From understanding work to moving work forward
Vibe devices could capture conversations and turn them into AI summaries, Vibe AI could access summaries like memory and understand work across meetings.
As Vibe prepared to introduce agentic capabilities to Vibe Dot, the team wanted to extend its AI value from summarizing conversations to executing work across connected apps like Slack, Notion, Google Drive, and more overtime.
After the business direction and a general agent flow across Vibe devices were defined, what remained unclear was: what users would actually delegate, how they would discover these capabilities, and how the interaction should work.
My challenge: Turn a broad AI agent capability into a meaningful product experience that users could understand, discover, and confidently use.
01 — What work should users delegate to AI agent?
Finding an opportunity beyond AI summaries
AI could really help many work across 3 themes like Understand, Coordinate, and Execute in personal work or team collaboration.
But what Vibe agent should confidently support was uncertain, I explored AI market trends, Vibe user feedback, and Vibe’s advantaged domain in capturing and understanding conversations.
With findings like AI summaries were becoming increasingly common while fully autonomous AI agents still faced barriers like data privacy, human trust, and delegation reliability in 2026.
I spotted an opportunity between the two ends: helping people move work forward after conversations without expecting AI to execute everything in entire workflows.
From technical AI skills to meaningful AI Jobs
Initially, I explored a wide range of AI SKILLS , abilities that agents use to get user work done, such as drafting an email, searching for information, or updating a document in connected apps, to scope in valuable ones that moved work forward.
But the enormous possibilities quickly exceeded what Vibe could realistically deliver. Discussions with the Dev Lead and Product Marketing narrowed the initial scope to 2 feasible scenarios: Customer follow-ups after sales calls and PRD updates after product meetings.
Another challenge remained was most AI skills described what an agent could do but didn’t had context to let users know WHY this specific skill could help them with what they want. The generic use made them hard to feel relevant to users.
Mental model: Users knew what outcome they want. But they wouldn’t know for sure which AI skill could get them there.
I shifted the product concept to AI JOBS, meaningful outcomes with several skills, meeting contexts, and required apps like following up with customers after today’s calls via Gmail or updating a PRD after today’s meetings on Notion, instead of showing technical AI skills.
Besides 2 executable Jobs, I also proposed repackaging existing LLM text-based work into useful jobs, such as tracing decisions across meetings, identifying unresolved issues, or preparing customer interviews from meeting context without more app integrations.
I developed a concept diagram and concrete scenarios that gave stakeholders a shared understanding of AI Jobs, helping us move from discussing abstract capabilities toward identifying feasible jobs and a clearer scope.
02 — How should users discover and delegate AI Jobs?
With a clearer product concept, the next challenge was making AI Jobs relevant and usable within Vibe's existing product.
I focused on 2 missions: helping users recognize a job worth delegating, then making it straightforward to review and authorize.
Make the next work recognizable through context
My early exploration included repositioning Vibe AI's homepage around agent since summary still needed users to plan actions themselves. But the proposed change was too broad for the initial scope. I returned to where users usually continued their work: Meeting summaries & Ask AI.
Simply listing available jobs in summaries wouldn't really help users imagine why the jobs were relevant. Without meeting context, wanting users to know what to ask AI about was also challenging, not to mention it was a meeting intelligence rather than general LLM AI.
I suggested surfacing AI Jobs close to the exact meeting information that made those actions relevant. In summaries, jobs showed alongside related discussions. In Ask AI, each suggestion showed essential context to help users recognize where it was from.
I also reduced the suggested jobs to 3 and developed criteria for when an AI Job should appear instead of filling the summaries with every possible action. The intention was to help users recognize a useful next step without having to understand a list of AI skills.
Make approval simpler than conversation
Once users chose a job, they still needed to know what exactly the agent would execute.
Many conversational AI rely on dialog exchanges to give instructions, refine changes, and confirm actions. But Vibe's initial jobs were relatively simple, such as analyzing content, sending messages, and updating documents.
Working with the Dev Lead, I explored whether those actions really needed a conversational review process. I designed a single review card that brought the proposed content, destination, and approval into one place whenever a JOB needed review.
My first card design laid out details one by one, but that weakened the visual hierarchy and users may not know what exactly needed to review. I refined the card to emphasize the content that need most attention while keeping recipients and destinations compact.
After users delegated their first Job from their laptops/mobiles, I had AI introduce delegation through Vibe devices like Vibe Dot and Vibe Bot. Without explaining everything upfront, users could discover Dot's “Hey Vibe” delegation once they understood what AI Jobs could do.
What I delivered
The final direction brought together outcome-based AI Jobs, contextual discovery, an easy review-approval experience, and cross-device delegation. It gained stakeholder buy-in and was handed off for development, with permissions and credit costs still under discussion.
My contribution was helping the team move from asking what an AI agent could do to defining what people could meaningfully delegate and how that experience should work.
What I learned
I learned to separate strategic exploration from the immediate design problem. My early research helped identify many opportunities, but it didn’t instantly mean a promising start with the existing product limitation and business constraints.
The project got moved forward smoothly once I translated those abstract possibilities into a concrete concept and interaction model.
You can also find me on 💁
- 👨🏻💻 Resume
- 🎨 Behance
- 📱 +886 909607406
- 📧 a870588@gmail.com