vidtheque.

read-only demo

Videos X1kp-ABIIxQ

Your agent architecture has a half-life of 6 months — Dan Farrelly, CTO, Inngest

index_state ready data_status ok

AI Engineer· published 2026-07-21· 0:19:20· en-US· indexed 2026-08-11 00:02

Open on YouTube

Scene timeline

  1. Shot 0, 0:00 to 0:03, 1 of 1 keyframes kept
  2. Shot 1, 0:03 to 0:05, 1 of 1 keyframes kept
  3. Shot 2, 0:05 to 0:12, 1 of 1 keyframes kept
  4. Shot 3, 0:12 to 0:22, 1 of 1 keyframes kept
  5. Shot 4, 0:22 to 0:33, 1 of 1 keyframes kept
  6. Shot 5, 0:33 to 1:07, 1 of 1 keyframes kept
  7. Shot 6, 1:07 to 1:38, 1 of 1 keyframes kept
  8. Shot 7, 1:38 to 2:25, 1 of 1 keyframes kept
  9. Shot 8, 2:25 to 3:01, 1 of 1 keyframes kept
  10. Shot 9, 3:01 to 3:36, 0 of 1 keyframes kept
  11. Shot 10, 3:36 to 4:06, 1 of 1 keyframes kept
  12. Shot 11, 4:06 to 4:35, 1 of 1 keyframes kept
  13. Shot 12, 4:35 to 5:11, 1 of 1 keyframes kept
  14. Shot 13, 5:11 to 5:48, 0 of 1 keyframes kept
  15. Shot 14, 5:48 to 6:11, 1 of 1 keyframes kept
  16. Shot 15, 6:11 to 6:46, 1 of 1 keyframes kept
  17. Shot 16, 6:46 to 7:12, 1 of 1 keyframes kept
  18. Shot 17, 7:12 to 7:37, 0 of 1 keyframes kept
  19. Shot 18, 7:37 to 8:02, 0 of 1 keyframes kept
  20. Shot 19, 8:02 to 8:31, 1 of 1 keyframes kept
  21. Shot 20, 8:31 to 9:00, 1 of 1 keyframes kept
  22. Shot 21, 9:00 to 9:37, 1 of 1 keyframes kept
  23. Shot 22, 9:37 to 10:03, 1 of 1 keyframes kept
  24. Shot 23, 10:03 to 10:29, 0 of 1 keyframes kept
  25. Shot 24, 10:29 to 10:37, 1 of 1 keyframes kept
  26. Shot 25, 10:37 to 11:05, 1 of 1 keyframes kept
  27. Shot 26, 11:05 to 11:33, 1 of 1 keyframes kept
  28. Shot 27, 11:33 to 12:11, 1 of 1 keyframes kept
  29. Shot 28, 12:11 to 12:41, 1 of 1 keyframes kept
  30. Shot 29, 12:41 to 13:11, 0 of 1 keyframes kept
  31. Shot 30, 13:11 to 13:49, 1 of 1 keyframes kept
  32. Shot 31, 13:49 to 14:26, 1 of 1 keyframes kept
  33. Shot 32, 14:26 to 14:52, 1 of 1 keyframes kept
  34. Shot 33, 14:52 to 15:18, 1 of 1 keyframes kept
  35. Shot 34, 15:18 to 15:44, 0 of 1 keyframes kept
  36. Shot 35, 15:44 to 16:33, 1 of 1 keyframes kept
  37. Shot 36, 16:33 to 17:05, 1 of 1 keyframes kept
  38. Shot 37, 17:05 to 17:31, 1 of 1 keyframes kept
  39. Shot 38, 17:31 to 17:58, 0 of 1 keyframes kept
  40. Shot 39, 17:58 to 18:24, 0 of 1 keyframes kept
  41. Shot 40, 18:24 to 18:52, 1 of 1 keyframes kept
  42. Shot 41, 18:52 to 19:03, 1 of 1 keyframes kept
  43. Shot 42, 19:03 to 19:19, 0 of 1 keyframes kept

43 shot(s).

keyframes kept every frame deduplicated

What was stored

cues
244
whisperx 244
chunks
34
from 244 cues
keyframes
33
kept of 43 captured
frames with text
33
680 lines read
chapters
0
from the source metadata
keyframe bytes
5.3 MB
word timings on 244 cues

Provenance

Each pipeline stage, its state and the model that produced it
stage state model started took
fetch done — 2026-08-10 23:58 1m 11s
stt done — 2026-08-10 23:59 23s
chunk done — 2026-08-10 23:59 0s
text_embed done — 2026-08-10 23:59 1s
keyframe done — 2026-08-10 23:59 2m 07s
ocr done — 2026-08-11 00:01 15s
frame_embed done — 2026-08-11 00:02 6s

