read-only demo

Videos qdZzND79mcg

Beyond the Harness: A Journey Towards Adaptative Engineering - Rajiv Chandegra, Annicha Labs

index_state ready data_status ok

AI Engineer· published 2026-07-07· 0:37:01· en-US· indexed 2026-08-11 04:49

Open on YouTube

Scene timeline

  1. Shot 0, 0:00 to 0:26, 1 of 1 keyframes kept
  2. Shot 1, 0:26 to 0:52, 1 of 1 keyframes kept
  3. Shot 2, 0:52 to 1:18, 0 of 1 keyframes kept
  4. Shot 3, 1:18 to 1:44, 0 of 1 keyframes kept
  5. Shot 4, 1:44 to 2:10, 0 of 1 keyframes kept
  6. Shot 5, 2:10 to 2:36, 0 of 1 keyframes kept
  7. Shot 6, 2:36 to 3:02, 0 of 1 keyframes kept
  8. Shot 7, 3:02 to 3:28, 1 of 1 keyframes kept
  9. Shot 8, 3:28 to 3:54, 0 of 1 keyframes kept
  10. Shot 9, 3:54 to 4:20, 0 of 1 keyframes kept
  11. Shot 10, 4:20 to 4:46, 0 of 1 keyframes kept
  12. Shot 11, 4:46 to 5:13, 0 of 1 keyframes kept
  13. Shot 12, 5:13 to 5:41, 1 of 1 keyframes kept
  14. Shot 13, 5:41 to 6:09, 0 of 1 keyframes kept
  15. Shot 14, 6:09 to 6:37, 0 of 1 keyframes kept
  16. Shot 15, 6:37 to 7:06, 1 of 1 keyframes kept
  17. Shot 16, 7:06 to 7:34, 0 of 1 keyframes kept
  18. Shot 17, 7:34 to 8:02, 0 of 1 keyframes kept
  19. Shot 18, 8:02 to 8:28, 1 of 1 keyframes kept
  20. Shot 19, 8:28 to 8:54, 0 of 1 keyframes kept
  21. Shot 20, 8:54 to 9:19, 0 of 1 keyframes kept
  22. Shot 21, 9:19 to 9:45, 0 of 1 keyframes kept
  23. Shot 22, 9:45 to 10:10, 0 of 1 keyframes kept
  24. Shot 23, 10:10 to 10:36, 0 of 1 keyframes kept
  25. Shot 24, 10:36 to 11:02, 1 of 1 keyframes kept
  26. Shot 25, 11:02 to 11:27, 0 of 1 keyframes kept
  27. Shot 26, 11:27 to 11:53, 0 of 1 keyframes kept
  28. Shot 27, 11:53 to 12:19, 0 of 1 keyframes kept
  29. Shot 28, 12:19 to 12:44, 0 of 1 keyframes kept
  30. Shot 29, 12:44 to 13:10, 0 of 1 keyframes kept
  31. Shot 30, 13:10 to 13:36, 0 of 1 keyframes kept
  32. Shot 31, 13:36 to 14:01, 0 of 1 keyframes kept
  33. Shot 32, 14:01 to 14:27, 0 of 1 keyframes kept
  34. Shot 33, 14:27 to 14:53, 0 of 1 keyframes kept
  35. Shot 34, 14:53 to 15:19, 1 of 1 keyframes kept
  36. Shot 35, 15:19 to 15:46, 0 of 1 keyframes kept
  37. Shot 36, 15:46 to 16:18, 1 of 1 keyframes kept
  38. Shot 37, 16:18 to 16:50, 0 of 1 keyframes kept
  39. Shot 38, 16:50 to 17:21, 0 of 1 keyframes kept
  40. Shot 39, 17:21 to 17:48, 1 of 1 keyframes kept
