Product Newsletter - September 2026
September was about seeing more of what your AI Colleagues do. Performance Insights turns AOP runs into a dashboard you can shape in plain English, and stuck runs can finally be cancelled. The employee web app has a new face. Knowledge answers can now stream in as they are written, and Conflict Analysis surfaces clashes as it finds them, ranked by severity. And Workflows Studio now flags nodes that only work for one employee before they ship.
The short version
Performance Insights builds the dashboard for you, then lets you reshape it in plain English. In beta.
Clear runs that will never finish, one at a time or in bulk, with a reason on record.
Side rail, welcome screen, a proper notifications panel. Live now for multi-chat assistants.
Knowledge answers in web chat appear as they are written. On long ones, the first words land 3–4 seconds sooner.
Conflict Analysis only flags clashes the same people can see, filtered by severity, and exportable.
Workflows Studio catches employee IDs and emails hardcoded into a node before it goes live.
Preview real records before an automated sync goes live, and add people one at a time.
Twilio joins Infobip as a delivery provider. Existing accounts stay put unless you ask.
🧠 AI Colleagues Studio
Your AOPs got a scoreboard, and a Cancel button for your In progress runs.
Every AOP run, turned into numbers you can track
Beta. Run History tells you a run happened. It does not tell you how many expense claims your AOP approved last week, for which departments, or why the rest stalled. That is business data, and until now it sat inside individual runs.
Performance Insights pulls it out. Point it at a published AOP and it reads the AOP's tools, its helper AOPs and its real runs, then proposes what to capture and how to chart it. No SQL. No schema to define. No ticket to the data team.
- Pick a published AOP with at least 5 real runs. Evaluation runs don't count, so your test traffic stays out of the numbers.
- Review the proposal. Every chart is rendered with sample data before anything goes live.
- Adjust it in plain language. "Show me approvals by department" is a perfectly valid instruction.
- Go live. Capture starts, and the charts fill in as runs happen.
- Six chart types: KPI, trend, bar, donut, funnel and table.
- Two cards on every report, Run completion rate and Why runs didn't complete, calculated the same way everywhere so you can compare one AOP against another.
- Drill-down on every number. Click any chart to see the exact runs behind it.
- Edit with AI. Describe changes to the proposed report in plain English.
- One landing screen listing every AOP by where it stands: Live, Action needed, Ready for reporting, or Paused. Helper AOPs roll up into their parent's report.
| You... | Your report... |
|---|---|
| Make a minor edit | Updates in place. Nothing to do. |
| Make a substantial change | Pauses capture and proposes a new version, with every change marked added, modified or deleted. Earlier versions stay frozen and viewable. |
| Switch reporting off | Keeps everything captured so far, ready for when you switch it back on. |
Set up Performance Insights → · Manage your reports →
A Cancel button for runs that will never finish
A run waiting on an approval nobody is ever going to give used to sit in Run History as In progress indefinitely, making your numbers quietly wrong. You can now cancel it.
Cancel one run from its row menu or the Execution details header, or choose Cancel runs from the Run History table menu to select several at once. Only in-progress runs can be cancelled, and only by people with AI Colleague Admin permission. Everyone else still sees the runs, just not the button.
What happens when you press it
- ✅ The run stops, along with any helper AOPs it called and any loops still handing out work
- ✅ It stays stopped. A late approval or callback cannot bring it back to life
- ✅ People in the run are told it was cancelled. Pending approvals and inputs close quietly, and anyone who acts on one later is told why, so nobody gets a flood of notifications
- ✅ Who cancelled it and why is on record: in the run header, on the Cancelled chip, and as columns in Run History. A note is optional for one run and required for a bulk cancel
Also in Studio
Hundreds of tools, found in seconds
Past a hundred tools, scrolling stops being a strategy.
Tags. Add free-text tags to any tool from its create or edit form, with suggestions drawn from tags already in use on that AI Colleague. The Tools list shows a Tags column, and search now matches tags as well as names. Tools built from a workflow share tags with that workflow in both directions, and existing ones have been filled in for you.