Frames, and what the machine read

  • 0:02 #0 done2 line(s)

    shot 0·sharpness 451.1

    1. AlEngineer0.95
    2. World's Fair0.97

  • 0:04 #1 done2 line(s)

    shot 1·sharpness 663.1

    1. AlEngineer0.96
    2. World's Fair0.99

  • 0:10 #2 done24 line(s)

    shot 2·sharpness 2728.7

    1. LAB & PLATINUM SPONSORS0.99
    2. Amazon AGI Lab0.98
    3. ANTHROP\C1.00
    4. Google DeepMind1.00
    5. MINIMAX0.98
    6. OpenAI0.92
    7. Akamai1.00
    8. arize1.00
    9. aws1.00
    10. Braintrust bright data0.99
    11. B1.00
    12. Browserbase1.00
    13. docker1.00
    14. :neo4j0.91
    15. ORACLE1.00
    16. PayPal1.00
    17. qodo1.00
    18. reducto1.00
    19. Sonar1.00
    20. Makers of0.99
    21. togetherai1.00
    22. Unblocked1.00
    23. WorkOS1.00
    24. SonarQube1.00

  • 0:20 #3 done3 line(s)

    shot 3·sharpness 146.5

    1. INNGEST0.98
    2. AlEngineer0.98
    3. World's Fair1.00

  • 0:28 #4 done15 line(s)

    shot 4·sharpness 1778.1

    1. AlEngineer0.98
    2. World'sFair1.00
    3. AI ENGINEER WORLD'S FAIR0.98
    4. SF1.00
    5. 20261.00
    6. OINNGEST0.95
    7. Your agent's architecture has a1.00
    8. PRESENTED·BY0.96
    9. Microsoft1.00
    10. HALF-LIFE1.00
    11. OF6 MONTHS0.97
    12. Dan Farrelly1.00
    13. CTO & Co-founder, Inngest1.00
    14. World's Fair0.98
    15. Engineering the future of Al0.98

  • 0:50 #5 done18 line(s)

    shot 5·sharpness 3455.8

    1. AlEngineer0.97
    2. World'sFair1.00
    3. CONTEXT1.00
    4. Who aml?0.93
    5. Dan Farrelly0.98
    6. PRESENTEDBY1.00
    7. CTO + Co-founder @ Inngest0.97
    8. Microsoft0.99
    9. ● Built Inngest - Durable execution platform used by thousands of teams to orchestrate agents, Al0.98
    10. workflows, background jobs, and multi-step pipelines.1.00
    11. ● Deep in Al infra daily - Shipping tools that handle the messy reality of running Al in production0.98
    12. (retries, orchestration, state management, observability)1.00
    13. • Practitioner, not just pundit - Actually building and operating the systems, not just talking about0.98
    14. them1.00
    15. INNGEST0.99
    16. INaGST0.70
    17. Engineering the future of Al0.98
    18. World'sFair1.00

  • 1:13 #6 done15 line(s)

    shot 6·sharpness 1364.8

    1. AlEngineer0.98
    2. World'sFair1.00
    3. The stack is moving under you0.98
    4. Function calling → Tool use → MCP → CLI →.0.97
    5. PRESENTED·BY0.95
    6. Microsoft1.00
    7. LangChain → LangGraph → Deep Agents → ..0.95
    8. AGAIN.1.00
    9. ReAct → Plan-and-execute → .0.96
    10. Sub-agents → Dynamic Workflows → Loops →0.98
    11. INNGEST1.00
    12. NNGEST0.88
    13. JULY 1,20260.96
    14. Expo Stage0.99
    15. World'sFair1.00

  • 1:43 #7 done12 line(s)

    shot 7·sharpness 1554.4

    1. AlEngineer0.97
    2. World'sFair1.00
    3. THE QUESTION1.00
    4. PRESENTED·BY0.97
    5. Microsoft1.00
    6. What part of your architecture1.00
    7. survives the next six months?1.00
    8. INNGEST1.00
    9. INGST0.75
    10. JULY 1,20260.99
    11. Expo Stage1.00
    12. World's Fair0.96

