Videos qdZzND79mcg
Beyond the Harness: A Journey Towards Adaptative Engineering - Rajiv Chandegra, Annicha Labs
Scene timeline
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
| 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
-
- Beyond the1.00
- Harness1.00
- The journey towards Adaptative Engineering1.00
- jiv Chandegra1.00
- ANICHHA0.95
- rajivchandegra.com0.99
- LIBS0.96
-
- Current AI Engineering: Use a fixed harnesses to steer agents0.99
- Fixed roles, topology, sequencing, tool access etc0.98
- Reliable, Replicable, Auditable0.98
- Great for well-defined engineering problems (most)0.99
- Engineer's role: Build/Use Harness + Steer Agents0.99
- But, future will see at least two explosions1.00
- Models so powerful that Harnesses constantly outdated1.00
- Exposure to real world - dynamic/messy - means a fixed harness is brittle0.99
- Enter Adaptative Engineering: Harness emerges from agents interacting0.99
- You allow the necessary harness to emerge and adapt mid-engineering1.00
- Engineer's role:1.00
- Design constraints, including rules of interaction amongst agents1.00
- Apply selection pressures0.99
- jiv Chandegra0.99
- ANICHHA0.90
- fajivchandegra.com0.98
- L\BS0.93
-
- Harness Engineering1.00
- Skills: Role, Typology, Capabilities, Rules0.99
- e.g. /PRD /to-issues /implement /front-end0.99
- Code, Documents, Issues, Handoffs0.99
- Agents1.00
- Output0.99
- LLM1.00
- Harness1.00
- Agents1.00
- Output1.00
- Outcome1.00
- Agents1.00
- Output0.99
- Stateless1.00
- Tools1.00
- Feature1.00
- Next-token prediction0.99
- System Prompt1.00
- Problem-Solved1.00
- -TrainingCut-0ffs0.95
- Agents.md1.00
- Human Review1.00
- Context Window1.00
- Permissions1.00
- and much more!1.00
- Looping1.00
- jiv Chandegra0.98
- NICHHA1.00
- rajivchandegra.com0.97
- L\BS0.92
-
- Harnesses1.00
- Harness properties:1.00
- A harness is the scaffold around a model that governs0.99
- how it operates. It turns model's behaviour into1.00
- useful work.0.99
- Orchestration0.98
- -Roles0.94
- The model is the engine; the harness is everything1.00
- Permissions and Rules0.98
- builtaround the engine.0.99
- Memory and Persistence0.98
- -Sequencing1.00
- e.g. Claude Code, Cursor, Codex, Cline, Goose,0.98
- Hermes, OpenClaw, LangChain, Pi0.97
- Tool Access0.96
- -Routing0.99
- Communication Protocols1.00
- Observability/Testing0.99
- It guides and is guided by your own0.98
- design andengineeringphilosophy1.00
- jiv Chandegra1.00
- NICHHA1.00
- rajivchandegra.com0.98
- L\BS0.97
-
- Fixed Harnesses and Taylorism0.99
- Leads to Outcomes that are:0.98
- Reliable / Reproducible0.98
- Auditable1.00
- -Linear Causality0.99
- Like a Factory Line. Efficient, but hard to vary1.00
- jiv Chandegra0.99
- ANICHHA0.92
- rajivchandegra.com0.98
- LIBS0.85
-
- Uses + Failure Modes of Fixed (factory) Harnessing0.98
- Useful For Fixed(complicated)problems1.00
- Closed, deterministic systems, where problems are well defined and fixed.0.99
- Where speed, reproducbility, auditability, certifability is needed0.99
- -i.e. non-realtime product releases where there is a clear temporal0.99
- seperation between problem - solution0.99
- Failure Modes - bad for moving(complex) problems0.98
- X1.00
- Where it comes in contact with the real world - messy and changing0.99
- Hard Ceiling on Novelty - its reliability is bought by supressing variance1.00
- Brittlness - Every unanticipated condition requires a human to update the harness.0.99
- Accelerating Model Capabilities will be limited by fixed Harnessing0.99
- jiv Chandegra0.99
- ANICHHA0.92
- fajivchandegra.com1.00
- L\BS0.95
Transcript
317 cues· 4,723 words· 27,562 chars
- 0:01 Hello, I'm Rajiv.
- 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.
- 0:22 And really the premise of my talk today is to introduce a new design philosophy
- 0:28 for the future of AI engineering, which is beyond a harness or beyond a fixed static harness and towards adaptive engineering.
- 0:41 This slide basically summarizes my entire talk.
- 0:46 And really, it's about exploring what the current AI engineering paradigm is, which is to use fixed harnesses to steer agents.
- 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.
- 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.
- 1:18 And it essentially produces very reliable and orderly work.
- 1:21 And it's great for most engineering problems and ideas.
- 1:26 And so your role as an AI engineering right now is to use a harness and steer agents.
- 1:32 But the future is going to be different in at least two ways.
- 1:37 Number one, models are going to become so powerful that existing fixed harnesses are constantly going to become outdated.
- 1:45 And number two is there's going to be a huge exposure to the real world.
- 1:50 Right now, a lot of AI engineering is software-based.
- 1:54 It's within behind a screen.
- 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.
- 2:09 And the real world is dynamic and messy.
- 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.
- 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.
- 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
- 2:50 to emerge, stabilize, change, and eventually dissolve as you go through the runtime of engineering.
- 3:03 So let's look firstly at the current paradigm for AI engineering.
- 3:10 And it's essentially where a harness serves as the primary method to guide an LLM, which is typically stateless,
- 3:21 and make it into something quite useful.
- 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.
- 3:34 And this essentially tells the harness what it can and can't do.
- 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.
- 3:52 And essentially, then you have the genesis of agents, which are specialized entities that have been harnessed.
- 4:01 granting them unique capabilities and allowing them to differentiate from other agents.
- 4:08 And these capabilities are typically manifest as specific skills when it comes to role, typology, their capabilities, the rules.
- 4:18 And that enables them to deliver targeted outputs and specified outputs, whether it be code,
- 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
- 4:42 whether it's a feature, a problem solved, which is eventually reviewed by a human.
- 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.
- 5:02 And it's been remarkable at getting the most out of the latest models.
- 5:16 And just sort of stepping a bit back, defining exactly what a harness is.
- 5:21 Essentially, the model is the engine, and the harness is everything built around to make that engine useful.
- 5:30 There's many different types of harnesses.
- 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.
- 5:42 You have IDEs like Cursor.
- 5:45 multi-agent orchestration through Langchain and things like Hermes as well, Klein and Goose.
- 5:52 And these are all differentiated based on number one, properties and how they're configured.
- 5:58 But actually, more importantly, number two, based on some sort of design and engineering philosophy that they believe in.
- 6:08 And what makes them so great and so distinct is that they are opinionated to a lesser or greater degree.
- 6:16 And they allow for, I guess, different use cases.
- 6:21 And depending on your temperament as an engineer, you'd pick one over the other.
- 6:30 But what we do see in all of these is that everything is predefined.
- 6:34 You can customize them, of course, like, for example, the PyAgent.
- 6:39 It's minimalist and it's maximally extensible, but everything is predefined.
loading