KnowledgeLake is document-centric AI workflow automation. That description puts the emphasis where our customers feel it most: on work reaching the right person and continuing to move until it is done. AI reads, classifies, and extracts. People review, decide, and approve. The value comes from how smoothly work moves between them.
Our Q3 2026 release, KnowledgeLake 4.3, is about that handoff. It focuses on how users pick up work, move through it, and finish it. Here is what users can now do, and why we built it this way.
Users can now assign, reassign, and unassign work from Monitor, Insight, Index, Tasks, and the Focused Inbox, all with one shared set of controls. Team leads can rebalance a queue in the middle of a shift, and nobody waits on a colleague's lock.
Underneath that is a change in how the platform thinks about ownership. Work items are now assigned to people instead of locked by them. A lock is passive. It records who opened something last, not who is responsible for finishing it, and when that person steps away, the work stops. Assignment is intentional. It says who owns the work, and someone with the authority to change that can do so.
Index adds My, Available, and Manage Work Items tabs, so users can see what is theirs, what is open, and (with the right permission) what the whole team is carrying. A new Manage Assignments permission decides who can hand work to others. One consistent way to assign work across every app also means users learn one behavior instead of several. Consistency is an underrated productivity feature.
Users can now work QC tasks and standard tasks side by side in the Focused Inbox, and Continuous Mode opens the next task the moment one is finished. QC in the Focused Inbox is generally available and out of preview.
Human review is part of the workflow, not a step bolted on after it. The corrections people make are how the process stays accurate and how it improves over time. So, QC work now gets the same tools as everything else: saved views, filtering, sorting, grouping, and assignment.
Continuous Mode is the piece I am most excited about. It works across QC and standard work, so a validator who is in a rhythm stays in it. Anyone who has done high-volume review knows what small interruptions cost. The click back to a list, the scan for the next item, the moment spent deciding what to pick up. Across thousands of tasks a day, those moments add up.
Workflow designers get more control too. The Start a QC Task step can now pass metadata as search terms and limit who can pick up a task, so specialized or sensitive review goes to the right people. Work items and documents also record who completed a QC task and when.
In AI-extracted tables, users can now see which cells are low confidence and confirm them without retyping. Every table opens in one rebuilt editor, and inline list and table editing in the index panel is generally available.
This is what AI and people working well together looks like. The model does the bulk of the extraction and flags where it is unsure. The user spends attention exactly there, and confirming a correct value takes one action instead of re-entering it.
Tables are where much of that attention goes: invoice line items, transcript course rows, statement transactions. The rebuilt editor lets users drag to reorder rows and columns, delete rows together, resize columns, and undo any edit. Line items and course rows get cleaned up in far fewer clicks.
Workflow designers with the new Manage Secret Workflow Variables permission can now mark a global variable as Secret, so keys and connection strings are not readable by anyone with Edit access.
Source Search gets a related refinement. A column can be hidden from the results grid while still filling the mapped index field, so users see only what they need to make a decision and the full record is still captured.
This update addresses many customer-found defects. The full list will be in the 4.3 release notes, which are on their way.
Every change in 4.3 serves the same goal. Work reaches the right person, that person has what they need to finish it, and the next piece of work is already waiting. AI handles what it does well, people handle what they do well, and the workflow connects the two.
That is what document-centric AI workflow automation should feel like for the people doing the work. If you are running a document workflow and want to compare notes on where work gets stuck, I would be glad to talk.