vidtheque.

read-only demo

Videos VktrqzQgytY

CI/CD Is Dead, Agents Need Continuous Compute and Computers — Hugo Santos and Madison Faulkner

index_state ready data_status ok

AI Engineer· published 2026-05-13· 0:18:37· en-US· indexed 2026-08-10 19:51

Open on YouTube

Scene timeline

  1. Shot 0, 0:00 to 0:05, 1 of 1 keyframes kept
  2. Shot 1, 0:05 to 0:09, 1 of 1 keyframes kept
  3. Shot 2, 0:09 to 0:14, 1 of 1 keyframes kept
  4. Shot 3, 0:14 to 0:47, 1 of 1 keyframes kept
  5. Shot 4, 0:47 to 1:27, 1 of 1 keyframes kept
  6. Shot 5, 1:27 to 1:32, 1 of 1 keyframes kept
  7. Shot 6, 1:32 to 1:57, 1 of 1 keyframes kept
  8. Shot 7, 1:57 to 2:22, 1 of 1 keyframes kept
  9. Shot 8, 2:22 to 2:26, 1 of 1 keyframes kept
  10. Shot 9, 2:26 to 2:33, 1 of 1 keyframes kept
  11. Shot 10, 2:33 to 2:50, 1 of 1 keyframes kept
  12. Shot 11, 2:50 to 3:14, 1 of 1 keyframes kept
  13. Shot 12, 3:14 to 3:42, 1 of 1 keyframes kept
  14. Shot 13, 3:42 to 4:00, 1 of 1 keyframes kept
  15. Shot 14, 4:00 to 4:29, 1 of 1 keyframes kept
  16. Shot 15, 4:29 to 4:39, 1 of 1 keyframes kept
  17. Shot 16, 4:39 to 5:11, 1 of 1 keyframes kept
  18. Shot 17, 5:11 to 5:50, 0 of 1 keyframes kept
  19. Shot 18, 5:50 to 6:06, 1 of 1 keyframes kept
  20. Shot 19, 6:06 to 6:11, 1 of 1 keyframes kept
  21. Shot 20, 6:11 to 6:30, 1 of 1 keyframes kept
  22. Shot 21, 6:30 to 7:20, 1 of 1 keyframes kept
  23. Shot 22, 7:20 to 7:47, 1 of 1 keyframes kept
  24. Shot 23, 7:47 to 8:14, 1 of 1 keyframes kept
  25. Shot 24, 8:14 to 8:41, 1 of 1 keyframes kept
  26. Shot 25, 8:41 to 9:03, 1 of 1 keyframes kept
  27. Shot 26, 9:03 to 9:29, 1 of 1 keyframes kept
  28. Shot 27, 9:29 to 9:59, 1 of 1 keyframes kept
  29. Shot 28, 9:59 to 10:03, 1 of 1 keyframes kept
  30. Shot 29, 10:03 to 10:32, 1 of 1 keyframes kept
  31. Shot 30, 10:32 to 11:01, 0 of 1 keyframes kept
  32. Shot 31, 11:01 to 11:29, 0 of 1 keyframes kept
  33. Shot 32, 11:29 to 11:58, 0 of 1 keyframes kept
  34. Shot 33, 11:58 to 12:27, 1 of 1 keyframes kept
  35. Shot 34, 12:27 to 12:53, 1 of 1 keyframes kept
  36. Shot 35, 12:53 to 13:20, 0 of 1 keyframes kept
  37. Shot 36, 13:20 to 13:46, 0 of 1 keyframes kept
  38. Shot 37, 13:46 to 14:13, 0 of 1 keyframes kept
  39. Shot 38, 14:13 to 14:40, 0 of 1 keyframes kept
  40. Shot 39, 14:40 to 15:06, 0 of 1 keyframes kept
  41. Shot 40, 15:06 to 15:33, 0 of 1 keyframes kept
  42. Shot 41, 15:33 to 15:50, 1 of 1 keyframes kept
  43. Shot 42, 15:50 to 16:23, 1 of 1 keyframes kept
  44. Shot 43, 16:23 to 16:48, 1 of 1 keyframes kept
  45. Shot 44, 16:48 to 17:10, 1 of 1 keyframes kept
  46. Shot 45, 17:10 to 17:20, 1 of 1 keyframes kept
  47. Shot 46, 17:20 to 17:46, 0 of 1 keyframes kept
  48. Shot 47, 17:46 to 18:12, 0 of 1 keyframes kept
  49. Shot 48, 18:12 to 18:22, 1 of 1 keyframes kept
  50. Shot 49, 18:22 to 18:36, 1 of 1 keyframes kept