Filters. A new Filters panel narrows the list by Category (Custom or Built-in), Type (Workflow, API or MCP), availability (Global or AIC-specific), Tags, Execution scope, and Application, meaning the app behind an MCP or Workflow tool. Filters stack with each other and with search, and they live in the URL, so a refresh keeps your view and a shared link shows a colleague exactly what you are looking at.
Build AOPs from the documents you already have
Most processes are already written down somewhere, in an SOP or a policy PDF. Retyping them into a prompt was the worst part of Create AOP via AI.
You can now attach files alongside your prompt in both Create AOP via AI and Improve AOP, with the same file types and limits as the manual builder. Each file is handled on its own merits:
| If the file is... | Leena AI... |
|---|---|
| Needed while the AOP runs, like a rate card or a limits table | Saves it on the AOP, where every run can read it |
| A description of the process, like an SOP | Works it into the instructions, then lets the file go |
Your prompt always wins. Say what you want done with a file, and that is what happens. Improve AOP also shows the AOP's existing attachments for reference, and both flows show live progress while they work: selecting resources, reading attachments, generating, finalising.
💻 Employee Web App
We promised it for mid-September. It is live.
The web app got a new face
The web app's layout dates from when an assistant mostly answered questions. Today it runs conversations, tasks and notifications side by side, and the layout needed to catch up. The redesign keeps everything your assistant can do and changes how it is arranged around your employees.
New chat, assistant switching, quick links, past conversations, notifications and settings, one click away. On mobile it folds into a drawer.
A centred starting point with a personal greeting and up to six suggested prompts.
Slides in from the side, with an unread filter and Mark all read.
Each file shows its type, size and a preview where available, and long names keep their extension visible.
A voice call sits where you type, instead of covering the conversation.
Font size, privacy terms and clearing chat history, in one small flyout.
Your colours and chat background come with you, and so does every message type: cards, forms, quick replies, webviews, streaming and voice. The new layout carries across to the embedded widget and the Teams and mobile apps too.
Can I switch today?
| If your assistant... | Then... |
|---|---|
| Already uses multi-chat | Yes. Turn on Enable webapp V3 under Admin console → Leena Channels → Settings → Display & interactions. Turning it off takes you straight back. |
| Uses single chat | Soon. The new layout is fully conversational, with no intro flow and no menu, so your SPOC will plan the move and the testing with you. |
| Runs Onboarding, Offboarding or Engage journeys | Not yet. These stay on the current layout for now. |
📚 Knowledge Management
Answers stopped keeping people waiting, and Conflict Analysis stopped crying wolf.
Answers that start before they finish
Ask the knowledge base a question and you used to watch a spinner until the entire answer had been written. Now answers stream in as they are generated, the same way AI Colleague responses already do.
- Words sooner. On longer answers, the first words appear 3–4 seconds earlier.
- A window into the search. The reasoning block explains, in plain language, what the knowledge search is doing while you wait.
- Fewer needless "I don't know"s. Questions are now routed more accurately, and we retired a rule that swapped any answer containing a hedge for a flat "I don't know".
Rolling out gradually. Streaming knowledge answers are available for assistants with multi-chat and streaming switched on, enabled on request while we watch the first adopters. Ask your SPOC to turn it on. It becomes the default for streaming assistants after that.
