vidtheque.

read-only demo

Videos UPwGaM2MKHY

The Log Is The Agent - Ishaan Sehgal, Omnara

index_state ready data_status ok

AI Engineer· published 2026-06-25· 0:15:11· en-US· indexed 2026-08-11 10:10

Open on YouTube

Scene timeline

  1. Shot 0, 0:00 to 0:25, 1 of 1 keyframes kept
  2. Shot 1, 0:25 to 0:51, 1 of 1 keyframes kept
  3. Shot 2, 0:51 to 1:19, 1 of 1 keyframes kept
  4. Shot 3, 1:19 to 1:32, 1 of 1 keyframes kept
  5. Shot 4, 1:32 to 1:36, 1 of 1 keyframes kept
  6. Shot 5, 1:36 to 2:22, 1 of 1 keyframes kept
  7. Shot 6, 2:22 to 2:49, 1 of 1 keyframes kept
  8. Shot 7, 2:49 to 3:24, 1 of 1 keyframes kept
  9. Shot 8, 3:24 to 3:55, 1 of 1 keyframes kept
  10. Shot 9, 3:55 to 4:26, 0 of 1 keyframes kept
  11. Shot 10, 4:26 to 4:31, 1 of 1 keyframes kept
  12. Shot 11, 4:31 to 5:03, 1 of 1 keyframes kept
  13. Shot 12, 5:03 to 5:36, 0 of 1 keyframes kept
  14. Shot 13, 5:36 to 6:12, 1 of 1 keyframes kept
  15. Shot 14, 6:12 to 6:49, 0 of 1 keyframes kept
  16. Shot 15, 6:49 to 6:55, 1 of 1 keyframes kept
  17. Shot 16, 6:55 to 7:21, 0 of 1 keyframes kept
  18. Shot 17, 7:21 to 7:47, 0 of 1 keyframes kept
  19. Shot 18, 7:47 to 8:13, 0 of 1 keyframes kept
  20. Shot 19, 8:13 to 8:39, 0 of 1 keyframes kept
  21. Shot 20, 8:39 to 9:07, 1 of 1 keyframes kept
  22. Shot 21, 9:07 to 9:35, 1 of 1 keyframes kept
  23. Shot 22, 9:35 to 10:02, 0 of 1 keyframes kept
  24. Shot 23, 10:02 to 10:32, 0 of 1 keyframes kept
  25. Shot 24, 10:32 to 11:01, 0 of 1 keyframes kept
  26. Shot 25, 11:01 to 11:42, 1 of 1 keyframes kept
  27. Shot 26, 11:42 to 12:31, 1 of 1 keyframes kept
  28. Shot 27, 12:31 to 13:05, 1 of 1 keyframes kept
  29. Shot 28, 13:05 to 13:38, 0 of 1 keyframes kept
  30. Shot 29, 13:38 to 14:11, 0 of 1 keyframes kept
  31. Shot 30, 14:11 to 14:40, 1 of 1 keyframes kept
  32. Shot 31, 14:40 to 15:10, 0 of 1 keyframes kept

32 shot(s).

keyframes kept every frame deduplicated

What was stored

cues
195
whisperx 195
chunks
27
from 195 cues
keyframes
19
kept of 32 captured
frames with text
19
111 lines read
chapters
0
from the source metadata
keyframe bytes
2.9 MB
word timings on 195 cues

Provenance

Each pipeline stage, its state and the model that produced it
stage state model started took
fetch done — 2026-08-11 10:07 1m 08s
stt done — 2026-08-11 10:08 31s
chunk done — 2026-08-11 10:09 0s
text_embed done — 2026-08-11 10:09 7s
keyframe done — 2026-08-11 10:09 1m 26s
ocr done — 2026-08-11 10:10 4s
frame_embed done — 2026-08-11 10:10 4s