50 shot(s).

keyframes kept every frame deduplicated

What was stored

cues
189
whisperx 189
chunks
32
from 189 cues
keyframes
38
kept of 50 captured
frames with text
38
872 lines read
chapters
14
from the source metadata
keyframe bytes
6.3 MB
word timings on 189 cues

Provenance

Each pipeline stage, its state and the model that produced it
stage state model started took
fetch done — 2026-08-10 14:19 2m 06s
stt done — 2026-08-10 14:21 19s
chunk done — 2026-08-10 14:21 0s
text_embed done — 2026-08-10 19:51 0s
keyframe done — 2026-08-10 14:21 1m 32s
ocr done — 2026-08-10 14:23 12s
frame_embed done — 2026-08-10 19:51 7s

Frames, and what the machine read

  • 0:03 #0 done2 line(s)

    shot 0·sharpness 658.4

    1. AlEngineer0.98
    2. EUROPE1.00

  • 0:08 #1 done2 line(s)

    shot 1·sharpness 829.1

    1. PRESENTINGSPONSOR1.00
    2. Google DeepMind1.00

  • 0:13 #2 done3 line(s)

    shot 2·sharpness 907.7

    1. PLATINUM SPONSORS0.98
    2. # Braintrust0.96
    3. WorkOS OpenAI0.95

  • 0:43 #3 done7 line(s)

    shot 3·sharpness 580.5

    1. AlEngineer0.99
    2. EUROPE1.00
    3. Madison Faulkner1.00
    4. NEA ∞Meta0.94
    5. NEA0.87
    6. AlEngineer0.99
    7. EUROPE1.00

  • 1:18 #4 done31 line(s)

    shot 4·sharpness 1864.6

    1. LJAZ CAJ0.72
    2. *★★0.58
    3. ★1.00
    4. AIE1.00
    5. ★1.00
    6. ★1.00
    7. ★1.00
    8. ★1.00
    9. ★1.00
    10. 41.00
    11. Madison Faulkner1.00
    12. Hugo Santos1.00
    13. NEA1.00
    14. ∞Meta1.00
    15. N0.73
    16. Google1.00
    17. Stanford1.00
    18. University1.00
    19. AlEngineer0.96
    20. NEA1.00
    21. 440.69
    22. 40.94
    23. X0.56
    24. EUROPE1.00
    25. /Trade Secrets © NEA 20250.89
    26. Madison Fau0.98
    27. NEA 0Meta0.76
    28. AlEngineer1.00
    29. Braintrust1.00
    30. WorkOS OpenAI0.98
    31. 20260.92

  • 1:28 #5 done21 line(s)

    shot 5·sharpness 1438.8

    1. Page 3 of 331.00
    2. *★★0.55
    3. AIE1.00
    4. ★0.99
    5. Agentic Software is1.00
    6. ★1.00
    7. ★1.00
    8. ★1.00
    9. ★1.00
    10. Breaking Traditional CI/CD0.99
    11. N0.85
    12. namespace1.00
    13. AlEngineer0.98
    14. AlEngineer1.00
    15. 40.90
    16. EUROPE1.00
    17. EUROPE1.00
    18. AlEngineer1.00
    19. Braintrust1.00
    20. WorkOS OpenAI0.96
    21. 20260.97