  41. Shot 40, 17:48 to 18:15, 0 of 1 keyframes kept
  42. Shot 41, 18:15 to 18:42, 0 of 1 keyframes kept
  43. Shot 42, 18:42 to 19:08, 0 of 1 keyframes kept
  44. Shot 43, 19:08 to 19:35, 0 of 1 keyframes kept
  45. Shot 44, 19:35 to 20:02, 0 of 1 keyframes kept
  46. Shot 45, 20:02 to 20:28, 1 of 1 keyframes kept
  47. Shot 46, 20:28 to 20:55, 0 of 1 keyframes kept
  48. Shot 47, 20:55 to 21:22, 0 of 1 keyframes kept
  49. Shot 48, 21:22 to 21:54, 1 of 1 keyframes kept
  50. Shot 49, 21:54 to 22:26, 0 of 1 keyframes kept
  51. Shot 50, 22:26 to 22:58, 0 of 1 keyframes kept
  52. Shot 51, 22:58 to 23:23, 1 of 1 keyframes kept
  53. Shot 52, 23:23 to 23:48, 1 of 1 keyframes kept
  54. Shot 53, 23:48 to 24:14, 1 of 1 keyframes kept
  55. Shot 54, 24:14 to 24:39, 0 of 1 keyframes kept
  56. Shot 55, 24:39 to 25:04, 1 of 1 keyframes kept
  57. Shot 56, 25:04 to 25:30, 1 of 1 keyframes kept
  58. Shot 57, 25:30 to 25:55, 0 of 1 keyframes kept
  59. Shot 58, 25:55 to 26:18, 1 of 1 keyframes kept
  60. Shot 59, 26:18 to 26:46, 1 of 1 keyframes kept
  61. Shot 60, 26:46 to 27:14, 0 of 1 keyframes kept
  62. Shot 61, 27:14 to 27:15, 1 of 1 keyframes kept
  63. Shot 62, 27:15 to 27:42, 1 of 1 keyframes kept
  64. Shot 63, 27:42 to 28:09, 0 of 1 keyframes kept
  65. Shot 64, 28:09 to 28:36, 0 of 1 keyframes kept
  66. Shot 65, 28:36 to 29:02, 0 of 1 keyframes kept
  67. Shot 66, 29:02 to 29:29, 0 of 1 keyframes kept
  68. Shot 67, 29:29 to 29:56, 0 of 1 keyframes kept
  69. Shot 68, 29:56 to 30:23, 0 of 1 keyframes kept
  70. Shot 69, 30:23 to 30:43, 1 of 1 keyframes kept
  71. Shot 70, 30:43 to 31:15, 1 of 1 keyframes kept
  72. Shot 71, 31:15 to 31:47, 0 of 1 keyframes kept
  73. Shot 72, 31:47 to 32:18, 1 of 1 keyframes kept
  74. Shot 73, 32:18 to 32:46, 1 of 1 keyframes kept
  75. Shot 74, 32:46 to 33:14, 0 of 1 keyframes kept
  76. Shot 75, 33:14 to 33:43, 1 of 1 keyframes kept
  77. Shot 76, 33:43 to 34:11, 0 of 1 keyframes kept
  78. Shot 77, 34:11 to 34:40, 0 of 1 keyframes kept
  79. Shot 78, 34:40 to 35:09, 0 of 1 keyframes kept
  80. Shot 79, 35:09 to 35:41, 1 of 1 keyframes kept
  81. Shot 80, 35:41 to 36:13, 0 of 1 keyframes kept
  82. Shot 81, 36:13 to 36:44, 0 of 1 keyframes kept
  83. Shot 82, 36:44 to 37:00, 1 of 1 keyframes kept

