From Models to Agents
The first wave of AI was primarily conversational.
A user asked a question. A model generated a response.
As capabilities improved, organisations connected models to documents, databases, tools, workflows, and external systems.
This created progressively more capable environments:
LLMs generate responses.
RAG systems retrieve context before generating responses.
Workflows connect models to repeatable business processes.
Agents reason across multiple steps and use tools.
Agent orchestration coordinates multiple agents towards defined objectives.
Each layer increases capability.
Each layer also increases complexity, dependency, and the importance of context.
The discussion is often framed around what AI can do.
The more important question may be:
Who determines what AI should do?
Capability Is Not Agency
AI capability and human agency are not the same thing.
A model may:
generate answers
analyse information
create code
investigate alerts
produce reports
None of these activities define purpose.
Capability determines what can be done.
Agency determines why it is being done.
This distinction becomes increasingly important as AI systems become more autonomous.
The Economics of Delegation
Many AI investment cases begin with a simple proposition:
increase productivity
reduce labour
improve efficiency
The logic is straightforward.
If a task can be automated, labour costs may fall.
This creates a powerful incentive to replace human activity with AI-assisted workflows and autonomous agents.
The relationship between AI and labour is often presented as a simple replacement equation. In practice, the outcome may be more nuanced. Some tasks become automated. Others become augmented. New responsibilities emerge around governance, orchestration, validation, and oversight. While some organisations have reduced staffing levels in anticipation of AI-driven productivity gains, the long-term balance between labour substitution and labour augmentation remains uncertain.
For some activities, this may be entirely appropriate.
For others, the relationship becomes more complex.
Labour performs tasks.
It also carries:
institutional memory
contextual understanding
informal knowledge
operational judgement
These qualities rarely appear in dashboards, but often become visible when they disappear.
The discussion also extends beyond labour alone.
Historical efficiency gains have not always reduced overall consumption. Improvements in computing, connectivity, and automation frequently increased demand for digital services rather than diminishing it. This phenomenon, often described through Jevons Paradox, raises an interesting question for AI. Greater capability may reduce the cost of individual tasks while simultaneously increasing overall reliance on computation, automation, and digital infrastructure.
The long-term outcome may therefore prove more complex than a simple reduction in labour costs. As capability expands, organisations may also increase their use of AI systems, infrastructure, orchestration layers, and computational resources. The relationship between efficiency, cost, and adoption remains an open question.
The Information Abundance Problem
Modern organisations rarely suffer from information scarcity.
They often suffer from information abundance.
Cyber security provides a useful example.
Organisations may deploy:
Endpoint Security
SIEM Platforms
Threat Intelligence
Vulnerability Management
Identity and Access Management
Cloud Security
Security Monitoring
Each platform produces observations.
Each platform generates alerts.
Each platform expands visibility.
The challenge is not detection.
The challenge is determining which observations matter.
AI can reduce the cost of generating information.
It does not automatically reduce the cost of interpretation.
False Positives and Context
False positives are often presented as a technical problem.
In practice, many are contextual problems.
An AI system may correctly identify:
unusual access
anomalous behaviour
large file transfers
privilege escalation
The observation may be accurate.
The interpretation may still be wrong.
Context determines whether an event represents:
legitimate activity
operational change
planned migration
business process variation
genuine risk
The issue is rarely visibility.
The issue is meaning.
Context helps determine which observations matter. Capability itself, however, depends upon a different set of considerations.
The Infrastructure Behind Capability
AI capability is often discussed as though it exists independently of physical constraints.
In reality, modern AI systems depend upon extensive infrastructure, including data centres, semiconductors, networking capacity, cloud platforms, and energy generation.
As capability expands, so too can the infrastructure required to support it.
Discussions around AI productivity therefore sit alongside broader questions concerning resource allocation, infrastructure investment, and long-term sustainability.
Context Before Automation
As AI systems become more capable, organisations often seek additional automation.
More workflows.
More orchestration.
More agents.
More execution.
This can create confidence theatre.
Outputs become cleaner.
Understanding becomes weaker.
A system may appear more sophisticated while becoming less interpretable.
Stronger decision-support begins elsewhere.
Context comes before automation.
Structure comes before scale.
Observation comes before execution.
Human Agency and Decision Architecture
The discussion is frequently framed as:
Human versus AI.
This is the wrong comparison.
The more useful distinction is:
Agency versus capability.
AI can:
analyse
recommend
investigate
orchestrate
execute
Humans define:
objectives
constraints
acceptable risk
ethical boundaries
success criteria
AI expands capability.
Human agency determines direction.
Capability answers how. Agency determines why.
Work, Adaptation, and Uncertainty
Predictions regarding the future of work often move between extremes. Some anticipate widespread labour displacement. Others expect entirely new categories of employment to emerge. History provides examples supporting both perspectives. Industrialisation, electrification, and computing altered labour markets while simultaneously creating opportunities that were difficult to anticipate beforehand. The long-term effects of AI remain uncertain. What appears more certain is that individuals, organisations, and institutions will continue adapting as capability evolves.
The Triangular Navigation Perspective
This principle extends beyond AI implementation.
The Triangular Navigation Program was designed around:
scenario framing
alternative perspective testing
optional analytical tooling
independent reasoning
The objective is not to remove judgement.
The objective is to strengthen it.
AI becomes more useful when context is preserved.
Not when certainty is assumed.
Decision-support should enhance agency, not replace it.
Distributed Control, Distributed Reasoning
Modern organisations increasingly distribute control across:
cloud providers
SaaS platforms
AI models
orchestration layers
external services
AI introduces a similar shift.
Reasoning becomes increasingly distributed.
Capability becomes increasingly externalised.
This creates new forms of dependency.
The challenge is no longer obtaining answers.
The challenge is preserving accountability, context, and agency across increasingly distributed environments.
AI as Reflection, Not Authority
AI can generate observations.
AI can expand analytical aperture.
AI can challenge assumptions.
AI can surface alternative interpretations.
These capabilities are valuable.
They do not remove uncertainty.
They reorganise it.
AI is not the decision-maker.
It is a reflection surface for structured judgement.
The objective is not to eliminate human judgement, but to improve the quality of the information, context, and reasoning that support it.
Visual Exploration
The following visual summary highlights the relationship between capability, context, and human agency within increasingly automated environments.
Human Agency in an Agentic World
As AI systems become increasingly capable, the conversation will continue to focus on automation.
The more important question may be agency.
Who defines purpose?
Who sets boundaries?
Who determines acceptable risk?
Who remains accountable when conditions change?
Capability may become abundant.
Context may become scarce.
Human agency remains the anchor between the two.
The objective is not to preserve human involvement for its own sake.
It is to preserve the capacity for meaningful judgement within increasingly automated environments.
Better decisions do not emerge from capability alone.
They emerge when capability operates within structure, context, and purpose.
That is where durable decision-making begins.
Related Exploration
Structure Insight. Think in Systems.
Explore Guides, Tools and Insights → blakewiltshire.com