Agentic Script
A natural language instruction + tools = an autonomous workflow that adapts.
How it works
Instruction
A natural language prompt that describes the goal, constraints, and expected output format.
Tools
The agent has access to a set of tools: APIs, scraping, browser, database, files.
AI Agent
The LLM orchestrates tool calls, handles errors, adapts to edge cases.
Output
Structured data, CRM export, Slack notifications, reports — in the defined format.
Classic Script vs Agentic Script
Classic script
- Breaks if data format changes
- No handling of ambiguous cases
- One bug = complete stop
- Heavy maintenance on every API change
Agentic script
- Adapts to format variations
- Understands context, makes decisions
- Handles errors, retries with alternatives
- The instruction stays stable even if the API changes
Script examples
Prospecting
Find 50 B2B SaaS companies in France, 10-50 employees, fundraised in 2024. For each: SIREN, director, email, CEO LinkedIn. Export to Google Sheets.
Tools: SIRENE, Pappers, LinkedIn, Google Sheets
Onboarding
For each new signup, verify the SIREN, extract legal info, score the risk. If OK, create the record in CRM and send the welcome email.
Tools: SIRENE, Pappers, CRM API, Email
Monitoring
Monitor the 15 competitors listed in Notion. Every morning, check new hires, fundraising rounds, press articles. Summarize in Slack #monitoring.
Tools: Notion, LinkedIn, Google News, Slack