Designing for AI Agents: Getting the Human Interface Right.

The Blank Page

Like most projects, we start with a blank page. 

This is your brain dump space—the place where you write fragmented thoughts, paste random Slack conversations, and scribble half-formed ideas about sprint planning - before the transformation.

Napkin AI demonstrates this perfectly, using its agents to transform text into visual diagrams and charts automatically. 

In our case, your "Neo" infrastructure, "Robin Williams" ceremony, and "Spock" analytics —watch everything you’ve written,  before organising it into structured notes and actions.

Drop in a Jira ticket:

• Your Integration Agent connects it to relevant GitHub PRs whilst 

• Context Agent flags related conversations

Mention Blockers:  

• Your Retrospective Agent starts adding this to your retro items for the next meeting

• Whilst your Analytics Agent spots patterns from previous sprints. 

The Mobile Experience

Preparing for Agile ceremonies happens everywhere: coffee shops, commutes - we’ll need a mobile version:  

• Voice Capture dominates mobile interactions. Research shows conversational AI adoption has increased by 35% year-on-year, with businesses rapidly adopting voice and chat interfaces for faster interactions.

• Progressive Disclosure. Your phone shows one agent's suggestions at a time, but your desktop / tablet reveals the full orchestration.

• Gesture-Based Sorting. Gestural interfaces are becoming popular, letting users control apps with hand movements - perhaps more for vloggers, less so, preparing for standup?! 

Design Principles That Actually Matter

Microsoft's Agent UX guidelines provide the theoretical framework, but for Master of Ceremonies to succeed, these principles need practical implementation:

1. Invisible Until Essential Background agents operate silently until they have something valuable, i.e. No spam! Your Compliance Agent only surfaces when governance issues arise.

2. Confidence Indicators Everywhere Show confidence levels for each result using percentages or visual indicators. When the Planning Agent suggests story estimates, you see "87% confidence based on 12 similar stories." When it's unsure, the confidence level reflects this. 

3. One Conversation, Multiple Personalities You interact with what feels like a single intelligent system, but beneath the surface, your agents debate amongst themselves. 

The Learning Agent might suggest a ceremony approach whilst the Compliance Agent raises concerns—you see the synthesis, not the argument.

Where The Agents Shine

Intelligent Ceremony Prep 

Monday morning, 9 AM: "Planning meeting in 2 hours." Your agents spring into action:

• Integration Agent pulls refined backlog items

• Context Agent highlights dependencies from last sprint

• Analytics Agent suggests capacity based on recent velocity

• Planning Agent structures the agenda

Arrive at the meeting with an agenda prepared by the agents. 

Diary Format Intelligence Not Adrian Mole. When you write "Need to discuss API delays with Sue next Tuesday," your agents: 

• Understand temporal context

• Use memory and connected databases to provide relevant results

• Connect "API delays" to previous tickets

• Link "Sue" to her recent commits, and 

• Automatically suggest Tuesday agenda items

The Enterprise Reality

Security. Your CISO needs to sleep at night - agents must operate within existing security boundaries. 

Integration complexity. Support for Model Context Protocol (MCP) and Agent2Agent (A2A) standards ensures ecosystem interoperability aka - your agents speak the same language as your existing tools. Perfect orchestration assumes agents have clean data, reliable APIs, and no conflicting recommendations. Integration Agents might find broken API connections whilst Context Agents lose track of conversations.

Imperfect as it is, it still beats manual coordination!

Change resistance.  Not everybody will use it. They’ll be concerns about data privacy  - “can anyone else see this?”,” Probably safer to manually write it down!”

Integration is Proven. Tools like AskConcierge.ai already demonstrate seamless integration across Gmail, Slack, Jira, and Notion through natural language. 

Remember the iPhone’s earlier problems?

The original iPhone had call dropping and battery issues but still revolutionised mobile computing. Why? The core interaction model was right. 

“Master of Ceremonies” doesn't need to be perfect on day one. It needs to be materially better than opening Jira, Slack, Confluence in separate tabs whilst trying to remember what happened in last week's retrospective.

The Bottom Line

36 days per function annually, 432 days total team capacity recovered (provide link to post). That's still the prize. But it only works if the interface feels natural, not overwhelming.

“Master of Ceremonies” succeeds when: 

• It feels like having incredibly competent assistants who anticipate your needs without interrupting your flow

• When ceremony preparation shifts from manual coordination to intelligent orchestration.

Next week: Interface design solved. User experience mapped. Technical complexity acknowledged. But building elegant solutions and building viable businesses are completely different challenges.

The final post in this series asks the uncomfortable question: is this a USD 40 billion market opportunity or just an expensive LinkedIn thought experiment?