Seeing The System · Published 2026-01-24

Systems Thinking Explained in Under 10 Min

privateReview: unreviewedConfidence: high
systems thinkingEmergenceconceptual modelsholismfeedback loopsystems-thinkingfeedback-loopsreductionismsystem-boundariesmental-modelsevaluation
Open on YouTube ↗

Summary

Overview

  • Speaker: Yassine Sellami
  • Channel: Seeing The System
  • Main topic: Systems Thinking and System Awareness
  • Purpose: To educate students, professionals, and leaders on how systems thinking helps navigate complex relationships, avoid reductionist pitfalls, and enhance daily decision-making. A 10-minute comprehensive overview of systems thinking, covering fundamental definitions, the shift from analytical to systems thinking, system awareness, and how personal mental models drive systemic change.

Topic Map

Introduction to Systems Thinking

  • Explanation: Introduces the speaker's background and sets the framework for why systems thinking is valuable for decision-making and problem-solving.
  • Key claims:
    • Systems thinking helps us navigate complex relationships and symptoms.
    • Engineers often zoom in on isolated parts, leading to unintended consequences.
  • Examples:
    • Decision trees, randomness, and business problem solving.
  • Terminology:
    • Systems Thinking
    • Decision Trees
    • Reductionism
  • Why it matters: Provides a foundation for understanding how separate parts combine to form a cohesive system.

What is a System?

  • Explanation: Defines systems across natural and human-made domains, highlighting coherence, interconnectedness, and characteristic behavior.
  • Key claims:
    • A system is a set of elements or parts coherently organized and interconnected in a pattern or structure.
    • Systems produce characteristic behaviors, called functions in man-made systems or purposes in human systems.
  • Examples:
    • Respiratory system, digestive system, and ecosystem.
  • Terminology:
    • System Elements
    • Interconnectedness
    • Function
    • Purpose
  • Why it matters: Establishes a universal definition of systems before applying systems thinking frameworks.

Why Systems Thinking?

  • Explanation: Contrasts analytical thinking with systems thinking, using the analogy of a magnifying glass over isolated parts versus the whole globe.
  • Key claims:
    • Business leaders and engineers often focus narrowly on the bottom line or isolated parts.
    • Solving a part without seeing the whole generates new, unexpected problems known as emergence.
  • Examples:
    • Treasury systems trying to solve economic problems and creating inflation.
    • A farmer, animals, trees, and transportation interacting within an ecosystem.
  • Terminology:
    • Analytical Thinking
    • Bottom Line
    • Emergence
    • Holism
  • Why it matters: Demonstrates the limitations of purely analytical problem solving and the danger of ignoring systemic context.

System Awareness and the Mental Model Loop

  • Explanation: Explains how conceptual models shape what we see, what we do, and the world we create in a continuous feedback loop.
  • Key claims:
    • Our paradigms and conceptual models shape everything we see.
    • How we see the world determines how we act in it, creating the world we live in which feeds back into our thinking.
  • Examples:
    • Glasses representing lenses and paradigms through which we view reality.
  • Terminology:
    • Conceptual Models
    • Paradigms
    • Feedback Loop
    • System Awareness
  • Why it matters: Connects personal psychology and mental frameworks directly to large-scale system outcomes.

Seeing Systems: Personal Development to Systems Change

  • Explanation: Binds personal development to systems change, advocating for systems innovation over mere problem-solving.
  • Key claims:
    • Changing mental models and paradigms leads to changes in how we look at the world.
    • Instead of just problem solving, we should focus on systems innovation and modifying existing systems.
  • Examples:
    • Shifting from problem solving to systems change.
  • Terminology:
    • Personal Development
    • Systems Change
    • Systems Innovation
  • Why it matters: Offers a practical philosophy for achieving lasting transformation by addressing underlying system structures rather than symptoms.

Key Points

Definition of a System

  • Explanation: A system is a set of elements or parts coherently organized and interconnected in a pattern or structure that produces characteristic behavior.
  • Evidence: Natural and human-made examples like the respiratory system or ecosystems.
  • Practical implication: Recognizing boundaries and relationships helps accurately define the scope of any problem.