  • 1:52 #6 done60 line(s)

    shot 6·sharpness 2620.0

    1. Foundation models are decomposing into agents1.00
    2. Model 1: Monolithic Agents0.99
    3. Model 2: Microservice Agents0.99
    4. (LLM as the engine)0.98
    5. (LLM as a tool)0.96
    6. Uasr Qusry0.93
    7. Aggregate & Respond0.98
    8. Client1.00
    9. LLM Query Understanding1.00
    10. Orchestration layer combines outputs0.99
    11. and delivers the completed task result0.96
    12. Human or agent0.99
    13. ★1.00
    14. AIE1.00
    15. ★1.00
    16. Query Planning1.00
    17. Query/Task1.00
    18. ★1.00
    19. Orchestration0.99
    20. System receives a complex task or0.95
    21. query0.99
    22. ★1.00
    23. ★1.00
    24. ★1.00
    25. Break down the task0.97
    26. Layer1.00
    27. Tool Selection1.00
    28. Assign Subtasks1.00
    29. Control Plane1.00
    30. Data1.00
    31. Relevant agents are identified0.98
    32. and tasknd concurrently0.95
    33. Monitor & Manage1.00
    34. Execution1.00
    35. Unstructured0.97
    36. Structured1.00
    37. Mossago Queue0.95
    38. Collect Results1.00
    39. Agents share results1.00
    40. Result Synthesis1.00
    41. LLM Rm0.87
    42. Data1.00
    43. Deliver Subtasks0.99
    44. LLM Racponse Genaration0.95
    45. Relevant agents are identified1.00
    46. and tasked1.00
    47. Tools (incl. LLMs)1.00
    48. Memory Update1.00
    49. AlEngineer1.00
    50. NEA1.00
    51. User Response1.00
    52. 440.75
    53. 40.87
    54. Return Results1.00
    55. EUROPE1.00
    56. TradleSecrets © NEA 20250.89
    57. AlEngineer0.99
    58. AlEngineer0.96
    59. 20260.99
    60. EUROPE1.00

  • 2:09 #7 done62 line(s)

    shot 7·sharpness 3030.0

    1. Software Development Lifecycle Now1.00
    2. Plan & Code0.98
    3. Build, Test, & Integrate1.00
    4. Release & Deploy1.00
    5. AATLASSIAN0.97
    6. Bitbucket1.00
    7. Dagger1.00
    8. BrowserStack1.00
    9. POSTHAN bauplan0.96
    10. LaunchDarkly1.00
    11. harness1.00
    12. E2B1.00
    13. Daytona1.00
    14. snyk1.00
    15. Entire1.00
    16. AIE1.00
    17. ★1.00
    18. ★1.00
    19. GitHub1.00
    20. Linear1.00
    21. ARCADE1.00
    22. Composio0.96
    23. NEON1.00
    24. namespace1.00
    25. STATSIG1.00
    26. TENSORZERO0.99
    27. ★1.00
    28. ★1.00
    29. ★1.00
    30. ★1.00
    31. GitHub0.97
    32. Actions0.99
    33. GitLab1.00
    34. harness1.00
    35. namespace1.00
    36. DevOps1.00
    37. armory1.00
    38. circleci1.00
    39. EARTHLY1.00
    40. depot1.00
    41. DATADOG dynatrace new relic0.96
    42. W Windsurf0.95
    43. CURSOR1.00
    44. FACTORY1.00
    45. Cognition1.00
    46. PagerDuty splunk> sumo logic0.99
    47. Claude1.00
    48. warp1.00
    49. Augment0.99
    50. Operate & Monitor1.00
    51. IDE1.00
    52. Autonomous Agent1.00
    53. AlEngineer1.00
    54. NEA1.00
    55. 40.55
    56. 40.95
    57. EUROPE1.00
    58. Trodle Serets © NEA 20250.88
    59. AlEngineer0.99
    60. AlEngineer0.97
    61. 20261.00
    62. EUROPE1.00

  • 2:22 #8 done23 line(s)

    shot 8·sharpness 1332.0

    1. Page 6 of 331.00
    2. ***0.54
    3. ★0.99
    4. AIE1.00
    5. ★0.92
    6. Explain to me why Cl/CD is dead1.00
    7. ★1.00
    8. ★1.00
    9. ★1.00
    10. ★1.00
    11. 40.99
    12. N1.00
    13. namespace1.00
    14. AlEngineer0.99
    15. AlEngineer1.00
    16. 40.92
    17. EUROPE1.00
    18. EUROPE1.00
    19. namespace1.00
    20. AlEngineer0.99
    21. AlEngineer0.97
    22. 20260.99
    23. EUROPE1.00

