Markdown source

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

Source Video

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

Relationship To World's Fair 2026

These slides are extracted from a public AI Engineer YouTube video connected to World's Fair 2026. Speaker-matched clips are supporting context unless later confirmed as exact session recordings; official livestream recordings are day-level/event-level source material.

Related Scheduled Sessions

Extracted Slides

slide-001.jpg

Slide text:

Beyond the Harness

The journey towards Adaptative Engineering

slide-002.jpg

Slide text:

Current AI Engineering: Use a fixed harnesses to steer agents

Fixed roles, topology. sequencing,tool access etc

Reliable,Replicable,Auditable

Great for well-defined engineering problems (most)

Engineer's role: Build/Use Harness + Steer Agents

But, future will see at least two explosions

Models sopowerful that Harnesses constantly outdated

r o

Enter Adaptative Engineering: Harness emerges from agents interacting

You allow the necessary harness to emerge and adapt mid-engineering Engineer's role:

Design constraints,including rules of interaction amongst agents

Apply selection pressures

uajivchandegra.com jiv Chandegra LIBS INICHHA

slide-003.jpg

Slide text:

Harnesses

Harness properties: A harness is the scaffold around a model that governs

how it operates. It turns model's behaviour into

Orchestration useful work.

-Roles Permissions and Rules The model is the engine; the harness is everything built around the engine.

-Sequencing Tool Access Memory and Persistence e.g.Claude Code,Cursor, Codex,Cline,Goose, Hermes,OpenClaw,LangChain,Pi

-Routing

Communication Protocols

-observability/Testing

design and engineering philosophy It guides and is guided by your own

ajivchandegra.com jiv Chandegra LIBS ANICHHA

slide-004.jpg

Slide text:

Uses + Failure Modes of Fixed (factory) Harnessing

Useful For Fixed (complicated) problems

Whero specd, teproducbilityo auditabllity. Certifabilityis needed Closed, deterministic Systems, Where problems are well defined and fixed.

Ee, non-realtime product releases where there fsa clear temporal Seperation between problen o Solution

Failure Modes o bad for moving (complex) problems

Where it comes in contact with the real worid o messy and changing

Hard Ceiling on Novelty o itsreliabilityis bought bysupressing variance

Brittlness o Every unanticipated condition requires a human to update the harness.

Accelerating Model Capabilities will be limited by fixed Harnessing

a2rajivchandegra.com jiu Chandegra LIBS ANICHHA

slide-005.jpg

Slide text:

How problems show up

"Managers are not confronted with problems that are independent of each other, but with dynamic situations that consists of complex systems of changing problems that interact with each other. I call such situations messes.

Managers do not solve problems, they manage messes."

-Russell Ackoff (1979, s. 93)

slide-006.jpg

Slide text:

Complicated Complex

in

Categorising

problem spaces

oxosvosoqoxd.uolxom.oyidrpe

same input,same output.You analyse-plan~execute,and a single principal can cer Complex: adaptive agents that respond to one another -a cat, a flock, a market. Beha emergent,loops back on itself （the feedback arrows).and a new order fors （the dash cluster) that you can't read off any part.You probe ~sense →respond.The factory Complicated: passive parts in fixed 1inkage -a clock.Knowable,decomposable,predic correct for the left-hand world;adaptive engineoring is only for the right-hand one.

tajivchandegra.com jiv Chandegra

slide-007.jpg

Slide text:

Principles of a Complex Adaptive System

slide-008.jpg

Slide text:

Adaptative Engineering

constraints to the ertent that the harness emerges on its own. envtronment, in ways that could not be specified in advance.? stabilises and adapts as needed, in 1 αAdaptive engineering is the discipline of designing response to the changing

The Charness becomes the output rather than an input.

You allow the agents to find a harness that is suitable for their environment or problem space. This may need to adapto mid-engeering.

