Appian Q3 Release Highlights: AI Speed Meets Enterprise Controls

AI is evolving faster than organizations can keep up, and often it’s the most important operations that fall the furthest behind. That’s exactly why Appian is focused on AI for complex, enterprise-grade processes—to make applying AI automation to important work easy.

Over the last quarter, Appian has continued to invest in capabilities that help you deploy AI fast without sacrificing control:

• AI-assisted development → 

• Controls for AI agent automation → 

• Agents that understand your operations → 

• Verify documents with agents → 

• Connect with anything using agentic RPA →

Over 80% of developers regularly use AI coding tools. That translates to about 40% of enterprise code today being generated with AI.

The benefits are too great to ignore: faster time to market, better adaptability, and increased IT bandwidth. But for regulated environments AI-generated apps can also create new risks.

If using AI tools for developing mission-critical applications feels impossible, Appian’s AI-assisted development approach can help. Composer gives you the speed of AI-assisted development with the control enterprises need.

It starts with requirements: Appian uses AI to extract business logic from system documentation and operating procedures, and turns it into clear requirements that business and IT can review before anything is built. From there, AI develops an implementation plan, allowing developers to review what needs to be created and reused to make the new application a reality. 

Once the plan is improved, AI implements the changes. AI creates everything using managed Appian objects, so capabilities are secure, pre-tested, easy to review, and deployable through the pipeline your developers already use. And while these capabilities are all natively available in Appian through Appian Composer, developers who prefer to work in their own AI coding tools can connect through MCP and work in the same environment without ever leaving the command line.

In this release, we’ve enhanced the user-story experience, so teams can now use AI to implement individual user stories at any point in the application lifecycle. New progress indicators show at a glance how many implementation tasks have been completed for each story. And Composer can now execute each task in a story in sequence, so you can review what it produces without having to supervise every step.

We are also introducing Plug-in Composer, allowing you to build custom plug-ins in minutes. Just install the extension in IntelliJ from the Appian AppMarket, authenticate to your site, and describe the plug-in you want. In minutes you get a project with the correct SDK version, resolved dependencies, build files, and implementation logic that all adhere to plug-in compliance standards.

These enhancements mean more developer productivity and faster application delivery.

Most organizations are still figuring out where to use AI agents, how much they’ll cost, and whether they can be deployed without sacrificing compliance. Appian’s Q3 innovations focus on giving organizations the controls they need to deploy agents in a compliant manner, gain greater visibility into agent activities, and reduce agent runtime costs.

Rather than forcing users to leave their workspace to engage with an AI agent, you can now embed conversational agents inside a user’s existing Appian workspace. For example, an employee reviewing expense reimbursements can hand a receipt to the agent, watch it extract the details, check them against policy, and prepare a recommendation, all in one place. This keeps agents in the flow of work, making people more productive.

Agent costs are commonly driven by the amount of context an agent is given, and that cost is compounded with each step an agent takes. In Q3, Appian introduced two new capabilities to help you reduce that cost:

• Automatic prompt caching caches the static parts of an agent’s context at the start of every run (like system instructions and tool schemas), so each step processes only what's new.

• You can now trim what a process tool sends back to the agent, excluding input parameters and hidden process variables, so that agents only work with the context that is relevant.

None of these enhancements change what your agents do. But they reduce what they cost to run.

Whatever safeguards you put in place, you need to see exactly what an agent did and why. Appian has introduced several new capabilities that give you more control over agents: 

• See how agents are using tools the moment each tool finishes, including duration, inputs, outputs, and reasoning. Rating each output helps tune how the agent behaves.

• Search across agent trace history to find the values for each run you care about in seconds. 

• Stream guardrail violation logs to Amazon S3 or a syslog receiver, providing you with centralized observability in the audit systems you already have.

Stream AI agent guardrail violation logs to Amazon S3 or any syslog receiver.

Your AI strategy has to fit your compliance and commercial commitments. Appian AI now supports even greater model and provider choice, including the ability to run Appian AI completely self-managed, with no cloud connection required: 

• Model choice. You can now use Gemini models in addition to the existing support for Claude, GPT, and Nova. Appian can now auto-route capabilities to the best available model, so your workflows keep running even when models are retired.

• Provider choice. You can now run Appian inference through Google Vertex AI in addition to Azure AI Foundry, Amazon Bedrock, or your own custom LLM endpoint. We also added support for the OpenAI Chat Completions protocol, so you can route Appian's AI traffic through gateways like Portkey, LiteLLM, and Kong, keeping model governance, cost tracking, and audit in one place.

• Self-managed. Appian’s AI capabilities can now be powered by self-managed inference. Connect your model and empower teams to build with Composer, analyze data through conversational interfaces, and deploy AI inside processes for task automation—no AWS dependency and no internet connection required. For regulated industries like defense, government, and financial services, that means no request ever has to leave your compliance boundary. 