  • 2:30 #9 done26 line(s)

    shot 9·sharpness 2473.7

    1. How Cl/CD pipelines work today0.99
    2. ★1.00
    3. Human developer with repo1.00
    4. Each PR takes ~10 minutes0.98
    5. AIE1.00
    6. permissions submits ~1 diff a1.00
    7. to verify1.00
    8. ★1.00
    9. day to 1 repo0.98
    10. ★1.00
    11. ★1.00
    12. 1 Repo1.00
    13. Runner1.00
    14. Human developer addresses1.00
    15. GitHub Actions uses1.00
    16. failed test cases1.00
    17. runners to run build, test,0.99
    18. and deploy steps1.00
    19. AlEngineer1.00
    20. NEA1.00
    21. 40.61
    22. EUROPE1.00
    23. AlEngineer0.99
    24. AlEngineer0.95
    25. 20261.00
    26. EUROPE1.00

  • 2:35 #10 done12 line(s)

    shot 10·sharpness 646.1

    1. How1.00
    2. elines work today0.97
    3. AlEngineer0.96
    4. EUROPE1.00
    5. Human developer with repo1.00
    6. permissions submits ~1 diff a0.97
    7. day to 1 repo1.00
    8. 1 Repo0.99
    9. Human developer addresses1.00
    10. failed test cases1.00
    11. AlEngineer0.99
    12. EUROPE1.00

  • 3:02 #11 done33 line(s)

    shot 11·sharpness 2233.0

    1. Where CI/CD fails at agent scale0.99
    2. AIE1.00
    3. ★1.00
    4. repo permissions which submit0.98
    5. Human manages agents with0.99
    6. N PRs to N repos0.97
    7. Each PR takes ~10 minutes to1.00
    8. verify1.00
    9. ★1.00
    10. ★1.00
    11. ★1.00
    12. ★1.00
    13. Repo1.00
    14. Repo1.00
    15. Runner1.00
    16. Runner1.00
    17. Repo1.00
    18. Repo1.00
    19. Runner1.00
    20. Runner1.00
    21. Agents correct failed cases in0.99
    22. GitHub Actions uses runners to1.00
    23. Cl/CD pipelines0.99
    24. run build, test, and deploy1.00
    25. steps1.00
    26. AlEngineer1.00
    27. NEA1.00
    28. 40.97
    29. EUROPE1.00
    30. adle Secrets © NEA 20250.90
    31. Engineering the future of Al0.99
    32. AlEngineer0.99
    33. 20260.93

  • 3:31 #12 done25 line(s)

    shot 12·sharpness 2968.5

    1. Bursts, fan-out, and ephemerality have material impact on cache0.99
    2. read/writes and hit rate over time1.00
    3. # of Read/Writes1.00
    4. # of Read/Writes0.99
    5. AIE1.00
    6. ★1.00
    7. ★1.00
    8. ★1.00
    9. Time1.00
    10. Time1.00
    11. Human1.00
    12. Agent1.00
    13. One PR/day1.00
    14. Thousands of short-lived1.00
    15. Predictable queues1.00
    16. branches/PRs1.00
    17. Local caches often warm1.00
    18. Runners queue; cache1.00
    19. miss rates spike0.98
    20. AlEngineer0.98
    21. NEA1.00
    22. EUROPE1.00
    23. AlEngineer1.00
    24. Engineering the future of Al0.99
    25. 20260.99

  • 3:58 #13 done39 line(s)

    shot 13·sharpness 2098.7

    1. Github activity over the last 3 mos (white line = commits)1.00
    2. Activity Over Time0.98
    3. *★★0.55
    4. ★1.00
    5. ★★0.96
    6. 800M0.95
    7. 400K1.00
    8. AIE1.00
    9. ★1.00
    10. 400M1.00
    11. 300K1.00
    12. ★1.00
    13. ★1.00
    14. ★1.00
    15. 40.99
    16. 200K1.00
    17. 400M1.00
    18. 100K1.00
    19. 800M1.00
    20. Dec 291.00
    21. Jan 160.99
    22. Feb 21.00
    23. Feb 190.98
    24. Mar 80.96
    25. Mar 251.00
    26. CommitsLines Added1.00
    27. Lines Deleted1.00
    28. N1.00
    29. namespace1.00
    30. AlEngineer0.98
    31. AlEngineer1.00
    32. 40.72
    33. EUROPE1.00
    34. EUROPE1.00
    35. namespace1.00
    36. AlEngineer0.97
    37. AlEngineer0.97
    38. 20260.96
    39. EUROPE1.00

