Review contributions from AI

Give people and AI a shared body of knowledge that improves with each reviewed contribution.

Why the working environment matters

Software teams give AI more than source files. Git supplies branches and commits. Pull requests, automated tests, continuous integration, and code reviews help teams check changes before accepting them. Issues, documentation, and architecture decision records preserve the reasons behind decisions. People and agents can contribute in parallel, inspect history, and build on each other's accepted work.

That is a useful model for other kinds of knowledge work. A folder of documents can hold the source material, while relationships, decisions, and the reasons for changes remain scattered across conversations and individual memory.

Legra combines structured and unstructured data in one encrypted workspace. Branches separate proposals from accepted knowledge. Commits, queryable history, journaling, and provenance preserve the record needed to inspect changes and their sources. Your team's review process decides which contributions become shared knowledge.

Inherit: start from accepted knowledge

Begin with the current shared branch in the intended workspace. Use its documents, facts, relationships, and recorded decisions as context. Follow sources and history when a statement needs explanation.

Agree on the task and the checks that will establish whether the result is useful. For a research task, those might include source coverage, consistent names, and evidence for each new claim. Existing checks can catch regressions when later work changes previously reviewed facts.

Work: propose and review on a branch

Create a branch for the task. People and AI can prepare changes there while other work continues on separate branches. Keep new claims connected to their sources and record the reason for the proposal so a reviewer can assess it.

Run the relevant queries and validation, compare the proposal with the shared branch, and review the outcome. Select the intended branch explicitly when giving an AI client a task. Workspace permissions and client approval controls still determine what it may do.

The Branching guide provides the commands; AI tools with MCP explains how to connect a client.

Compound: carry improvements forward

Merge accepted changes into the shared branch. Preserve useful findings, sources, decisions, and reusable checks in the workspace. The next person or agent can begin with those improvements and inspect the history behind them.

The benefit accumulates as reviewed contributions become shared context. Review remains essential: accepting more material only helps when it improves the knowledge that future work depends on.

Example: a law firm's research

A firm keeps matter documents alongside structured facts about parties, agreements, and events. An AI works on a research branch to propose a timeline and connect each event to supporting documents. A lawyer checks the evidence, corrects the proposal, and approves the changes for merging.

In a later session, another lawyer or agent starts with that reviewed timeline, its sources, and its change history. Further findings can extend or correct it through another branch and review. The firm's knowledge improves across sessions, with a record of how it changed.

Try the workflow with the Review data changes tutorial.