Frames, and what the machine read

  • 0:19 #0 done2 line(s)

    shot 0·sharpness 876.7

    1. The Log Is the Agent0.99
    2. Ishaan Sehgal · CEO, Omnara (YC S25)0.99

  • 0:48 #1 done5 line(s)

    shot 1·sharpness 1547.5

    1. What is your character?1.00
    2. game engine1.00
    3. playstation1.00
    4. controller1.00
    5. save file0.98

  • 0:54 #2 done2 line(s)

    shot 2·sharpness 892.0

    1. Your character is the save file.0.99
    2. 40.57

  • 1:28 #3 done6 line(s)

    shot 3·sharpness 886.1

    1. what people think the agent is1.00
    2. model1.00
    3. runtime / sandbox0.99
    4. tools1.00
    5. loop1.00
    6. framework1.00

  • 1:33 #4 done2 line(s)

    shot 4·sharpness 992.3

    1. The agent is its data1.00
    2. Specifically: the log1.00

  • 1:46 #5 done14 line(s)

    shot 5·sharpness 1652.7

    1. What is the log?0.98
    2. Event History1.00
    3. example1.00
    4. User Input1.00
    5. user asks1.00
    6. → model responds1.00
    7. Model Input/Output1.00
    8. → tool called0.98
    9. Permissions1.00
    10. → permission request0.95
    11. → human approves0.98
    12. Tool call and tool result1.00
    13. → tool result1.00
    14. → turn completed1.00

  • 2:38 #6 done9 line(s)

    shot 6·sharpness 1322.8

    1. Every operation reads from or appends to the log0.99
    2. Log0.99
    3. Determine1.00
    4. Next1.00
    5. step0.88
    6. whatever you want0.99
    7. Handle Next Step0.99
    8. Append to Log1.00
    9. adapted from dex's 12-factor agents0.99

  • 2:56 #7 done8 line(s)

    shot 7·sharpness 3103.7

    1. Simplified Loop1.00
    2. while session_has_work:1.00
    3. state = reconstruct_from_log(session_id)1.00
    4. response = model.next(state)0.99
    5. append(response)1.00
    6. if response.requests_tool:1.00
    7. result = run_tool(response.tool_call)1.00
    8. append(result)1.00

  • 3:31 #8 done10 line(s)

    shot 8·sharpness 1368.8

    1. Databases learned this first0.99
    2. 目0.65
    3. Log1.00
    4. Tables1.00
    5. Indexes0.98
    6. Caches1.00
    7. 止0.96
    8. Materialized Views0.99
    9. Search1.00
    10. Analytics1.00

  • 4:08 #9 skipped

    shot 9·duplicate of #8

  • 4:29 #10 done1 line(s)

    shot 10·sharpness 469.4

    1. Objections1.00

  • 4:44 #11 done1 line(s)

    shot 11·sharpness 826.2

    1. Compaction is not the log1.00

  • 5:10 #12 skipped

    shot 12·duplicate of #11

  • 5:54 #13 done13 line(s)

    shot 13·sharpness 1295.4

    1. The log isn't the whole world0.99
    2. its the agents view of it1.00
    3. World1.00
    4. Agent Log1.00
    5. filesystem1.00
    6. event history1.00
    7. GitHub1.00
    8. a. what it did0.99
    9. email1.00
    10. b. what changed1.00
    11. browser1.00
    12. APls0.91
    13. database1.00

  • 6:27 #14 skipped

    shot 14·duplicate of #13

  • 6:54 #15 done2 line(s)

    shot 15·sharpness 939.3

    1. Once the log is the primitive,1.00
    2. the properties fall out.1.00

  • 7:00 #16 skipped

    shot 16·duplicate of #8

  • 7:39 #17 skipped

    shot 17·duplicate of #8

  • 8:10 #18 skipped

    shot 18·duplicate of #8

  • 8:16 #19 skipped

    shot 19·duplicate of #8

  • 8:47 #20 done8 line(s)

    shot 20·sharpness 460.8

    1. Multiplayer1.00
    2. SHARED SESSION LOG0.98
    3. Alex1.00
    4. Jordan1.00
    5. 00:240.86
    6. Designer1.00
    7. Maya1.00
    8. Product Manager1.00