  • 4:14 #14 done31 line(s)

    shot 14·sharpness 2119.4

    1. Cl/CD acceleration is the starting point1.00
    2. ★1.00
    3. Human manages agents with1.00
    4. AIE1.00
    5. repo permissions which submit0.99
    6. N PRs an hour to N repo0.98
    7. ★1.00
    8. Repo1.00
    9. Repo1.00
    10. Runner1.00
    11. Runner1.00
    12. ★1.00
    13. ★1.00
    14. Agent1.00
    15. Cache1.00
    16. 四0.67
    17. Repo1.00
    18. Repo1.00
    19. Runner1.00
    20. Runner1.00
    21. Agents correct failed cases in0.98
    22. Cl/CD pipelines0.99
    23. AlEngineer1.00
    24. NEA1.00
    25. 40.98
    26. EUROPE1.00
    27. Traodle Secrets © NEA 20250.90
    28. AlEngineer0.99
    29. AlEngineer0.95
    30. 20260.96
    31. EUROPE1.00

  • 4:35 #15 done27 line(s)

    shot 15·sharpness 2527.8

    1. Cl/CD cache becomes the orchestration layer for agents1.00
    2. Orchestration layer can be accelerated through hardware-software co-design directing parallelizable or real-time tasks0.99
    3. ★1.00
    4. AIE1.00
    5. ★1.00
    6. Runners1.00
    7. ★1.00
    8. ★1.00
    9. ★1.00
    10. repo permissions which submit1.00
    11. Human manages agents with1.00
    12. N PRs an hour to N repos0.99
    13. Agent Cache1.00
    14. Build Tools1.00
    15. Agents correct failed cases in0.98
    16. Cl/CD pipelines0.99
    17. Source Code0.99
    18. Management1.00
    19. AlEngineer1.00
    20. NEA1.00
    21. 40.95
    22. EUROPE1.00
    23. deSmrnts © NEA20250.78
    24. AlEngineer0.99
    25. AlEngineer0.96
    26. 20260.90
    27. EUROPE1.00

  • 4:55 #16 done59 line(s)

    shot 16·sharpness 3162.5

    1. Caching and orchestration expand into agent identity & more1.00
    2. ★1.00
    3. AIE1.00
    4. Intake1.00
    5. Cache1.00
    6. Orchestration1.00
    7. Identity1.00
    8. Retries1.00
    9. ★1.00
    10. ★1.00
    11. ★1.00
    12. ★1.00
    13. Ingress shaping &1.00
    14. de-dupe code0.99
    15. Global cache1.00
    16. orchestration1.00
    17. Route to the right0.99
    18. runners (self-0.99
    19. Short-lived creds1.00
    20. for agents1.00
    21. Parse logs1.00
    22. Auto-fix known1.00
    23. Rate-limit1.00
    24. Content-addressed1.00
    25. hosted/cloud,1.00
    26. becoming the Okta1.00
    27. classes (flaky tests,0.99
    28. upstream calls;1.00
    29. caches for1.00
    30. Intel/ARM, GPU)1.00
    31. for agents0.98
    32. missing cache1.00
    33. coalesce near-1.00
    34. dependencies,1.00
    35. with warm caches0.96
    36. Pre-flight checks at1.00
    37. seeds), otherwise0.99
    38. duplicate PRs1.00
    39. Docker layers,1.00
    40. Avoid emulation1.00
    41. the edge0.97
    42. surface RCA hints0.99
    43. before they hit0.99
    44. compiled outputs,1.00
    45. for multi-arch1.00
    46. (GitLab Duo1.00
    47. GitHub/GitLab1.00
    48. test results0.99
    49. (native Intel & Arm0.97
    50. demonstrates Al-1.00
    51. speedups)1.00
    52. RCA).1.00
    53. AlEngineer0.99
    54. NEA1.00
    55. EUROPE1.00
    56. adeSecrets © NEA 20250.91
    57. Engineering the future of Al0.99
    58. AlEngineer0.99
    59. 20260.98

