A agent<\/strong> is an authorized assistant to act<\/strong>, and not just to respond. Prefix your message with The agent takes control until the task is completed, then returns control to the A workspace set up in mode Automatic<\/strong> entrusts each message to a A planned task<\/strong> executes an agent according to a schedule, without anyone The progress is visible in the conversation: you can see which tools were used. A maximum number of tool calls<\/strong> limits loops: a blocked agent That's the main factor. Operating tools is more demanding than conversing. Each space can therefore assign an agent different model<\/strong> of that of An agent follows instructions. Vague instructions produce vague results.<\/p>\n Ineffective:<\/strong> \u00ab"Take care of competitive intelligence."\u00bb<\/p>\n Effective :<\/strong> \u00ab"Consult the Pricing pages of the following three sites, note the An agent can only do what their activated skills allow. Depending on the skills activated:<\/p>\n
\nWhere an ordinary conversation produces text, an agent strings together actions
\nconcrete examples: searching the web, reading a file, querying a database,
\nprepare an email, produce a graph.<\/p>\n
\nTrigger an agent<\/h2>\n
In a conversation<\/h3>\n
@agent<\/code> :<\/p>\n@agent Research the latest regulatory developments on electronic invoicing in Switzerland, then summarize them in ten lines.\n<\/code><\/pre>\n
\nnormal conversation.<\/p>\nPermanently<\/h3>\n
\nagent. Reserved for spaces dedicated to automation: this mode is slower
\nand less predictable.
\n\u2192 Conversations<\/a><\/p>\nWithout human intervention<\/h3>\n
\neither in front of the screen. \u2192 Planned tasks<\/a><\/p>\n
\nHow an agent works<\/h2>\n
\n
\n the next step.<\/li>\n
\ncalled and what they returned. A failing agent says so, rather than
\nto invent a result.<\/p>\n
\nstops instead of spinning indefinitely.<\/p>\n
\nWhat determines the quality of an agent<\/h2>\n
The model<\/h3>\n
\nYou need to choose the right tool, build valid arguments, and interpret
\na sometimes unexpected result. A model that responds very well to
\nquestions can fail as an agent.<\/p>\n
\nconversations. If agents matter to you, give them the best model.
\ncapable of what you have at your disposal.
\n\u2192 Choosing your models<\/a><\/p>\nThe formulation of the request<\/h3>\n
\nprices displayed for the entry-level offer, and present them in a table
\ncomparison with the date of consultation.\u00bb<\/p>\nAvailable skills<\/h3>\n
\nA request that assumes a tool is missing fails \u2014 and the agent has to say so.
\n\u2192 SKILLS<\/a><\/p>\n
\nWhat an agent can do<\/h2>\n