  • 2:50 #8 done25 line(s)

    shot 8·sharpness 1779.0

    1. AlEngineer1.00
    2. World's Fair0.99
    3. THE HARNESS0.99
    4. Three layers1.00
    5. YOUR AGENT + APPLICATION0.99
    6. under the agent1.00
    7. Execution1.00
    8. THE BRAIN0.96
    9. PRESENTED BY0.95
    10. Not your product — the0.99
    11. FLOW · STATE· DURABILITY · RETRIES0.96
    12. infrastructure it sits on.0.99
    13. Microsoft1.00
    14. THE1.00
    15. Context1.00
    16. THE KNOWLEDGE1.00
    17. HARNESS1.00
    18. MODELS · PROMPTS · TOOLS · MEMORY · RAG0.93
    19. Compute0.96
    20. THE HANDS1.00
    21. SANDBOXES · RUNTIMES · BROWSERS0.95
    22. OINNGEST0.94
    23. JULY 1,20260.97
    24. Expo Stage0.98
    25. World's Fair1.00

  • 3:19 #9 skipped

    shot 9·duplicate of #8

  • 3:48 #10 done25 line(s)

    shot 10·sharpness 1544.5

    1. AlEngineer0.99
    2. World's Fair0.98
    3. THE THESIS1.00
    4. HALF-LIFE1.00
    5. Relevance1.00
    6. Every layer decays.1.00
    7. PRESENTED BY0.95
    8. Not at the same rate.1.00
    9. Microsoft1.00
    10. Decouple the layers.1.00
    11. The part that survives is the0.98
    12. part you designed to survive.0.99
    13. Context1.00
    14. O0.71
    15. Compute1.00
    16. Execution1.00
    17. Time1.00
    18. weeks1.00
    19. months1.00
    20. years1.00
    21. INNGEST1.00
    22. INGEST0.86
    23. JULY 1, 20260.97
    24. Expo Stage0.98
    25. World's Fair0.99

  • 4:17 #11 done24 line(s)

    shot 11·sharpness 1547.1

    1. AlEngineer0.99
    2. World's Fair0.99
    3. THE THESIS1.00
    4. HALF-LIFE1.00
    5. Relevance1.00
    6. Every layer decays.1.00
    7. PRESENTED BY0.95
    8. Not at the same rate.1.00
    9. Microsoft1.00
    10. Decouple the layers.1.00
    11. The part that survives is the0.99
    12. part you designed to survive.0.99
    13. Context1.00
    14. Compute1.00
    15. Execution1.00
    16. Time1.00
    17. weeks1.00
    18. months1.00
    19. years1.00
    20. OINNGEST0.94
    21. *NGCST0.68
    22. JULY 1, 20260.97
    23. Expo Stage0.98
    24. World's Fair1.00

  • 4:39 #12 done13 line(s)

    shot 12·sharpness 1991.2

    1. AlEngineer0.97
    2. World'sFair1.00
    3. DESIGN FOR EVOLUTION0.99
    4. Don't fight it. Embrace it.0.96
    5. PRESENTED·BY0.95
    6. Magical frameworks→Very high level abstractions, difficult to adapt0.98
    7. Microsoft1.00
    8. Pre-built harnesses →Complete lack of control0.98
    9. Custom rolled→Complexity grows, abstractions leak across responsibilities1.00
    10. INNGEST0.99
    11. World'sFair0.92
    12. JULY 1,20260.98
    13. Expo Stage0.97

  • 5:36 #13 skipped

    shot 13·duplicate of #12

  • 5:57 #14 done13 line(s)

    shot 14·sharpness 2654.5

    1. AlEngineer0.97
    2. World'sFair1.00
    3. The Execution Layer is the system1.00
    4. PRESENTED·BY0.98
    5. responsible for running your code reliably —1.00
    6. Microsoft1.00
    7. managing how, when, and whether each0.99
    8. piece of work completes, independent of the0.99
    9. infrastructure it runs on.1.00
    10. INNGEST1.00
    11. JULY 1,20260.98
    12. Expo Stage0.98
    13. World's Fair0.96

  • 6:39 #15 done30 line(s)

    shot 15·sharpness 1822.3

    1. AlEngineer0.99
    2. World'sFair1.00
    3. THE BRAIN0.98
    4. The harness in motion.1.00
    5. PRESENTED·BY0.95
    6. Execution1.00
    7. Context1.00
    8. SWAPPABLE1.00
    9. Microsoft1.00
    10. CALL MODEL1.00
    11. CLAUDE · GPT · MCP · RAG0.92
    12. trigger1.00
    13. 21.00
    14. RUN CODE1.00
    15. YOUR APP1.00
    16. Compute1.00
    17. SWAPPABLE1.00
    18. 31.00
    19. INVOKE SUB-AGENT1.00
    20. E2B · MODAL · DOCKER0.91
    21. 41.00
    22. CALL MODEL1.00
    23. 51.00
    24. RETURN RESULT1.00
    25. SUB-AGENTS· WORKFLOWS· SCHEDULED TASKS0.98
    26. Execution connects the components. Context and Compute are swappable.1.00
    27. INNGEST1.00
    28. JULY 1,20260.95
    29. Expo Stage0.97
    30. World'sFair1.00