These enhancements help prevent AI cost overruns, reduce cloud vendor lock-in, and provide the audit-ready governance required for highly regulated operations.

Agents, like people, need operational context to make effective, well-grounded decisions. Appian provides that context through two core capabilities: 

• Process intelligence shows you how work actually flows across teams and systems, with cycle times, process variants, and bottlenecks all surfaced for easy consumption. 

• Data fabric connects disparate records and documents into a unified ontology, with rich metadata to help agents find and interpret the information they need.

This quarter we made both sets of context easily accessible to Appian and third-party AI agents. Appian AI agents gained system tools that performantly join data across complex data relationships. External agents access the same context through MCP Server. Record-level and field-level security ensure that data access is securely governed, mitigating the risks of data leaks and making compliance audits seamless.

We’ve also made enhancements to data fabric that give agents and developers more precise answers:

• New system tools let an agent resolve complex queries in one operation rather than sequencing multiple data queries to build a complex picture. That's fewer AI actions with no additional development time.

• Focused smart search limits semantic search to the records that matter, like a single customer's cases, ensuring these searches remain cost-effective with low latency.

• Federated queries now include Oracle as a supported source, so you can connect directly to Oracle data sources and configure relationships and record-level security without syncing any data.

And since high-stakes operations benefit from both human and AI oversight, this context is visible to your people, too:

• The record knowledge graph component lets users see how data connects together, visualizing complex relationships to help people make better decisions.

• Business-friendly display names ensure what users see matches their operations.

• When a relationship returns many records, the graph groups them into a single cluster that users can easily browse individually. 

• Scorecards combine multiple process KPIs into a composite score, so an operations leader can see performance against targets at a glance. 

• Assisted analysis enables analysts to drill into KPI performance. Ask about bottlenecks and receive activity and sequence durations with suggested actionable steps, so every analyst can follow the trail to a root cause.
  

A lot of document work isn't about pulling data out. It's about checking whether a document meets your requirements.

With new verification models for DocCenter, you can define business rules in plain language and reference materials, enabling AI agents to autonomously validate documents. Agents return a pass or fail result for every rule, citing the reason for each judgment. And the results return to your process, routing exceptions to human review and keeping everything else moving.

Organizations have spent decades accruing older systems, and people have spent as much time learning how to work with them. They master the green screens, swivel-chair between applications, and codify workarounds in their heads.

Agents lack this institutional knowledge. 

And according to Harvard Business Review's research, 69% of organizations say legacy systems are limiting their ability to scale AI. 

To use agents in the processes that run on legacy systems, those agents need a reliable way to connect with those same green screens. And that’s exactly what Appian provides.

Agentic RPA (available in the 26.10 release) brings the power of computer use agents to Appian RPA. Agentic RPA overcomes scripting shortfalls: when a step fails, an agent takes over. Using computer vision and LLM reasoning, the robot can intelligently move past the block, and then continue again with the deterministic script. Agentic RPA also empowers RPA bots to navigate applications they've never seen before. 

This means you can automate actions in systems that only offer a legacy user interface, with greater resilience than scripted bots can provide alone. And this power pairs with other process and governance controls available on the Appian Platform.

Organizations have spent decades accruing older systems, and people have spent as much time learning how to work with them. They master the green screens, swivel-chair between applications, and codify workarounds in their heads.

Agents lack this institutional knowledge. 

And according to Harvard Business Review, 69% of organizations say legacy systems are limiting their ability to scale AI. 

To use agents in the processes that run on legacy systems, those agents need a reliable way to connect with those same green screens. And that’s exactly what Appian provides.

Agentic RPA (available in the 26.10 release) brings the power of computer use agents to Appian RPA. Agentic RPA overcomes scripting shortfalls: when a step fails, an agent takes over. Using computer vision and LLM reasoning, the robot can intelligently move past the block, and then continue again with the deterministic script. Agentic RPA also empowers RPA bots to navigate applications they've never seen before. 

This means you can automate actions in systems that only offer a legacy user interface, with greater resilience than scripted bots can provide alone. And this power pairs with other process and governance controls available on the Appian Platform.

Since many important processes still run through mainframe applications, we are also introducing new RPA actions to make those connections easy to configure. Mainframe automations can now be added to any robotic task with drag-and-drop. Screen timing, one of the hardest things to get right when automating mainframe UIs, is handled for you. And the mainframe interaction is expressed in plain language, so developers never have to read the underlying code.

These enhancements connect core legacy applications to modern AI workflows, significantly lowering RPA maintenance costs while improving automation coverage.

Watch the Product Innovation Highlights Webinar below or read the 26.9 Release Notes.