and set the output name and type for a brand new field the AI will generate. Clicking update adds a row for an existing field, so the AI can modify or enrich a value that’s already in your conversation. You can drag and drop rows to reorder them as needed.
Before going live, use the verify button to test your prompt with sample inputs and fine-tune the output before it reaches your end users.
Now let’s discuss the Solve a Problem model under the Ask Krista AI conversation step. Similar to Manage Info AI Assist, it uses AI to solve problems and generate outputs. But the key difference lies in how each handles the confidence of the AI’s predictions.
With Solve a Problem, you can define different paths based on the confidence score, giving you full control over what happens next depending on how confident the AI is in its answer.
With Manage Info AI Assist, the system automatically throws an error if the confidence score is too low, keeping your data accurate without any additional logic needed.
So depending on your use case, whether you need flexible branching or straightforward error handling, you now have the right tool for the job.
Now let’s go back to the Solve a Problem model configuration. Here we’re asking the model to rearrange emails by priority, and we’ll show you how to configure different paths based on the prediction’s confidence.
First, go to the configure tab. Here you can set the percentage thresholds for high, moderate, and low confidence. Once that’s set, we’ve added the required steps under each of the three paths, so the conversation responds differently depending on how confident the AI is in its prediction.
Next up, one of the most exciting AI features of the Studio: the Agentic AI conversation step, currently available in preview.
Unlike traditional steps where every decision is predefined by an author, the Agentic step lets you simply describe what you want the AI to accomplish: the objective, the rules, and the expected outcome, all in plain language. The AI takes it from there.
Here’s a real example. We’ve added an Agentic AI step inside a support agent conversation. In the instructions, we’ve defined exactly what the agent needs to accomplish as a help desk agent. Make sure the instructions are clear and descriptive, since they define the entire scope of work for the Agentic AI step.
To accomplish this, the agent has been given some input values and requests: Jira Project Management for creating support tickets, and an AI QA extension for answering questions from the knowledge base. The agent decides which tool to use and when, based on what the user is asking. To add input values and requests, click the plus icon.
Now let’s scroll down to the Process Controller. Authors can configure checkpoints that pause the task for approvals, additional input, or intervention from a person before the flow continues.
The step also supports Results and Data Extraction, so any outputs the AI generates, like the ticket ID from Jira or the answer from the knowledge base, can be captured and passed seamlessly into the next steps of the workflow.
And finally, Execution Rules give you guardrails to control AI behavior: setting limits on tool invocations, conversations invoked, and request timeouts, keeping everything within the boundaries you define.
This is Krista operating as a true AI agent: reasoning, acting, and delivering results. And since this feature is currently in preview, there’s more to come in upcoming releases.
Now let’s move to the final AI feature: Conversation Title and Description Generation, and Step Title Generation. You can also generate intuitive icons for your conversation.
All of this happens by clicking the generate with AI icon next to each field. Based on your conversation steps and flow, once you click generate, the AI automatically produces the title, description, step titles, and a custom icon that visually represents what your conversation does.
No manual naming, no searching for icons. The AI reads your flow and takes care of it all for you.
That’s a look at the AI capabilities now available inside Krista Studio, all designed to make conversation building faster, smarter, and simpler. If you’d like to go deeper on any of these features, in-depth training videos are available to Krista customers.
Before we go, we’re excited about what’s coming next in Krista. Here’s a quick look at an upcoming feature: the Conversation Builder. With Conversation Builder, you can create entire conversations by simply explaining your use case in natural language. No steps to configure, no canvas to build. Just describe what you need, and Krista builds it for you. Your enterprise brain, built even faster.
Thanks for watching, and stay tuned. The best is still to come.
What is an LLM Gateway? Prev post 