  • 6:59 #16 done29 line(s)

    shot 16·sharpness 1831.6

    1. AlEngineer0.99
    2. World's Fair0.99
    3. ORCHESTRATION - BUILD DURABLE AGENTS1.00
    4. Resume.1.00
    5. Don't restart.1.00
    6. STEPS1.00
    7. PRESENTED·BY0.96
    8. An agent 45 minutes in shouldn't lose 451.00
    9. const result = await step.run("analyze",0.98
    10. minutes to a single failure.0.99
    11. async () => {0.94
    12. Microsoft1.00
    13. return llm.chat({0.98
    14. model: "claude-opus-4",0.99
    15. messages: [{ role, content }],0.99
    16. });0.89
    17. });0.89
    18. Steps 1-370.99
    19. Step 380.94
    20. STEP 391.00
    21. CACHED · DONE0.97
    22. RATE LIMIT0.98
    23. CONTINUE1.00
    24. RETRYABLE· IDEMPOTENT· CACHED0.97
    25. Completed steps are cached — the agent picks up at 38, not step 1.0.98
    26. INNGEST1.00
    27. JULY 1,20260.96
    28. Expo Stage0.98
    29. World's Fair1.00

  • 7:29 #17 skipped

    shot 17·duplicate of #16

  • 7:55 #18 skipped

    shot 18·duplicate of #16

  • 8:17 #19 done28 line(s)

    shot 19·sharpness 1923.9

    1. AlEngineer1.00
    2. World'sFair1.00
    3. EXECUTION - INVOCATION & COORDINATION0.98
    4. Triggered anywhere. Coordinated async.1.00
    5. PRESENTED·BY0.94
    6. FLEXIBLE INVOCATION0.99
    7. SUB-AGENTS: SYNC + ASYNC COORDINATION0.99
    8. Microsoft1.00
    9. CRON1.00
    10. EVENT1.00
    11. AGENT A1.00
    12. DONE1.00
    13. AGENT1.00
    14. step.invoke({...})0.99
    15. step.send({...})0.99
    16. SYNC1.00
    17. ASYNC1.00
    18. API1.00
    19. APPROVAL1.00
    20. AGENT B1.00
    21. TRIGGERED1.00
    22. AGENT C0.99
    23. TRIGGERED1.00
    24. INNGEST1.00
    25. INGEST0.94
    26. JULY 1,20260.96
    27. Expo Stage0.99
    28. World'sFair1.00

  • 8:40 #20 done28 line(s)

    shot 20·sharpness 1926.5

    1. AlEngineer0.99
    2. World'sFair1.00
    3. EXECUTION - INVOCATION & COORDINATION0.98
    4. Triggered anywhere. Coordinated async.0.99
    5. PRESENTED·BY0.95
    6. FLEXIBLE INVOCATION0.99
    7. SUB-AGENTS: SYNC + ASYNC COORDINATION0.99
    8. Microsoft1.00
    9. CRON1.00
    10. EVENT1.00
    11. AGENT A1.00
    12. DONE1.00
    13. AGENT1.00
    14. step.invoke({...))0.96
    15. step.send({...})0.99
    16. SYNC1.00
    17. ASYNC1.00
    18. API1.00
    19. APPROVAL1.00
    20. AGENT B1.00
    21. TRIGGERED1.00
    22. AGENT C0.99
    23. TRIGGERED1.00
    24. INNGEST1.00
    25. a0.94
    26. JULY 1,20260.96
    27. Expo Stage0.99
    28. World'sFair1.00

  • 9:25 #21 done32 line(s)

    shot 21·sharpness 1571.2

    1. AlEngineer0.98
    2. World'sFair1.00
    3. EXECUTION - OBSERVABILITY0.98
    4. See the whole session.0.99
    5. Not one LLM call — the entire run, or session. Models, tools, errors, pauses.0.98
    6. PRESENTED·BY0.95
    7. Microsoft1.00
    8. agent.run1.00
    9. 2h 14m0.97
    10. plan1.00
    11. sandbox.exec1.00
    12. 11m.call0.99
    13. sandbox.exec0.99
    14. 11m.call0.97
    15. browser1.00
    16. finalize1.00
    17. fs.write1.00
    18. navigate1.00
    19. 38m0.87
    20. 1h1.00
    21. 1h 30m1.00
    22. 2h1.00
    23. 1 failure — caught, retried, recovered0.95
    24. model call1.00
    25. tool / sandbox0.95
    26. retry1.00
    27. wait1.00
    28. INNGEST1.00
    29. INGEST0.85
    30. JULY 1,20260.97
    31. Expo Stage0.97
    32. World'sFair1.00

