The starting concept
Monitoring, analysis, and proposed updates were closely tied together. I wanted to make their different roles explicit.
Enterprise AI / Independent exploration
AI Strategic Planning: Independent Redesign
An independent redesign of an early strategic-planning concept. I explored how to separate agent roles and let people review suggestions before changing a plan.
- Type
- Independent concept redesign
- Intended audience
- Leadership teams
- Focus
- Agent roles and human review
- Status
- Exploration / not shipped
My contribution
I took an early concept and redesigned the experience as a separate exploration. My focus was making agent activity, suggested changes, and approval steps easier to follow.
My original project work is confidential. This is a separate, independent exploration, not the client product or my original deliverables. The screens and prototype show this redesign; they are not evidence of a launch or user testing.
Starting Point
What should an AI agent be allowed to change?
I started with an early concept for AI-assisted strategic planning. In reviewing it, I focused on a question: could someone tell the difference between an agent noticing a change and changing the plan?
My design question
How could AI help a leadership team notice changes without making decisions on its behalf? This was an exploration, not a finding from user research.
AI assists thinking. It never auto-changes your map.
Interface principle
I treated unclear permissions as a design risk and used that to guide the redesign. I have not tested whether users experience the same problem.
Architecture
Separate monitoring from making changes.
One agent watches for changes. The other analyzes them when asked. Neither can change the strategic plan without review, so an irrelevant signal or a poor suggestion cannot silently become a decision.
Signal Agent / always on
Monitors market events and scenario assumptions on a schedule. Its outputs appear in Monitoring and Dashboard; it does not publish changes to the plan.
Analysis Agent / on demand
Runs when the user asks. Proposed scores and narrative changes go through review. Approval and publishing remain separate actions.
Draft before commitment
Route generated output into a reviewable draft. Confidence and relevance help users focus on changes that merit attention.
Keep scores provisional
Make pending changes visibly different from committed analysis. Filter minor shifts so only meaningful deltas demand review.
Offer context before a blank prompt
Use surface-specific question cards to help users get started while retaining custom input.
Show what each agent is allowed to do.
The rules define when an agent can act, what confidence is required, and when it needs a person to decide.
Product Design
See the evidence before approving a change.
Leaders can check a signal, review a proposed change, ask a question, and decide what to keep. Those steps remain separate so it is clear when the plan actually changes.
See what changed around the strategy.
PESTLE-tagged signals and risk status make the Signal Agent's work visible without silently rewriting the plan.
Review a proposed rewrite before committing.
An agent-proposed endstate remains provisional until a person reviews and approves the change.
Bring relevant signals into one view.
Scores connect market signals with the scenario, helping leadership focus on assumptions that may be shifting.
Ask with the current context in view.
Pre-framed prompt cards guide inquiry. The analyst can explain, but cannot publish without a human decision.
Review a decision together.
An additional concept brings two-lead approval, agent discussion, and a decision record into one place. It is part of the exploration, not a shipped feature.
Reflection
What this exploration shows, and what it does not.
The redesign shows how I would approach agent permissions and review flows. It is not commissioned delivery, validated research, or a claim of business impact.
The main design decision
I separated monitoring from analysis and kept proposed changes distinct from approved ones. That is my proposed solution, not a proven improvement.
What I would test
Can people explain each agent's role, distinguish a suggestion from an approved change, and find the evidence they need to decide?
Next, I would test whether leaders can tell a signal from a proposed change and review it without publishing by mistake. I would also ask why they rejected suggestions before deciding how that feedback should affect the agent.