Conflict Analysis, with a sense of proportion
Conflict Analysis used to make you wait for the whole run, then flag every clash it could find, including between articles no single employee would ever see side by side.
| Before | Now | |
|---|---|---|
| When you see results | When the whole run finished | As each conflict is found, with a progress bar showing how many article pairs have been checked |
| What gets compared | Every pair of articles | Only articles the same people can actually see |
| How much noise | All of it | Filtered by severity, with fewer false positives |
| Sharing it | Stayed in the dashboard | Export it for offline review and internal distribution |
Everything else in Knowledge Management
| What changed | Why you care |
|---|---|
| Simpplr connector | Your Simpplr knowledge base can now feed your assistant directly. Connect Simpplr → |
| SharePoint works out its own expiry date | Credential expiry reminders for SharePoint take the date from when your certificate was generated. You just choose who gets reminded, and how many days ahead. Set up reminders → |
| ServiceNow source links, your format | Set the format of the source article link employees see, per knowledge base. |
| Write tests whenever you like | Smart Testing now saves test cases without starting a run, so a run in progress no longer stops you preparing the next one. Smart Testing → |
| Related articles, your call | Show or hide related articles from Knowledge Management General settings. |
| Parsing, rebuilt | One parsing flow instead of two, so documents behave consistently. A document parsed for one assistant can be reused by your others instead of being processed again. And files over 16MB no longer fail silently. |
⚙️ Workflows Studio
A guardrail for "but it worked when I tested it".
Workflows that work for everyone, not just the person who built them
You build an Action node and test it with your own employee ID. It works. You forget to swap the ID for a dynamic reference, and publish. It still works, perfectly, for exactly one person.
Action and Function nodes are now checked on save for employee and company details typed straight into the configuration. How loudly the check objects depends on what it finds:
| What it finds | Flag | What happens |
|---|---|---|
| Employee ID, email, name or phone | 🔴 Error | The first save is blocked and the field is highlighted with the value it found. Swap in a dynamic reference, or click Save again to keep it as it is. |
| Department, designation, location or OU | 🟡 Warning | The node saves. A dynamic reference is recommended. |
| Employee data in a Function node's test output | 🔵 Info | Nothing to fix. That is expected when a node works with employee data. |
The check covers everywhere a value can hide: the API request, input fields, output mapping and success or error messages on Action nodes, and the code, arguments, output mapping, messages and test output on Function nodes. Edit the node and it checks again.
🛠️ Platform & Global Settings
Setup screens that show their work.
Employee sync, with a test drive
Saving an automated sync used to start it straight away, so you found out whether the data looked right after it had already landed. Setup is now a guided flow, with a testing step in the middle.
- Preview before you commit. See a sample of the records your source will send, with problem rows flagged, such as records with neither an employee ID nor an email, which a real sync would skip. Nothing is written until you start the sync.
- Access settings, where you would look for them. Authorisation settings now live inside Employee sync, so when someone cannot get in, the explanation sits next to the data behind it.
- See who has actually signed in. One toggle switches from synced employees to authorised users, the people who have signed in to the assistant, along with their channel details.
- Add one person without a spreadsheet. New employees can be added through a form instead of an Excel upload. Automated syncs keep taking their employees from the source.
Virtual assistant settings, decluttered
Leena Channels → Settings has been rebuilt for a cleaner setup, and the settings almost nobody used are gone. Three things are new:
- File uploads: decide whether employees can attach files, and set limits on size and number.
- A disclaimer banner, for the line your legal team would like every employee to see.
- Quick replies next to webviews, so both can go inside the same folders.
Customise your virtual assistant →
WhatsApp, now via Twilio too
WhatsApp messages can now be delivered through Twilio as well as Infobip. New accounts start on Twilio. Existing accounts stay on Infobip, and nothing changes unless you ask. Want to switch? Your SPOC can arrange it today, and a switch in the dashboard is on the way.
🔢 September, by the numbers
Lines of SQL needed to build a Performance Insights report.
Real runs an AOP needs before its report can be set up.
Chart types, plus two built-in cards on every report.
Sooner the first words of a long knowledge answer appear.
Setting between a multi-chat assistant and the new web app.
WhatsApp delivery providers, up from one.
👋 That's September
Want a walkthrough of any of this, or help planning the web app switch? Your Leena SPOC is the person to ask: your Implementation PM if you are still onboarding, your CSM once you are live. Leena AI support is always there too.