  • 9:21 #21 done5 line(s)

    shot 21·sharpness 806.7

    1. Migration / Model Portability1.00
    2. Claude1.00
    3. GPT1.00
    4. Qwen1.00
    5. Different models, same agent.0.99

  • 9:51 #22 skipped

    shot 22·duplicate of #21

  • 10:12 #23 skipped

    shot 23·duplicate of #8

Transcript

195 cues· 2,547 words· 14,053 chars

  1. 0:00 Hey, everyone.
  2. 0:01 I'm Ishan, the CEO of Omnara, and today I'm going to be talking about the log is the agent.
  3. 0:06 The basic idea of the talk is simple, and that is most people think of an agent as the model or the execution environment that it's running in, and I think that that's the wrong abstraction.
  4. 0:17 I think that the thing that actually gives an agent its identity is its log, and that's what I'm going to be arguing today.
  5. 0:26 So think about a character you've spent 100 hours playing in your favorite video game, in this case Skyrim.
  6. 0:32 What exactly is your character?
  7. 0:34 Is it the game engine?
  8. 0:36 Is it the PlayStation?
  9. 0:37 Is it the controller?
  10. 0:40 No, it's not.
  11. 0:41 Those things matter, and those things are what we'll interact with, and they'll run the character, but none of those things are your character.
  12. 0:48 Your character is data.
  13. 0:49 It's the save file.
  14. 0:52 And this is important because if your PlayStation bursts into flames, your character isn't gone.
  15. 0:58 You can buy another PlayStation, you can download your save file from the cloud, and you can resume exactly where they were.
  16. 1:06 And that's because the agent and its identity and history and its state is all captured in its data.
  17. 1:13 The character lives in the data.
  18. 1:14 And this is the framing that I want to bring to agents.
  19. 1:19 Today, when people talk about agents, they usually point at the wrong thing.
  20. 1:24 They'll say that the agent is the model or they'll say that it's the runtime.
  21. 1:28 And again, as I mentioned earlier, those things matter, but they're not the agent.
  22. 1:32 The agent is its data.
  23. 1:34 It's specifically the log.
  24. 1:37 So what actually is the log?
  25. 1:39 At the simplest level, the log is the append-only event history of the agent.
  26. 1:45 It's every user input, every model output, every tool call, tool result, permission, failure.
  27. 1:52 And the idea is that every state transition that the agent takes is written to the log.
  28. 1:58 This is important because it means that the identity of the agent isn't tied to the runtime or the model or the tools.
  29. 2:06 Those things are all just interpreting and appending to the log.
  30. 2:09 They're reading the log, acting on it, and writing the next event back.
  31. 2:14 And that's important because then just using the log on its own is enough to resume the agent.
  32. 2:22 Once you define the agent as the log, the rest of the system becomes a whole lot easier to reason about because every operation is either reading from or appending to the log.
  33. 2:33 The model is reading from the log and then determining the next action.
  34. 2:38 The tool runner is then executing that action and then it's appending that result.
  35. 2:43 And this is all operating in a loop.
  36. 2:46 Everything coordinates itself around the log.
  37. 2:49 In practice, a simplified loop can look something like this.
  38. 2:52 You can reconstruct the state from the log.
  39. 2:54 You can pass that state to the model.
  40. 2:56 The model can propose the next step and then append that response to the log.
  41. 3:00 If the response asks for a tool, you can run that tool and also append that response to the log, and then you can repeat.
  42. 3:06 The important insight is not that this loop is complicated.
  43. 3:10 The important insight is that the loop is disposable.
  44. 3:13 A worker can claim the session, read the log, advance the agent one step, write the result, and then just completely disappear.
  45. 3:20 And then that means that any other worker can pick it up later.
  46. 3:25 This pattern should feel familiar.
  47. 3:27 Databases had to learn this first.
  48. 3:29 For years, databases looked like these non-transparent systems that were hard to reason about with tables and indexes and materialized views.
  49. 3:36 But underneath every serious database is a log.
  50. 3:41 And

Open at this second