The Startup Podcast

Why Claude never stops listening to our meetings (w/ Minikai's Keoki Alexander-Chang)

9 snips
Mar 16, 2026
Keoki Alexander-Chang, founder and CEO of Minikai and former Deloitte AI lead, builds AI agents to cut admin in disability and aged care. He talks about person-centred AI, fine-grained access control and provenance for safety. He explains using continuously running agents to update sales pipelines, wearing lapel mics to turn meetings into code, and rethinking build vs buy in an AI-native stack.
Ask episode
AI Snips
Chapters
Transcript
Episode notes
INSIGHT

Design For Access Control And Provenance

  • Fine-grained access control and provenance are foundational to using AI with sensitive health data to meet privacy obligations.
  • Keoki enforces grounded citations and forces clinicians to provide source context so notes can't be fabricated by the agent.
ADVICE

Add Friction To Keep Humans Responsible

  • Keep humans deliberately in the loop by adding friction where responsibility matters, e.g., require clinician-provided context to create clinical notes.
  • This prevents an agent from synthesizing unstamped or fabricated progress notes in high-stakes settings.
INSIGHT

Let Users Peek Into Agents' Work

  • Agents should expose their intermediate work so humans can 'peek' into what the agent is doing and invite review at the right moment.
  • Keoki frames this as a new product surface: visible traces of background agent activity for review.
Get the Snipd Podcast app to discover more snips from this episode
Get the app