83 shot(s).

keyframes kept every frame deduplicated

What was stored

cues
317
whisperx 317
chunks
64
from 317 cues
keyframes
28
kept of 83 captured
frames with text
28
464 lines read
chapters
0
from the source metadata
keyframe bytes
7.1 MB
word timings on 317 cues

Provenance

Each pipeline stage, its state and the model that produced it
stage state model started took
fetch done 2026-08-11 04:45 1m 09s
stt done 2026-08-11 04:46 34s
chunk done 2026-08-11 04:47 0s
text_embed done 2026-08-11 04:47 1s
keyframe done 2026-08-11 04:47 1m 42s
ocr done 2026-08-11 04:49 13s
frame_embed done 2026-08-11 04:49 5s

Frames, and what the machine read

  • 0:18 #0 done7 line(s)

    shot 0·sharpness 1257.2

    1. Beyond the1.00
    2. Harness1.00
    3. The journey towards Adaptative Engineering1.00
    4. jiv Chandegra1.00
    5. ANICHHA0.95
    6. rajivchandegra.com0.99
    7. LIBS0.96
  • 0:46 #1 done17 line(s)

    shot 1·sharpness 3758.6

    1. Current AI Engineering: Use a fixed harnesses to steer agents0.99
    2. Fixed roles, topology, sequencing, tool access etc0.98
    3. Reliable, Replicable, Auditable0.98
    4. Great for well-defined engineering problems (most)0.99
    5. Engineer's role: Build/Use Harness + Steer Agents0.99
    6. But, future will see at least two explosions1.00
    7. Models so powerful that Harnesses constantly outdated1.00
    8. Exposure to real world - dynamic/messy - means a fixed harness is brittle0.99
    9. Enter Adaptative Engineering: Harness emerges from agents interacting0.99
    10. You allow the necessary harness to emerge and adapt mid-engineering1.00
    11. Engineer's role:1.00
    12. Design constraints, including rules of interaction amongst agents1.00
    13. Apply selection pressures0.99
    14. jiv Chandegra0.99
    15. ANICHHA0.90
    16. fajivchandegra.com0.98
    17. L\BS0.93
  • 1:05 #2 skipped

    shot 2·duplicate of #1

  • 1:41 #3 skipped

    shot 3·duplicate of #1

  • 1:49 #4 skipped

    shot 4·duplicate of #1

  • 2:16 #5 skipped

    shot 5·duplicate of #1

  • 2:49 #6 skipped

    shot 6·duplicate of #1

  • 3:23 #7 done30 line(s)

    shot 7·sharpness 2567.7

    1. Harness Engineering1.00
    2. Skills: Role, Typology, Capabilities, Rules0.99
    3. e.g. /PRD /to-issues /implement /front-end0.99
    4. Code, Documents, Issues, Handoffs0.99
    5. Agents1.00
    6. Output0.99
    7. LLM1.00
    8. Harness1.00
    9. Agents1.00
    10. Output1.00
    11. Outcome1.00
    12. Agents1.00
    13. Output0.99
    14. Stateless1.00
    15. Tools1.00
    16. Feature1.00
    17. Next-token prediction0.99
    18. System Prompt1.00
    19. Problem-Solved1.00
    20. -TrainingCut-0ffs0.95
    21. Agents.md1.00
    22. Human Review1.00
    23. Context Window1.00
    24. Permissions1.00
    25. and much more!1.00
    26. Looping1.00
    27. jiv Chandegra0.98
    28. NICHHA1.00
    29. rajivchandegra.com0.97
    30. L\BS0.92
  • 3:51 #8 skipped

    shot 8·duplicate of #7

  • 4:12 #9 skipped

    shot 9·duplicate of #7

  • 4:26 #10 skipped

    shot 10·duplicate of #7

  • 4:55 #11 skipped

    shot 11·duplicate of #7

  • 5:27 #12 done24 line(s)