Analytical vs. Systems Thinking

  • Explanation: Analytical thinking zooms into parts to troubleshoot, while systems thinking evaluates the whole and interactions between parts.
  • Evidence: Business leaders focusing solely on profit metrics often create emergent problems elsewhere.
  • Practical implication: Decision-makers must evaluate broader systemic impacts before implementing localized solutions.

The Feedback Loop of Awareness

  • Explanation: Conceptual models shape what we see, which determines what we do, which creates the world we live in, feeding back into our thinking.
  • Evidence: Diagram cycle showing Conceptual Models -> What We See -> What We Do -> The World We Create.
  • Practical implication: Self-awareness and examination of personal biases are mandatory for effective leadership.

Frameworks, Models & Processes

System Awareness Loop

  • How it works: A four-stage continuous feedback loop connecting mental paradigms to real-world actions and outcomes.
  • Components:
    • Conceptual Models
    • What We See
    • What We Do
    • The World We Create
  • When to use: When evaluating how personal perspectives and institutional paradigms influence organizational outcomes.

Examples & Case Studies

Treasury systems attempt to solve an economic problem and unintentionally create inflation.

  • Illustrates: Emergence
  • Lesson: Solving a localized part of a system can generate unintended emergent problems across the broader economy.

A farmer manages trees, interacts with animals, and utilizes transportation for market sales.

  • Illustrates: Interconnected ecosystems and business systems
  • Lesson: Individual stakeholders cannot be analyzed in isolation from their environmental and economic networks.

Actionable Takeaways

  • Immediate:
    • Recognize that your conceptual models and paradigms shape everything you observe.
    • Look beyond isolated symptoms to understand the broader system structure.
  • Strategic:
    • Shift from traditional problem solving to systems innovation.
    • Connect personal development and mindset shifts directly to organizational and systems change.
  • Questions to investigate:
    • What unseen parts of the system are affected by our current performance metrics?
    • How do my personal biases and mental models limit my understanding of complex problems?

Claims Worth Verifying

  • A system is a set of elements or parts coherently organized and interconnected in a pattern or structure. (Definition)

Notable Quotes

"A system is a set of elements or parts that are coherently organized and interconnected in a pattern or structure that produces a characteristic behavior." "Systems thinking is a way of looking at the world. It's about understanding how things are linked together and how they affect each other within a whole system."

Compressed Summary

  • Systems consist of organized, interconnected parts producing characteristic behaviors.
  • Analytical thinking focuses on parts; systems thinking examines whole interactions.
  • Solutions to localized problems often create unintended emergent issues.
  • Personal mental models drive observation, action, and systemic reality.
  • Systems innovation replaces narrow problem solving with structural modifications.
  • Keywords: systems thinking, emergence, conceptual models, holism, feedback loop
  • Core insight: Systems thinking requires shifting focus from isolated parts and symptoms to holistic interconnections, recognizing that our mental models actively shape the systemic reality we create.

Core insights

6
Failure Modemedium noveltymoderate evidence

Local fixes inside a system predictably generate new failure classes elsewhere ('emergence'); the summary's treasury example shows one subsystem solving an economic problem while creating inflation — i.e., the fix becomes a new defect at the whole-system level.

Why it matters

Any agentic architecture that optimizes per-component metrics (per-tool accuracy, per-agent reward, per-module SLO) is structurally exposed to the same pattern: component tests pass while the pipeline acquires a new failure mode that only appears at the interaction level.

Generalization

Treat emergence as an expected cost of decomposition-based optimization rather than an anomaly; every localized optimization needs a whole-system side-effect check before being accepted.

Solving a part without seeing the whole generates new, unexpected problems known as emergence.
Open source video
Treasury systems attempt to solve an economic problem and unintentionally create inflation.
Open source video
Architecturemedium noveltymoderate evidence

Defining the system's boundary and interconnections is treated as a prerequisite for correctly scoping any problem, not as a downstream detail; the system definition centers on elements coherently organized and interconnected producing a characteristic behavior.

