---
schema: 1
type: video
title: 'Don''t Let the LLM Drive - Ornella Bahidika & Joel Allou, Microsoft'
description: >-
  The LLM in my voice tutor doesn't decide when the lesson is over. It doesn't
  decide whether the user got the answer right. It doesn't decide which step
  comes next. A harness does all of that. The LLM just shows up and talks. Every
  engineer who's tried to ship a multi-step flow agent has felt this: the model
  declares itself done before it should, skips a check, loops on a step, or
  quietly drops half the procedure. Prompting gets you most of the way. Tool-use
  discipline gets you closer. The last stretch, the difference between a demo
  and a system real users sign into every day, is owning the flow outside the
  model. Ace is a voice tutor in production. The lesson is a small state
  machine: intro, teach, check, grade, advance, wrap. Each node hands the LLM a
  narrow contract: do this, return that. The harness validates the return,
  advances the state, decides what comes next. When the LLM tries to skip ahead
  the harness ignores it. When the LLM tries to declare the lesson finished the
  harness checks the actual completion signal. Same pattern for the shared
  canvas the agent draws on, for grading, for interruption handling. Seven
  minutes. The state machine, the contract shape, a few places where I tried to
  give the LLM more authority and rolled it back, and a short list of decisions
  the LLM should never own in any flow agent. Speakers: - Ornella Bahidika
  (Microsoft): Ornella Bahidika is a Product Manager at Microsoft, where she
  develops solutions that help organizations optimize collaboration, workplace
  technology, and AI-driven experiences. LinkedIn:
  https://www.linkedin.com/in/ornella-bahidika/ - Joel Allou: Joel builds
  voice-first AI tutors. Solo founder focused on agentic systems for
  personalized learning, with a particular interest in infrastructure that makes
  flow agents reliable. LinkedIn: https://www.linkedin.com/in/joel-allou/
canonical: >-
  https://liquidgold.cc/channel/aidotengineer/explore/video/don-t-let-the-llm-drive-ornella-bahidika-joel-allou-f03da3cd95a3
markdown: >-
  https://liquidgold.cc/channel/aidotengineer/video/don-t-let-the-llm-drive-ornella-bahidika-joel-allou-f03da3cd95a3.md
llms: null
parent: 'https://liquidgold.cc/channel/aidotengineer.md'
channel: aidotengineer
updated: '2026-10-24T07:08:07.026Z'
language: null
children: []
videoId: m24UKZomm7k
durationSeconds: 368
publishedAt: '2026-07-20T06:25:18.000Z'
searchChunks:
  - ordinal: 0
    text: null
    startSeconds: 13
    endSeconds: 103
    watch: 'https://liquidgold.cc/watch/m24UKZomm7k?t=13'
    heldReason: rights
  - ordinal: 1
    text: null
    startSeconds: 13
    endSeconds: 103
    watch: 'https://liquidgold.cc/watch/m24UKZomm7k?t=13'
    heldReason: rights
  - ordinal: 2
    text: null
    startSeconds: 13
    endSeconds: 103
    watch: 'https://liquidgold.cc/watch/m24UKZomm7k?t=13'
    heldReason: rights
  - ordinal: 3
    text: null
    startSeconds: 13
    endSeconds: 103
    watch: 'https://liquidgold.cc/watch/m24UKZomm7k?t=13'
    heldReason: rights
  - ordinal: 4
    text: null
    startSeconds: 13
    endSeconds: 103
    watch: 'https://liquidgold.cc/watch/m24UKZomm7k?t=13'
    heldReason: rights
  - ordinal: 5
    text: null
    startSeconds: 114
    endSeconds: 242
    watch: 'https://liquidgold.cc/watch/m24UKZomm7k?t=114'
    heldReason: rights
  - ordinal: 6
    text: null
    startSeconds: 114
    endSeconds: 242
    watch: 'https://liquidgold.cc/watch/m24UKZomm7k?t=114'
    heldReason: rights
  - ordinal: 7
    text: null
    startSeconds: 114
    endSeconds: 242
    watch: 'https://liquidgold.cc/watch/m24UKZomm7k?t=114'
    heldReason: rights
  - ordinal: 8
    text: null
    startSeconds: 253
    endSeconds: 368
    watch: 'https://liquidgold.cc/watch/m24UKZomm7k?t=253'
    heldReason: rights
  - ordinal: 9
    text: null
    startSeconds: 253
    endSeconds: 368
    watch: 'https://liquidgold.cc/watch/m24UKZomm7k?t=253'
    heldReason: rights
  - ordinal: 10
    text: null
    startSeconds: 253
    endSeconds: 368
    watch: 'https://liquidgold.cc/watch/m24UKZomm7k?t=253'
    heldReason: rights
  - ordinal: 11
    text: null
    startSeconds: 253
    endSeconds: 368
    watch: 'https://liquidgold.cc/watch/m24UKZomm7k?t=253'
    heldReason: rights
---


# Don't Let the LLM Drive - Ornella Bahidika & Joel Allou, Microsoft

The LLM in my voice tutor doesn't decide when the lesson is over. It doesn't decide whether the user got the answer right. It doesn't decide which step comes next. A harness does all of that. The LLM just shows up and talks. Every engineer who's tried to ship a multi-step flow agent has felt this: the model declares itself done before it should, skips a check, loops on a step, or quietly drops half the procedure. Prompting gets you most of the way. Tool-use discipline gets you closer. The last stretch, the difference between a demo and a system real users sign into every day, is owning the flow outside the model. Ace is a voice tutor in production. The lesson is a small state machine: intro, teach, check, grade, advance, wrap. Each node hands the LLM a narrow contract: do this, return that. The harness validates the return, advances the state, decides what comes next. When the LLM tries to skip ahead the harness ignores it. When the LLM tries to declare the lesson finished the harness checks the actual completion signal. Same pattern for the shared canvas the agent draws on, for grading, for interruption handling. Seven minutes. The state machine, the contract shape, a few places where I tried to give the LLM more authority and rolled it back, and a short list of decisions the LLM should never own in any flow agent. Speakers: - Ornella Bahidika (Microsoft): Ornella Bahidika is a Product Manager at Microsoft, where she develops solutions that help organizations optimize collaboration, workplace technology, and AI-driven experiences. LinkedIn: https://www.linkedin.com/in/ornella-bahidika/ - Joel Allou: Joel builds voice-first AI tutors. Solo founder focused on agentic systems for personalized learning, with a particular interest in infrastructure that makes flow agents reliable. LinkedIn: https://www.linkedin.com/in/joel-allou/

Canonical: https://liquidgold.cc/channel/aidotengineer/explore/video/don-t-let-the-llm-drive-ornella-bahidika-joel-allou-f03da3cd95a3 · Channel: [AI Engineer](https://liquidgold.cc/channel/aidotengineer.md)

> Search chunk text is held until public-source provenance is verified.

## Document chunks

> Missing: no stored document chunks are available.