‘Find me a good train connection.’ One system answers with suggestions. Another checks a timetable, compares options and prepares the next step. What changed?
From words to a sequence
In this hypothetical travel example, the system has tools for searching real information and can use their results to decide what to do next. Anthropic’s engineering guide distinguishes predefined workflows from agents whose models dynamically direct their processes and tool use. Terminology varies across the industry. [1]
A system following a fixed recipe can still be useful. An agent becomes valuable when the next step depends on what it discovers: an unavailable train, a changed departure time or an option outside your budget.
The question that matters
A fluent sentence saying ‘your trip is ready’ does not establish that anything was reserved. Look for the actual result: a retrieved timetable, a prepared itinerary or a booking confirmation.
Permission matters too. Searching and purchasing are different actions. A travel assistant should know which it may do and when to ask you. Tool access expands what a system can attempt; it does not make every attempt correct. [1]
What to keep in mind
The travel scenario is illustrative, not a promise about a product. Anthropic is an AI vendor; its guide explains an architectural distinction, not independent proof of performance.
Sources & context
Numbered references link the explanation to its sources. Hypothetical examples and visual metaphors are labelled in the text. Checked 2 October 2026.
Our editorial principles · Suggest a correctionKEEP EXPLORING
From Chatbots to AI AgentsExplore · 3 min read
Could Moon Dust Become a Search for Ancient Technology?Newsroom · 3 min read
When Waste Plastic Becomes a Structural ShapePeek · 1 min readCURIOUS MINDS, SHARED IDEAS
Join the conversation
What surprised you? Share a thought or a question.
Loading reader responses…