Why it matters

For agent systems, where you draw the boundary determines what counts as context, what counts as an external environment, and therefore what gets instrumented and evaluated — boundary choice silently determines the failure modes you can even see.

Generalization

Make boundary selection an explicit, versioned design decision (what is inside the agent, what is a tool, what is environment) rather than an implicit consequence of framework defaults.

A system is a set of elements or parts that are coherently organized and interconnected in a pattern or structure that produces a characteristic behavior.
Open source video
Recognizing boundaries and relationships helps accurately define the scope of any problem.
Open source video
Mechanismmedium noveltyweak evidence

The awareness loop is a closed circuit: conceptual models determine what is seen, seeing determines action, action constructs the world, and that constructed world feeds back into the models — with no external ground-truth checkpoint named anywhere in the cycle.

Why it matters

Agents whose observations are mediated by their own prior context, memory, or filters operate the same closed loop: a wrong internal model can drive actions that make the environment look consistent with the wrong model, and nothing in the loop falsifies it.

Generalization

Any system that both perceives and acts on a shared environment needs an out-of-loop verification channel (independent evaluator, external probe, held-out ground truth) or its self-consistency becomes indistinguishable from correctness.

Our paradigms and conceptual models shape everything we see.
Open source video
How we see the world determines how we act in it, creating the world we live in which feeds back into our thinking.
Open source video
Tradeoffmedium noveltymoderate evidence

Analytical thinking (zoom into parts to troubleshoot) and systems thinking (evaluate the whole and the interactions between parts) are presented as complementary rather than substitutable, with narrow part-focus explicitly linked to unintended consequences.

Why it matters

This maps directly onto the split between unit/component testing and end-to-end interaction evaluation in agent stacks: component debugging is tractable and precise but blind to interaction effects that only exist when modules run together.

Generalization

Budget for two distinct evaluation layers — decomposition-level diagnostics for localizing faults, and interaction-level evaluation for detecting effects no component test can express.

Analytical thinking zooms into parts to troubleshoot, while systems thinking evaluates the whole and interactions between parts.
Open source video
Engineers often zoom in on isolated parts, leading to unintended consequences.
Open source video
Mental Modellow noveltyweak evidence

The speaker draws a distinction between functions (characteristic behavior of man-made systems) and purposes (characteristic behavior of human systems), implying the two must be specified differently.

Why it matters

Agent and multi-agent designs that include human stakeholders conflate a declared objective with the actual purpose being pursued; treating purpose as an explicit, separately specified artifact is a different design obligation than specifying a function.

Generalization

Separate the machine-checkable objective (function) from the stakeholder intent (purpose) in the system spec, since only the former can be optimized against directly.

Systems produce characteristic behaviors, called functions in man-made systems or purposes in human systems.
Open source video
Practicelow noveltyweak evidence

The advocated intervention point is structural: shift from problem solving to systems innovation, i.e., modifying the existing system rather than relieving symptoms, with mental-model change named as the lever that propagates into system change.

Why it matters

Framed for engineering, this is the argument for changing loops, roles, and control structure instead of repeatedly patching prompts, tool descriptions, or single outputs — a recurring tension in agent maintenance.

Generalization

Track patch frequency per symptom; repeated symptom-level fixes are a signal that the structural cause (agent topology, feedback path, boundary) is the correct level of intervention.

Instead of just problem solving, we should focus on systems innovation and modifying existing systems.
Open source video
Changing mental models and paradigms leads to changes in how we look at the world.
Open source video

Deep dives

5

Interaction-level detection of emergent cross-component failures

Research question

Can emergent, cross-component failures be detected before deployment using interaction-level instrumentation and side-effect probes, rather than observed post hoc?

Why

Component-level tests can pass while the assembled pipeline acquires a failure mode that exists only at the interaction level; without a pre-deployment detection layer, decomposition-based optimization structurally imports emergence as a hidden cost.

Solving a part without seeing the whole generates new, unexpected problems known as emergence.
Open source video
Treasury systems attempt to solve an economic problem and unintentionally create inflation.
Open source video
Source video

Out-of-loop verification for model-conditioned perception

Research question

Does adding an independent, out-of-loop verification channel make a self-consistent but wrong internal model detectable, when the model itself gates what the system observes?