The harness becomes @ self-organising. constantly evolving. multi-agent system

trajivchandegra.com jiv Chandegra LIBS ANICHHA

slide-009.jpg

Slide text:

Phase 4: Order without Centralised Control

slide-010.jpg

Slide text:

Phase 5: New Order

slide-011.jpg

Slide text:

the engineer; it Adaptive: engineering of engineering relocates does not the emphasis abolish

Exploit the model's learn and change. have power to interact, capabilities so agents You engineer the rule of the game constraints - the deciding what can harness to emerge and You allow the new based on the changing change continually

does. happen not what environment.

najivchandegra.com jiv Chandegra LIBS ANICHHA

slide-012.jpg

Slide text:

Purely Fixed Harness Claude Code Langchain Hermes. Cursor Goose Codex Cline Vertical vs horlzontal Entelligence Fully Adaptive Harness

P:v.:r+.

adaptative pre-run, vs adaptive 'during, xuntino'

[tur]

(ingosid ceatraisos -Clond) eme:gmt distilbuted

prescriptive, immovable archietcture the loop learning Single agent In emergent, multi-agent self organisation

jrajivchandegra.con jivChandegra LIBS INICH

slide-013.jpg

Slide text:

2

Engineer Open or close? QuesTION 1 Wall or goal? QUestION 2 How fast? t norisano

Constraints new moves enabling door), This s just Dhat the rule does. A rule either opens market, @ scoreboard) limit) or closes options governing (a speed (a shared language,a cano governing "you eay not? enabling you now alocked wants to win, so fit fenced playground: do Some of both. apart you either anything inside) no fence), Usually To stop itflying build a wall around give it a shared goal holds together with to orbit coherence (a tean that all) ito containerO hold by a boundary ocoherence pul1 hold by shared a container(a xor once: open one small, thing firsto watchy then widen. That is the dial (the Iate of coupling). tenpo. Don't switch This one isn'ta rule at alloft's the fullyreversible everything on at time reversible step ata scrimnage before the chaapionship - one

tvchandegra-com Chandegra LIBS ANICHHA

slide-014.jpg

Slide text:

PrOpLAiT HARHESS CHGINEIRING.THE FACTORY ADAPTIVE CNGINEERIRC. IHE FOREST

fron Hhere ordel coaes Iposed up front, by dosigner interaction Emerges afterwaids,fron

Harness Properties Unlt o2 cosign ittell Control The tarm+!! An input = you author It The agent and Its wixing. Centralised Osinglo principal Tho rolationship coupling. becone the harness Partly an output energent noras exchange, constraint pxincipal Distributed - no central

VS Boundary Closed - porutes its past Opon co.adapts with its world

Adaptative Engineer the outcome cultivate tho macro Engineer tho unit Oconstraints:

Properties Causality "hettor" [+ans) Linear, Enspectable paths Optinise tomard. fixed etric Hon-linear, energent, Irreducible Co-adapt as the landscape royes (Red Queen)

Adaptatlon Done to ito offline, between releases 'continuously Done by It = in the field,

Men it faills The failure bas an owner The failure has only a shape

Best for Closed, deterainistic. cortifiable tasks Open, shlfting, novelty-seoking frontios

rajivchandegra.con jiv Chandegra LIBS ANICHHA

slide-015.jpg

Slide text:

Uses + Failure Modes

Useful For Messy, Exploratory Work

Open ended discoveryo environments that shift fast.

Where novelty and continuous adaptation is needed for changing conditions

Failure Modes

Without genuine selection pressure, you get drift. Emergence leans towards stability, not necessarily good.

Monocultuxe isk o agents trained on similar data

Legibilitycollapses o as adaptabilityfncreases, abilityto explain decreases.

Irreversiblity olock-ins, Stable bad equilibria Accountability gap o harm fron a pattern with no culpabie author

No reproducibility or predictability ahead of run time.

cajivchandegra.com jiv Chandegra LIBS ANICHHA

slide-016.jpg

Slide text:

Takeaway

As models improve exponentially, and as AI engineering moves more into real-world, the discipline will be forced to rethink itself around continual production.

The limiting factor is not the strength of the model - it will be the adaptability of the harness.

Adaptability means multi-agent (horizontal), decentralised, mid-runtime intelligence.

Classification audit: raw/sources/slide-ai-classification/slides/qdZzND79mcg/audit.json

Slide-Derived Subjects To Review

Subject extraction uses video title, related session titles/descriptions, transcript context, and OCR text when available. OCR is best-effort and should be reviewed against the embedded slide images.