  • 5:19 #17 skipped

    shot 17·duplicate of #12

  • 5:56 #18 done6 line(s)

    shot 18·sharpness 622.1

    1. Okay1.00
    2. real?1.00
    3. AlEngineer0.99
    4. EUROPE1.00
    5. AlEngineer0.99
    6. EUROPE1.00

  • 6:09 #19 done8 line(s)

    shot 19·sharpness 567.2

    1. AlEngineer0.99
    2. CI/CD1.00
    3. EUROPE1.00
    4. (it's all ager1.00
    5. N1.00
    6. namespace1.00
    7. AlEngineer0.99
    8. EUROPE1.00

  • 6:27 #20 done18 line(s)

    shot 20·sharpness 1541.2

    1. CI/CD is dead1.00
    2. AIE1.00
    3. ★1.00
    4. ★1.00
    5. ★1.00
    6. 40.99
    7. (it's all agents now)1.00
    8. namespace1.00
    9. AlEngineer0.98
    10. AlEngineer0.99
    11. 40.86
    12. EUROPE1.00
    13. EUROPE1.00
    14. (it')0.83
    15. AlEngineer1.00
    16. AlEngineer0.96
    17. 20260.93
    18. EUROPE1.00

  • 6:45 #21 done33 line(s)

    shot 21·sharpness 3620.4

    1. Before1.00
    2. After1.00
    3. humans write code1.00
    4. code generation is0.97
    5. slowly1.00
    6. cheap1.00
    7. *★*0.57
    8. AIE1.00
    9. changes are packaged0.99
    10. work is continuous1.00
    11. ★1.00
    12. as PRs1.00
    13. ★1.00
    14. ★1.00
    15. ★1.00
    16. validation moves into1.00
    17. validation runs1.00
    18. the inner loop1.00
    19. after push1.00
    20. latency becomes a1.00
    21. machine latency hides1.00
    22. first-class problem1.00
    23. behind human latency1.00
    24. N0.99
    25. namespace1.00
    26. AlEngineer0.99
    27. AlEngineer0.99
    28. 40.56
    29. EUROPE1.00
    30. EUROPE1.00
    31. Engineering the future of Al1.00
    32. AlEngineer0.99
    33. 20260.99

  • 7:36 #22 done24 line(s)

    shot 22·sharpness 1765.8

    1. Page 18 of 331.00
    2. today, the human is the agent0.99
    3. Intent1.00
    4. AIE1.00
    5. ★0.99
    6. ★1.00
    7. ★1.00
    8. ★1.00
    9. Code1.00
    10. Pull Request1.00
    11. Cl0.97
    12. Human1.00
    13. Review1.00
    14. Merge1.00
    15. N1.00
    16. namespace1.00
    17. AlEngineer0.99
    18. AlEngineer0.99
    19. 40.95
    20. EUROPE1.00
    21. EUROPE0.96
    22. AlEngineer0.99
    23. Engineering the future of Al1.00
    24. 20260.91

  • 8:08 #23 done22 line(s)

    shot 23·sharpness 2056.5

    1. today, the human is the agent1.00
    2. Intent1.00
    3. *★★0.65
    4. AIE1.00
    5. ★1.00
    6. ★1.00
    7. ★1.00
    8. Code1.00
    9. Pull Request1.00
    10. Cl0.97
    11. Human1.00
    12. Review1.00
    13. Merge1.00
    14. namespace1.00
    15. AlEngineer0.99
    16. AlEngineer1.00
    17. 40.96
    18. EUROPE1.00
    19. EUROPE1.00
    20. AlEngineer0.99
    21. Engineering the future of Al1.00
    22. 20260.90