Why

If paradigms and conceptual models shape everything seen, then self-consistency of an agent's world model is not evidence of correctness; a falsification path must be structurally external to the perception-action loop.

Our paradigms and conceptual models shape everything we see.
Open source video
How we see the world determines how we act in it, creating the world we live in which feeds back into our thinking.
Open source video
Source video

Boundary placement as a versioned architectural decision

Research question

What criteria determine where the system boundary is drawn in an agentic stack, and how does moving a tool, memory store, or evaluator across the boundary change emergent failures, observability, and cost?

Why

Boundary choice silently determines what is instrumented, what counts as context versus environment, and therefore which failure modes are even visible; treating it as an implicit framework default makes it unauditable.

A system is a set of elements or parts that are coherently organized and interconnected in a pattern or structure that produces a characteristic behavior.
Open source video
Recognizing boundaries and relationships helps accurately define the scope of any problem.
Open source video
Source video

Structural intervention versus repeated symptom patching

Research question

Do structural changes (topology, feedback path, control flow) outperform repeated prompt/tool patching on long-horizon reliability and symptom-recurrence metrics?

Why

If symptom-level fixes leave the generating structure intact, patch frequency per symptom becomes a diagnostic that the intervention is aimed at the wrong level; this is a recurring maintenance tension in agent stacks.

Instead of just problem solving, we should focus on systems innovation and modifying existing systems.
Open source video
Changing mental models and paradigms leads to changes in how we look at the world.
Open source video
Source video

Two-layer evaluation: decomposition diagnostics plus interaction evaluation

Research question

Can analytical part-level debugging and whole-system interaction evaluation be budgeted as complementary layers without one substituting for the other?

Why

Part-focus localizes faults precisely but is blind to interaction effects; whole-system evaluation catches those effects but cannot localize them, so a single evaluation layer is insufficient for agent stacks.

Analytical thinking zooms into parts to troubleshoot, while systems thinking evaluates the whole and interactions between parts.
Open source video
Engineers often zoom in on isolated parts, leading to unintended consequences.
Open source video
Source video

Article ideas

4

Every Local Fix Buys You a New Failure Mode

Emergence should be treated as an expected cost of decomposition-based optimization, not an anomaly: any architecture optimizing per-component metrics is structurally exposed to component tests passing while the pipeline acquires an interaction-level defect, so each localized optimization needs a whole-system side-effect check before acceptance.

Angle

Reframe emergence from a philosophical systems concept into an engineering budget line item that every decomposition decision must pay for.

Source video

Your Agent's Confidence Is Its Own Reflection

An agent whose observations are mediated by its own context and memory runs a closed perception-action loop in which a wrong internal model can drive actions that make the environment look consistent with that wrong model; without an out-of-loop verification channel, self-consistency becomes indistinguishable from correctness.

Angle

Argue that internal-model self-consistency is not a validity signal and that falsification must be structurally external to the loop.

Source video

Where You Draw the Boundary Decides Which Failures You Can See

Boundary selection should be an explicit, versioned design decision (what is the agent, what is a tool, what is environment) rather than an implicit result of framework defaults, because the boundary determines what is instrumented and therefore which failure modes are even observable.

Angle

Show that boundary placement is an evaluation-power decision disguised as an architecture-style preference.

Source video

Patch Frequency Is a Diagnostic, Not a Workload

Track the number of symptom-level patches per symptom; when the same symptom recurs, that repetition is a signal the correct intervention level is structural (topology, feedback path, boundary), and prompt or tool-output iteration is treating the symptom rather than the cause.

Angle

Turn a maintenance metric into an escalation rule for when to stop editing prompts and change the structure.

Source video

Project ideas

4

Interaction-Level Emergence Probe

beyond-evals

For a multi-component agent pipeline, counterfactual runs that hold one subsystem fixed while varying another will surface failure modes that are absent from every component-level test suite.

