How to turn a YouTube video into notes

Start from the video's transcript, split it into chapters by topic, and write a short summary and key points for each chapter with its timestamp. You can do that by hand, or have an AI model do it from the transcript in about a minute.

A real example

We ran Keynote: What Agentic AI Asks of Open Source by The Linux Foundation (15:08, Creative Commons Attribution license) through Klaara with GPT-6.1 Sol. It took 38 seconds and cost $0.054. Here are two of its 9 chapters, unedited.

5:01 Beyond the Model: Understanding the Agent Harness

The harness is presented broadly as everything surrounding the model that enables an agent to operate. Using a car analogy, the model is the engine, while the harness includes the equivalent of wheels and brakes: instructions, context, tools, memory, execution loops, permissions, policy enforcement, and observability.

9:54 The Role of Open Source Program Offices

Open source program offices, or OSPOs, help organizations move from ad hoc software consumption to deliberate management of compliance, standards, and contributions. For agentic AI, they can assess dependencies, ecosystem health, governance, and where participation would provide strategic value.

The full output

Everything Klaara made from that talk, including the takeaways for the whole video. The speaker and The Linux Foundation have nothing to do with Klaara.

TL;DR

Agentic AI strategy should look beyond model performance to the surrounding infrastructure—the “harness”—that enables agents to use tools, access systems, and take action. Ana Jíménez Santamaria argues that open source program offices can help organizations evaluate these layers, preserve model choice, manage risk, and participate strategically in emerging ecosystems alongside legal, security, and engineering teams.

Key takeaways

  • Evaluate the entire agent system, not just its model: instructions, tools, memory, execution loops, permissions, policy enforcement, and observability all shape its capabilities and risks.
  • Designing portable infrastructure and using open protocols can help keep models interchangeable, reduce vendor lock-in, and avoid costly future migrations.
  • Openness must be assessed layer by layer because an open model, harness, protocol, or infrastructure component does not automatically make the rest of the system open.
  • Open source program offices can apply their experience with compliance, dependencies, community health, standards, and contribution strategy to agentic AI governance.
  • Effective governance requires collaboration across open source, legal, security, and platform engineering teams, with attention to resilience, cost avoidance, risk foresight, and strategic influence.

0:00 Introduction and Session Roadmap

Unlike systems that only generate answers, AI agents can use tools, access systems, and take actions, creating new organizational choices and responsibilities. The session examines production challenges, the infrastructure surrounding models, open components across the agentic stack, and the implications for open source strategy.

  • Agentic AI systems go beyond generating answers: they can use tools, access systems, and take actions.
  • The talk examines how these capabilities change organizational open source strategy, including infrastructure choices and governance.
  • Organizations face questions about technology selection, policies, risk, accountability, and standards—not just model performance.

2:11 AI Governance and the Strategic Value of Open Source

AI adoption brings questions about regulation, technology choices, policy, accountability, and standards. The speaker connects these challenges to four established dimensions of open source strategy: resilience through avoiding lock-in, cost avoidance through portability, risk foresight before deployment, and strategic influence through early participation in standards.

  • The speaker connects open source strategy to four dimensions: resilience, cost avoidance, risk foresight, and strategic influence.
  • Examples include avoiding vendor lock-in, planning portability before migration becomes expensive, assessing risks before production, and participating in emerging standards.
  • The 'harness' is the infrastructure surrounding a model: instructions, context, tools, memory, execution loops, permissions, policy enforcement, and observability.

5:01 Beyond the Model: Understanding the Agent Harness

The harness is presented broadly as everything surrounding the model that enables an agent to operate. Using a car analogy, the model is the engine, while the harness includes the equivalent of wheels and brakes: instructions, context, tools, memory, execution loops, permissions, policy enforcement, and observability.

7:16 Harness Controls, Portability, and Model Choice

Although model behavior is probabilistic, the surrounding harness can provide more deterministic controls and guardrails. Investing in this infrastructure can support security, efficiency, and portability, shifting the strategic question from which model to choose toward how to keep model choice available.

  • The agent stack can be considered in separate layers: the harness, communication protocols, and shared infrastructure such as authentication, policy, and observability.
  • Model Context Protocol (MCP) is introduced for connecting AI applications to tools and external systems; A2A addresses communication between agents.
  • Openness is layer-specific: using an open protocol does not automatically make the model, harness, or infrastructure open.

8:27 Open Layers of the Agentic Stack

The stack can include an open harness, communication protocols, and shared infrastructure for authentication, policy, and observability. Model Context Protocol supports interaction with tools and external systems, while A2A addresses communication between agents; organizations must evaluate openness independently at each layer.

9:54 The Role of Open Source Program Offices

Open source program offices, or OSPOs, help organizations move from ad hoc software consumption to deliberate management of compliance, standards, and contributions. For agentic AI, they can assess dependencies, ecosystem health, governance, and where participation would provide strategic value.

10:58 Preliminary Survey Findings on OSPOs and AI Governance

Preliminary findings from a 2026 survey conducted with Altimeter Research indicate that OSPOs are participating in AI governance, with 35% reported as leading or co-leading these initiatives. Compliance is the most prominent contribution area discussed, while the speaker identifies opportunities for greater involvement in open protocols, standards, and ecosystem participation.

12:57 Transferring Open Source Expertise Through Collaboration

OSPO expertise can help evaluate whether agentic AI dependencies are sustainably maintained, well governed, and interoperable. The speaker stresses that OSPOs should contribute this knowledge alongside legal, security, and platform engineering teams rather than own AI governance alone.

  • OSPO skills transfer to evaluating agentic AI dependencies, governance, sustainability, interoperability, and strategic participation.
  • This work should be shared with security, legal, and platform engineering rather than assigned exclusively to the OSPO.
  • The closing recap connects agentic AI decisions to resilience, cost, risk, and strategic influence.

13:57 Closing Takeaways

Treat agentic AI as a strategic issue involving resilience, cost, risk, and influence—not simply model selection. Understand the harness, identify which stack layers are open, and assess openness granularly to preserve organizational choice.

Four things that make notes from video useful

  1. Chapters by topic. A long transcript is hard to return to. Topic breaks let you find a part again.
  2. A timestamp on every section. Notes lose detail. A timestamp takes you to the exact place to check it.
  3. Takeaways from the whole video. Our first version took them from the first few chapters, which missed how the talk ended. Check this in any tool you use.
  4. Your own notes beside the generated ones. Add what you thought while watching, which isn't in any transcript.

Doing it in Klaara

  1. Paste the link and choose Read it.
  2. Skim the summary and the chapters, then open Watch for the parts you want to see.
  3. In Watch, write a note at any moment. It keeps the timestamp.
  4. Click Copy to get Markdown, and paste it into Obsidian, Notion or wherever your notes live.

When to take notes by hand

A generated summary tells you what was said. If you need to learn the material for an exam or a job, read the summary first to get the structure, then watch the parts that matter and write in your own words. Also check any number or claim you plan to reuse against the video, because models make mistakes. In our tests one model added a qualification the video didn't contain.

Klaara for Mac

Klaara is a Mac app that turns YouTube videos into summaries with chapters, using your own AI API key. It costs €19 once. 3 free summaries first.

Download for Mac

Related