cl kj clkjxclk xjclcxj
Subtext
Product C3
C3 helps teams manage customer communication and automation through configurable workflows. As workflows become more complex, AI can dramatically reduce the effort required to build and modify them.
Problem The Trust Gap
AI made workflow creation faster, but introduced a gap between what users asked for and what AI interpreted. A workflow could look correct while containing the wrong logic, assumptions, or actions, leaving users unsure whether they could safely trust the result.
Solution The AI Collaborator
I designed AI as a collaborator, not an executor. It proposes workflow changes, explains key decisions, surfaces uncertainty, and lets users review, correct, or reject them before applying.
Result Confidence by Design
Users can understand what AI interpreted, what it changed, and why, while staying in control of the final workflow. The experience makes AI faster to use without making it harder to trust.
CONTEXT
C3 is building a smarter way to automate customer conversations.
As C3 introduced AI into the Workflow Builder, users could describe what they wanted instead of manually constructing every step. I was brought in to shape this new AI experience and solve the challenge of letting AI move quickly while keeping users confident and in control.
PROBLEM
AI could build the workflow in seconds. But could users trust what it built?
AI could turn a simple instruction into a complex workflow with multiple triggers, conditions, and actions. But users couldn’t easily see how their request had been interpreted, what assumptions AI had made, or whether the resulting logic matched their intent.
The real challenge wasn’t getting AI to build workflows faster. It was making its decisions understandable enough for users to trust, verify, and stay in control of the outcome.
GOAL
To create an AI experience that could move quickly while keeping its reasoning visible, its changes inspectable, and the user in control of what ultimately gets applied.
Make intent visible: Show users how AI interpreted their request.
Make changes inspectable: Clearly communicate what AI added, removed, or modified.
Keep users in control: AI proposes changes; users decide what gets applied.
Surface uncertainty: Ask for clarification instead of confidently making risky assumptions.
RESEARCH & KEY INSIGHTS
We mapped how users interpret and verify AI-generated workflows, using workflow analysis, user interviews, and AI interaction reviews to uncover where trust breaks down.
Key Insight N01: Users trust outcomes they can verify. Seeing how a request translates into a workflow builds more confidence than an AI explanation alone.
Key Insight No2: Change visibility matters more than AI confidence. Users don’t need AI to sound certain. They need to clearly understand what changed and why.
Key Insight No3: Ambiguity should interrupt automation. When AI isn’t confident about an important decision, asking for clarification is better than silently making an assumption.
DESIGN 1/4
AI Trace
Making AI’s reasoning visible while it builds.
While AI works on the request, a lightweight step trace shows how it is interpreting the request, mapping requirements, and constructing the workflow. This gives users visibility into the process without exposing raw reasoning.
DESIGN 2/4
Reviewable Generation
Previewing and editing the workflow before applying it.
AI generates the proposed blocks inside a dedicated canvas within the response. Users can inspect the structure, edit or remove blocks, and approve the result before applying it to their main workflow.
DESIGN 3/4
Clarification
Resolving ambiguity before making the wrong decision.
When a request is unclear, AI pauses and asks a focused question with relevant options. Users can select an answer or provide their own, giving AI the context it needs to continue confidently.
DESIGN 4/4
Decision Context
Explaining why each AI-generated block was added.
Once generated blocks are added to the workflow, a contextual indicator lets users understand their origin. Selecting it reveals the specific request or decision that led to the block, making AI-generated logic easier to audit, trust, and modify later.
IMPACT
Reducing uncertainty around AI-generated workflow changes.
30%
fewer manual steps AI-assisted workflow creation reduces the effort needed to build and modify automation.
100%
visible changes Every AI-generated change is surfaced for review before it becomes part of the workflow.
3x
clearer decision points Users can understand the request, proposed logic, and final action before committing.
REFLECTIONS
Trust comes from verification.
Making AI visible isn’t enough. Users need to understand and verify its decisions.
Uncertainty is part of the experiencea.
When AI isn’t confident, asking for clarification is safer than making a hidden assumption.
Control should remain with the user.
The best AI experiences don’t remove decisions from users. They make those decisions faster and easier.