Venture · Enterprise SaaS · AI

ContextOS AI recommendations, with the evidence in view.

Inspect why a risk was flagged, check the source quality, then confirm or disagree. The decision stays in the timeline.

Open live ContextOS prototype

Interactive prototype · Simulated sources · No production integrations

AudienceB2B

DomainEnterprise SaaS · AI

RoleSolo product designer

PlatformResponsive web

Year2026

ScopeStrategy · Interaction · Prototype

How review works

Every recommendation carries the evidence needed to challenge it.

The drawer treats confidence as a signal, not an answer. It keeps the sources, supporting evidence, permissions, and review controls together so a teammate can question the recommendation and understand the decision later.

Source trail shows where the signal came from.
Supporting evidence shows which connected artifacts point to the same issue.
Confidence includes a reason, not only a percentage.
Source health shows whether the evidence is current, and permissions show who can see it.
Confirm and Disagree stay visible so a person can challenge the recommendation.
Decision Timeline preserves the review state and later outcome.

Interaction

Confirm or disagree. The history updates with the human decision.

The most important interaction is the ability to challenge the recommendation in place. Disagree changes to Noted, and the Decision Timeline immediately records that the user flagged the recommendation as incorrect.

The review control stays beside confidence and evidence.
Disagree records the recommendation as flagged incorrect.
Human judgment is recorded in the decision history. The interface preserves that review without implying automated learning.

Workspace context

Source quality, permissions, and prioritization matter too.

The rest of the product answers practical questions without turning into a feature catalog. How was this ranked? Are the connected sources healthy? Does a component conflict with the system? What evidence can the workspace actually see?

Sorted by impact exposes the factors behind queue order.
Component Debt shows conflicting implementation signals instead of collapsing them into one answer.
Source health and permission clarity stay attached to evidence.
Resolved recommendations keep their outcome in history.
AI & Trust preferences let teams set the confidence threshold directly.

Ask ContextOS

Answers should carry their evidence too.

Ask ContextOS applies the same approach to conversation. Each answer keeps its sources, source health, confidence, and permissions visible so a reviewer can check it before acting.

The answer keeps its sources visible.

User needAsk a workspace question without losing the evidence behind the answer.

DecisionKeep Sources used, freshness, confidence, and who can see the source attached to the response.

EnablesA reviewer can inspect where the answer came from before using it in a product decision.

People need a way to check and challenge AI.

ContextOS changed how I think about AI product design. Generating a recommendation is only the first step. The harder work is designing the evidence, review states, permissions, and history that let someone verify it, challenge it, and still understand what happened later.

Behind the product: why evidence comes first

Problem

The tools were connected. The context was not.

Figma carried design decisions. Jira carried delivery pressure. Slack held exceptions. Research and rationale lived elsewhere. Teams still had to reconstruct the story before acting, and an AI recommendation without evidence would only add another layer to verify.

Context was fragmented by tool.

A handoff risk could begin in a Figma frame, show up again in a Jira ticket, and be discussed differently in Slack. Each source was useful on its own, but the relationship between them was easy to lose.

Priority needed evidence.

A ranked issue is only useful when a designer can see what triggered it, which sources agree, how fresh they are, and whether the system had permission to inspect them.

Design question

How might we help product teams act on workflow risks flagged by AI without turning AI into a black box?

Product strategy

Four moves shaped the product.

I kept the strategy intentionally small so every screen could be traced back to one job: surface a risk, explain it, let a person judge it, and remember the decision.