    shot 12·sharpness 3486.1

    1. Harnesses1.00
    2. Harness properties:1.00
    3. A harness is the scaffold around a model that governs0.99
    4. how it operates. It turns model's behaviour into1.00
    5. useful work.0.99
    6. Orchestration0.98
    7. -Roles0.94
    8. The model is the engine; the harness is everything1.00
    9. Permissions and Rules0.98
    10. builtaround the engine.0.99
    11. Memory and Persistence0.98
    12. -Sequencing1.00
    13. e.g. Claude Code, Cursor, Codex, Cline, Goose,0.98
    14. Hermes, OpenClaw, LangChain, Pi0.97
    15. Tool Access0.96
    16. -Routing0.99
    17. Communication Protocols1.00
    18. Observability/Testing0.99
    19. It guides and is guided by your own0.98
    20. design andengineeringphilosophy1.00
    21. jiv Chandegra1.00
    22. NICHHA1.00
    23. rajivchandegra.com0.98
    24. L\BS0.97
  • 6:00 #13 skipped

    shot 13·duplicate of #12

  • 6:13 #14 skipped

    shot 14·duplicate of #12

  • 7:02 #15 done10 line(s)

    shot 15·sharpness 2325.8

    1. Fixed Harnesses and Taylorism0.99
    2. Leads to Outcomes that are:0.98
    3. Reliable / Reproducible0.98
    4. Auditable1.00
    5. -Linear Causality0.99
    6. Like a Factory Line. Efficient, but hard to vary1.00
    7. jiv Chandegra0.99
    8. ANICHHA0.92
    9. rajivchandegra.com0.98
    10. LIBS0.85
  • 7:20 #16 skipped

    shot 16·duplicate of #15

  • 7:37 #17 skipped

    shot 17·duplicate of #15

  • 8:08 #18 done16 line(s)

    shot 18·sharpness 4022.8

    1. Uses + Failure Modes of Fixed (factory) Harnessing0.98
    2. Useful For Fixed(complicated)problems1.00
    3. Closed, deterministic systems, where problems are well defined and fixed.0.99
    4. Where speed, reproducbility, auditability, certifability is needed0.99
    5. -i.e. non-realtime product releases where there is a clear temporal0.99
    6. seperation between problem - solution0.99
    7. Failure Modes - bad for moving(complex) problems0.98
    8. X1.00
    9. Where it comes in contact with the real world - messy and changing0.99
    10. Hard Ceiling on Novelty - its reliability is bought by supressing variance1.00
    11. Brittlness - Every unanticipated condition requires a human to update the harness.0.99
    12. Accelerating Model Capabilities will be limited by fixed Harnessing0.99
    13. jiv Chandegra0.99
    14. ANICHHA0.92
    15. fajivchandegra.com1.00
    16. L\BS0.95
  • 8:38 #19 skipped