  • 9:42 #22 done27 line(s)

    shot 22·sharpness 1764.2

    1. AlEngineer0.97
    2. World'sFair1.00
    3. COMPUTE - THE HANDS0.99
    4. A sandbox does work. It doesn't think.1.00
    5. PRESENTEDBY1.00
    6. Microsoft1.00
    7. Sandbox1.00
    8. Execution1.00
    9. THE BRAIN1.00
    10. SEQUENCE · STATE · RETRIES0.95
    11. COORDINATION1.00
    12. E2B · MODAL · DAYTONA0.95
    13. MANAGES↓0.97
    14. STATELESS1.00
    15. NO MEMORY1.00
    16. NO RETRIES1.00
    17. EPHEMERAL1.00
    18. NO SEQUENCING1.00
    19. NO RECOVERY0.97
    20. Compute1.00
    21. THE HANDS1.00
    22. SANDBOXES · RUNTIMES · BROWSERS0.96
    23. OINNGEST0.93
    24. ∞0.57
    25. JULY 1,20260.97
    26. Expo Stage0.98
    27. World's Fair0.99

  • 10:24 #23 skipped

    shot 23·duplicate of #22

Transcript

244 cues· 2,803 words· 15,562 chars

  1. 0:12 Going good?
  2. 0:13 All right.
  3. 0:14 There we go.
  4. 0:16 All right.
  5. 0:17 Hello, everyone.
  6. 0:18 All right.
  7. 0:19 Just checking voice.
  8. 0:20 All right.
  9. 0:20 Good.
  10. 0:22 All right.
  11. 0:22 Well, my name is Dan, and I'm here to talk to you about how your agent architecture has a half-life of six months.
  12. 0:33 But first, who am I?
  13. 0:36 I'm Dan.
  14. 0:36 I'm the CTO and co-founder at Ingest.
  15. 0:39 And why should you listen to me?
  16. 0:41 Well, first, I lead an amazing team over at Ingest.
  17. 0:45 We build a system that reliably executes anything from agents to workflows to context pipelines, whatever you want.
  18. 0:56 I'm deep building AI infra every day.
  19. 0:58 And on top of that, I'm also building agents.
  20. 1:01 I'm not just pontificating up here in theory about how you should build them.
  21. 1:07 So speaking of building agents, if you've been building agents for more than six months, you've likely rewritten something, maybe more than once.
  22. 1:18 A new model, a new framework or framework version,
  23. 1:22 new tool calling standard, a new pattern, suddenly your architecture doesn't fit.
  24. 1:29 It's not a complaint.
  25. 1:30 It's just the reality.
  26. 1:32 Things move faster than ever.
  27. 1:34 You can see if you've been to any of the sessions or keynotes.
  28. 1:38 So I want you to think about what you shipped six months ago and how much of it still runs the way that you originally wrote it.
  29. 1:52 Some parts of your code likely survived, but did they survive by accident or did you design it that way?
  30. 1:59 Did they survive by design?
  31. 2:03 How did you actually architect your agent?
  32. 2:06 Did you architect it at all?
  33. 2:08 It's okay.
  34. 2:10 Did you use a framework?
  35. 2:11 Did you custom roll something?
  36. 2:14 Was the architecture an intentional design that you really thought through or just kind of evolved into what you have now?
  37. 2:21 These are all various ways of where people are these days.
  38. 2:25 So a lot of folks have talked about harness architecture, building harnesses.
  39. 2:33 But mostly, I think a lot of people draw diagrams, and they talk about specific components.
  40. 2:39 Draw just nice lines.
  41. 2:40 It looks very simple.
  42. 2:44 But I want to talk about the conceptual layers when you're building a system like this.
  43. 2:49 This is maybe the mental model, not specific components.
  44. 2:54 So in my opinion, there are three discrete layers.
  45. 2:58 First, the execution layer.
  46. 3:01 I think of this as the brain.
  47. 3:03 It's where flow, state, durability, retries happen.
  48. 3:09 Then there's the context layer.
  49. 3:11 This is the knowledge.
  50. 3:13 This is models, prompts, tools, memory.

Open at this second