Episode 7: Chatbot, workflow, or agent?

1. 1 Chatbot: you ask, it answersIn plain terms: it writes a reply, and the next move is yours.Try: Ask for restaurant ideas. You still do the booking.
2. 2 Workflow: a path set in advanceAnthropic: LLMs and tools orchestrated through predefined code paths. Predictable for well-defined tasks.Try: Every morning: read emails, summarize, save to a folder.
3. 3 Agent: it directs its own stepsAnthropic: LLMs dynamically direct their own processes and tool usage. OpenAI's docs: an AI system with instructions, guardrails and tools to take action on the user's behalf.Try: Goal: a table for four on Friday. It picks the steps and tools.
4. 4 Is a simpler solution enough?Anthropic recommends the simplest solution possible, and notes that agentic systems often trade latency (waiting) and cost for better task performance.General good practice: ask whether one good prompt or a fixed workflow could do this.
5. 5 Can you check the result, and what if it is wrong?Anthropic warns of the potential for compounding errors. In plain terms: one early mistake can carry into later steps.General good practice: start with low-stakes tasks.
6. 6 Set limits and test safely firstAnthropic recommends extensive testing in sandboxed environments, with appropriate guardrails. It also describes agents pausing for human feedback at checkpoints.General good practice: decide in advance what it may not do.

Tap a step when you have done it. Your ticks stay on this device.