00 · IN THREE MINUTES

The answer in three steps

  1. 1An agent has a goal, an environment and actions that can change that environment.
  2. 2Memory, planning and tool use extend a language model but do not automatically create autonomy.
  3. 3More autonomy increases both capability and the need for permissions, monitoring and recovery.

01 · CONVERSATION IS NOT ENOUGH

Conversation is not enough

A conversational interface can call tools, and an agent can communicate without chat. The defining shift is from producing one message to repeatedly choosing actions in response to changing state.

02 · THE LOOP CARRIES STATE FORWARD

The loop carries state forward

A typical system observes, proposes a step, invokes a tool, receives a result and updates its plan. Memory may store intermediate facts, but its accuracy, scope and lifetime must be designed.

FIG. 02How the system changes state
A conceptual mechanism map. Geometry, scale and timing are explanatory unless labelled otherwise.

03 · TOOLS MAKE ERRORS CONSEQUENTIAL

Tools make errors consequential

A wrong sentence is different from a wrong database write, purchase or code deployment. Agent design therefore needs least-privilege access, explicit approval boundaries and logs that show which action came from which evidence.

04 · PLANNING DOES NOT ENSURE JUDGMENT

Planning does not ensure judgment

Breaking a goal into tasks can improve reliability, yet plans may drift, repeat or optimize an incomplete objective. Tests must cover long trajectories and failure recovery, not only whether the final answer looks good.

05 · AUTONOMY IS A SPECTRUM

Autonomy is a spectrum

Systems range from a single approved tool call to long-running workflows. Calling all of them “agents” hides important differences, so a useful description states available actions, duration, supervision and stopping conditions.

06 · SOURCES AND EVIDENCE

Sources and evidence

Claims are linked to foundational papers, standards or the primary study behind the update.

  1. 01
    The Rise and Potential of Large Language Model Based Agents: A Survey

    Supports a defined mechanism, measurement or evidence boundary in this article.

    SCHOLARLY REVIEW ↗
  2. 02
    Intelligent agents: Theory and practice

    Supports a defined mechanism, measurement or evidence boundary in this article.

    SCHOLARLY REVIEW ↗
CHANGE LOG29 Aug 2026 · First five-language edition; mechanisms, limits, diagrams and sources checked.