Transcript

189 cues· 2,877 words· 15,559 chars

  1. 0:15 All right, can you all hear me?
  2. 0:18 Great, there we go.
  3. 0:20 All right, well we're only 35 to 40 minutes late, but thank you for sticking around.
  4. 0:26 We're gonna talk about why CICD is dead, and we're gonna propose that continuous compute is gonna be the next thing.
  5. 0:38 Maybe.
  6. 0:42 All right, so just quick introduction.
  7. 0:44 We're gonna have two speakers, one's getting mic'd up.
  8. 0:48 My name's Madison, and I'm a partner at NEA investing in technology.
  9. 0:51 I do focus in infra and dev tools, and I formerly used to be a meta AI researcher, so I used to lead data and AI teams.
  10. 1:00 I got really frustrated by the state of infrastructure, and so I jumped into venture to do something about it from the top down.
  11. 1:07 And then I'm also gonna introduce on behalf of my partner here, Hugo Santos.
  12. 1:13 So he's the CEO of Namespace, which is building high-performance compute infrastructure, and at this point, what we believe is going to eclipse the new CICD wave.
  13. 1:22 He also formerly led microservices at Google.
  14. 1:26 Yeah, great to be here with you folks.
  15. 1:28 So we're gonna talk about why agentic software is breaking traditional CICD.
  16. 1:34 Obviously, we're not gonna get through this today, but the point is, on the left side, what started off in agentic software was really monolithic agents.
  17. 1:45 We were really using the LLM as one engine.
  18. 1:48 But now we're moving into the right side, which is microservices with agents.
  19. 1:53 And that's how we really need to think about software development in an agentic world.
  20. 1:59 So the lifecycle is very fragmented.
  21. 2:02 This is quite a mess, right?
  22. 2:04 We've really kind of brought together all these traditional CI, CD systems, build, test, deploy, DevOps, but we also now have new IDEs, we have autonomous agentic engineering solutions, and then we have our traditional DevOps in the middle, which we believe is really going to innovate in the next year.
  23. 2:23 So let's explain why we think it's dead.
  24. 2:27 So first, how do CICD pipelines work today?
  25. 2:30 Well, we all know human developers are currently submitting one, maybe a couple of diffs when they're just writing it themselves.
  26. 2:37 And those PRs then take your colleagues a bunch of time to review.
  27. 2:41 Then you have to go through GitHub Actions and run build, test, and deploy steps.
  28. 2:45 And then finally, you're addressing those failed test case and maybe you're iterating on the diff.
  29. 2:49 So in that scenario, it was really just one or two a week.
  30. 2:55 So now, how do we think about this at agent scale?
  31. 2:57 You've got agents using the exact same systems, but they have n number of PRs, maybe n number of repos.
  32. 3:05 Still takes a similar amount of time to verify, unless you're using review bots, which gets a little crazy.
  33. 3:12 And then we correct those failed cases, just like we did in the past scenario.
  34. 3:16 So what ends up happening?
  35. 3:17 With a human, pretty predictable.
  36. 3:20 And you've got local caches, which are often warm.
  37. 3:23 With an agent, this starts to get really complicated.
  38. 3:26 You have thousands of short-lived branches.
  39. 3:29 It's all trying to pull the same code base in a few different directions.
  40. 3:34 You start to get to a point where merging all these different versions together is really impossible.
  41. 3:39 And that is where we start to have a huge problem.
  42. 3:43 So let's look at in real time, GitHub activity has gotten absolutely crazy.
  43. 3:48 The white line here is the actual number of commits in the last couple of months.
  44. 3:53 And then the number of lines added versus deleted.
  45. 3:56 I mean, this is just an unbelievable spike.
  46. 4:01 So how do we start with replacing CICD?
  47. 4:05 Well, the starting point should be at the acceleration.
  48. 4:08 So obviously right now, I know a lot of you are struggling with very slow build test and deploy times for your CICD solutions.
  49. 4:16 This is a very common problem.
  50. 4:19 But where we're headed is being able to first speed that up by inserting over the existing GitHub actions and other underlying infrastructure for CICD.

Chapters

  1. 0:00 Introduction and speaker bios
  2. 1:28 Why agentic software is breaking traditional CI/CD
  3. 1:59 The fragmented lifecycle of modern software development
  4. 2:25 How traditional CI/CD pipelines work
  5. 2:55 The problems with CI/CD at agent scale
  6. 4:04 Replacing CI/CD with acceleration and orchestration
  7. 6:12 Real-world solutions and the future of agentic loops
  8. 7:23 The role of the human as the agent
  9. 8:43 Why Pull Requests (PRs) are becoming a bottleneck
  10. 10:00 A new architecture: Intent and plan-based development
  11. 11:58 Moving toward fully automated internal/external validation
  12. 13:46 The premerge queue and human-in-the-loop review
  13. 15:20 The future: Parallel development in the multiverse
  14. 16:51 Conclusion: The shifting role of CI and governance

Open at this second