    shot 19·duplicate of #18

  • 9:11 #20 skipped

    shot 20·duplicate of #18

  • 9:27 #21 skipped

    shot 21·duplicate of #18

  • 10:00 #22 skipped

    shot 22·duplicate of #18

  • 10:18 #23 skipped

    shot 23·duplicate of #18

Transcript

317 cues· 4,723 words· 27,562 chars

  1. 0:01 Hello, I'm Rajiv.
  2. 0:03 I'm a practicing medical doctor here in London and been doing some AI engineering for the past few years, particularly interested in real-world application and really the future, which is multi-agent, multi-human, multi-institutional collaboration through a company called Anitcha Labs.
  3. 0:22 And really the premise of my talk today is to introduce a new design philosophy
  4. 0:28 for the future of AI engineering, which is beyond a harness or beyond a fixed static harness and towards adaptive engineering.
  5. 0:41 This slide basically summarizes my entire talk.
  6. 0:46 And really, it's about exploring what the current AI engineering paradigm is, which is to use fixed harnesses to steer agents.
  7. 0:56 So we either use or build an existing harness like Pi, Cloud Code, Cursor, Codex, whatever it is, ahead of runtime that remains fairly consistent throughout the engineering process.
  8. 1:08 and is largely constrained with fixed roles and tools and sequencing, which, again, ahead of the process, beginning of the process, we can tweak.
  9. 1:18 And it essentially produces very reliable and orderly work.
  10. 1:21 And it's great for most engineering problems and ideas.
  11. 1:26 And so your role as an AI engineering right now is to use a harness and steer agents.
  12. 1:32 But the future is going to be different in at least two ways.
  13. 1:37 Number one, models are going to become so powerful that existing fixed harnesses are constantly going to become outdated.
  14. 1:45 And number two is there's going to be a huge exposure to the real world.
  15. 1:50 Right now, a lot of AI engineering is software-based.
  16. 1:54 It's within behind a screen.
  17. 1:57 But actually, as things get more powerful, as models get more powerful, we're going to have exposure to the real world, where I'd argue that this is where the real challenges are.
  18. 2:09 And the real world is dynamic and messy.
  19. 2:12 It's full of multi-agent, multi-human, cross-institutional, and really touching the physical world, which means that a fixed harness is quite brittle.
  20. 2:23 And so that's where we enter what I coin as adaptive engineering, where you actually allow the harness to emerge and adapt mid engineering to find its most optimal position and structure.
  21. 2:40 And essentially the engineer's role is going to be to design some of the constraints, which are kind of the rules of play and really allowing the harness
  22. 2:50 to emerge, stabilize, change, and eventually dissolve as you go through the runtime of engineering.
  23. 3:03 So let's look firstly at the current paradigm for AI engineering.
  24. 3:10 And it's essentially where a harness serves as the primary method to guide an LLM, which is typically stateless,
  25. 3:21 and make it into something quite useful.
  26. 3:23 And there's a couple of aspects of a harness, which I won't go to in detail, but essentially things like system prompts, which is established by the harness vendor, meaning that users can't actually modify it.
  27. 3:34 And this essentially tells the harness what it can and can't do.
  28. 3:41 You've also got the agents.md files or the claude.md if you're using claude code, which is loaded into every context window at the start of a session, tool calling.
  29. 3:52 And essentially, then you have the genesis of agents, which are specialized entities that have been harnessed.
  30. 4:01 granting them unique capabilities and allowing them to differentiate from other agents.
  31. 4:08 And these capabilities are typically manifest as specific skills when it comes to role, typology, their capabilities, the rules.
  32. 4:18 And that enables them to deliver targeted outputs and specified outputs, whether it be code,
  33. 4:26 documentation issues handoffs and and most recently we've seen the genesis of kind of loops loop engineering has become a thing um so that eventually we reach some sort of outcome
  34. 4:42 whether it's a feature, a problem solved, which is eventually reviewed by a human.
  35. 4:49 So what you see from this is that before running the engineering process, the entire harness has been predetermined or pre-engineered, and that's harness engineering.
  36. 5:02 And it's been remarkable at getting the most out of the latest models.
  37. 5:16 And just sort of stepping a bit back, defining exactly what a harness is.
  38. 5:21 Essentially, the model is the engine, and the harness is everything built around to make that engine useful.
  39. 5:30 There's many different types of harnesses.
  40. 5:32 I mean, there's hundreds, actually, but I guess the most common ones are, you know, CLI coding, such as Claude Code, Codex, Pi.
  41. 5:42 You have IDEs like Cursor.
  42. 5:45 multi-agent orchestration through Langchain and things like Hermes as well, Klein and Goose.
  43. 5:52 And these are all differentiated based on number one, properties and how they're configured.
  44. 5:58 But actually, more importantly, number two, based on some sort of design and engineering philosophy that they believe in.
  45. 6:08 And what makes them so great and so distinct is that they are opinionated to a lesser or greater degree.
  46. 6:16 And they allow for, I guess, different use cases.
  47. 6:21 And depending on your temperament as an engineer, you'd pick one over the other.
  48. 6:30 But what we do see in all of these is that everything is predefined.
  49. 6:34 You can customize them, of course, like, for example, the PyAgent.
  50. 6:39 It's minimalist and it's maximally extensible, but everything is predefined.

Open at this second