Proof of concept

Build a harness that runs the assembled pipeline, then re-runs it with one subsystem frozen at its baseline output; diff the whole-system outcomes and flag divergences that no per-component metric detects.

Measurement

Count of interaction-level failures detected by the probe that component-level tests missed, and the false-positive rate of flagged divergences across repeated runs.

Source video

Out-of-Loop Falsification Channel

gatehouse

Adding an independent verification path that is structurally outside the agent's perception-action loop makes self-consistent-but-wrong internal models detectable, whereas in-loop self-checks do not.

Proof of concept

Instrument the perception layer, define observable proxies for the agent's internal model, and compare detection rates of an in-loop self-check versus an external evaluator on scenarios where the model is wrong but internally consistent.

Measurement

Detection rate of wrong-but-consistent internal models, and the added latency/cost per verification step versus the in-loop baseline.

Source video

Boundary Ablation Bench

new

Moving a tool, memory store, or evaluator inside versus outside the declared system boundary measurably changes which emergent failures are observable and at what observability and cost.

Proof of concept

Define three boundary configurations of the same agent stack, run identical task suites under each, and record emergent failures, instrumentation coverage, and cost per run.

Measurement

Delta in detected emergent failures and instrumentation coverage per boundary configuration, normalized by cost per run.

Source video

Structural vs Patch A/B

movement-lab

Structural changes to agent topology or feedback path reduce symptom recurrence over a long horizon more than repeated prompt/tool patching, at the cost of longer validation cycles.

Proof of concept

Run two maintenance arms on the same recurring failure: one applying prompt/tool-output patches, one modifying topology or feedback structure; track patch frequency per symptom over a fixed number of iterations.

Measurement

Symptom recurrence rate and patch count per symptom over the horizon, plus validation cycle length per intervention.

Source video

Architectural implications

4

The system definition privileges organization and interconnection that produce a characteristic behavior, so the design unit is the boundary plus the relations, not the isolated component.

Before

Design and test each agent, tool, or module against its own local metric.

After

Define the system boundary explicitly and make interconnections (who reads whose output, what feeds back) first-class design artifacts.

Consequence

Evaluation harnesses must measure interaction-level and emergent behavior, not only component accuracy.

Source video

Perception is model-conditioned in the awareness loop, so what a system can observe is partly a function of its own conceptual model.

Before

Treat observations/context as an objective input and evaluate on downstream task accuracy.

After

Instrument the perception layer itself and add an independent, out-of-loop verification path.

Consequence

Self-consistent but wrong internal models become detectable; adds cost and a second evaluator to the pipeline.

Source video

The advocated intervention level is structural modification (systems innovation) rather than symptom relief (problem solving).

Before

Iterate on prompts, tool output shaping, and single-response corrections as the primary change surface.

After

Treat topology, control flow, and feedback structure as the primary change surface, with prompt changes as secondary.

Consequence

Change management must version structure and handle longer validation cycles, since structural changes have wider blast radius than prompt edits.

Source video

Functions (man-made systems) and purposes (human systems) are named as different characteristic behaviors of the same system abstraction.

Before

Encode one objective that is assumed to represent both machine-checkable behavior and stakeholder intent.

After

Maintain declared function and stakeholder purpose as separate, explicitly linked specifications.

Consequence

Optimization targets stay machine-checkable while intent drift becomes visible as a spec mismatch rather than an unexplained production failure.

Source video

Tradeoffs and failure modes

3

Local/component optimization vs whole-system evaluation

Benefit

Narrow focus (bottom line, isolated parts, single metrics) is tractable, fast, and directly measurable.

Cost or risk

The unchecked fix produces new problems in other parts of the system — emergence — which surface only after deployment and outside the optimized metric.

Solving a part without seeing the whole generates new, unexpected problems known as emergence.
Open source video
Source video

Analytical decomposition vs systems view

Benefit

Zooming into parts supports troubleshooting and fault localization in isolated components.

Cost or risk

Part-level focus misses interaction effects and, per the summary, leads to unintended consequences.

Engineers often zoom in on isolated parts, leading to unintended consequences.
Open source video
Source video

Problem solving vs systems innovation

Benefit

Problem solving delivers immediate symptom relief and is easier to scope and ship.

Cost or risk

The underlying system structure is left unchanged, so the conditions that generated the symptom persist; structural change is slower and broader in impact.

Instead of just problem solving, we should focus on systems innovation and modifying existing systems.
Open source video
Source video

Open questions

4

How can emergent, cross-component failure be detected before deployment rather than observed post hoc as an unintended consequence?

Why unresolved

The summary asserts emergence and illustrates it with the treasury/inflation case, but names no detection mechanism, metric, or test procedure.

Research direction

Build interaction-level instrumentation and side-effect probes (counterfactual runs holding one subsystem fixed) that flag effects absent from component-level tests.

Source video

How can the conceptual-model loop be broken or at least measured, given that the model determines what is observed?

Why unresolved

The summary gives only a four-stage diagram (Conceptual Models to What We See to What We Do to The World We Create) and no operational definition or measurement method.

Research direction

Define observable proxies for a system's internal model (e.g., probe tasks, prediction of its own environment changes) and test whether changing the proxy changes action selection.

Source video

What criteria should determine where the system boundary is drawn for an agentic system?

Why unresolved

The summary states that recognizing boundaries and relationships scopes a problem but supplies no criteria for drawing or validating a boundary.

Research direction

Boundary-ablation experiments: move a tool, memory store, or evaluator inside vs outside the boundary and measure effect on emergent failures, observability, and cost.

Source video

Is systems innovation measurably better than symptom-level problem solving over time?

Why unresolved

The recommendation is asserted; the summary reports no comparison, metric, or longitudinal result.

Research direction

A/B structural changes (topology, feedback path) against repeated prompt/tool patching on long-horizon reliability and recurrence-of-symptom metrics.

Source video

Key claims

7
factualVerification needed

A system is a set of elements or parts coherently organized and interconnected in a pattern or structure that produces a characteristic behavior.

Evidence

A system is a set of elements or parts that are coherently organized and interconnected in a pattern or structure that produces a characteristic behavior.

Question

What operational criteria distinguish a set of interacting parts from a 'system' — is coherence, pattern, and characteristic behavior measurable or only stipulated?

Source video
causalVerification needed

Solving a part of a system without seeing the whole generates new, unexpected problems called emergence.

Evidence

Solving a part without seeing the whole generates new, unexpected problems known as emergence.

Question

Is emergence an inevitable consequence of localized fixes, or a probabilistic outcome dependent on coupling density between subsystems?

Source video
causalVerification needed

Treasury systems attempting to solve an economic problem unintentionally created inflation.

Evidence

Treasury systems attempt to solve an economic problem and unintentionally create inflation.

Question

Is there a documented case where a specific treasury intervention was shown to be the causal driver of the resulting inflation, rather than a contributing factor among many?

Source video
causalVerification needed

Our paradigms and conceptual models shape everything we see, and how we see determines how we act, which constructs the world that feeds back into our thinking.

Evidence

How we see the world determines how we act in it, creating the world we live in which feeds back into our thinking.

Question

Can the direction and strength of this perception-action-world loop be measured, or is it an unfalsifiable framing?

Source video
comparativeVerification needed

Analytical thinking zooms into parts to troubleshoot, while systems thinking evaluates the whole and the interactions between parts.

Evidence

Analytical thinking zooms into parts to troubleshoot, while systems thinking evaluates the whole and interactions between parts.

Question

Do the two modes produce measurably different outcomes on the same complex problem, and can they be combined without trade-off?

Source video
opinionVerification not requested

Practitioners should shift from problem solving to systems innovation and modifying existing systems.

Evidence

Instead of just problem solving, we should focus on systems innovation and modifying existing systems.

Source video
causalVerification needed

Recognizing boundaries and relationships helps accurately define the scope of any problem.

Evidence

Recognizing boundaries and relationships helps accurately define the scope of any problem.

Question

Does explicit boundary definition measurably reduce scope errors, or does it merely relocate ambiguity to the boundary decision itself?

Source video

Connections

4