# 08. Put one AI step in a workflow. Turn a fictional request into a short reviewable summary. ## Facilitator preflight Bring laptops, internet, a projector and these local course files. Check tool access before the meetup. Keep samples free of personal information. Quick win: Run a prepared model connection on one fictional sentence. Prerequisites: Module 07. Facilitator has prepared n8n and a working local Ollama model. ### Tools and cost notes - **n8n Basic LLM Chain, Ollama Chat Model and local model credentials.**: Prepared before the meetup.. Local inference uses your hardware. Hosted API alternatives may incur charges. ## Run of show | Slide | Minutes | Activity | |---|---:|---| | 1 | 3 | Put one AI step in a workflow. | | 2 | 3 | Before you start. | | 3 | 5 | Your first small win. | | 4 | 3 | What happens between input and output? | | 5 | 3 | Step 01. Check the prepared model. | | 6 | 3 | Step 02. Create the input item. | | 7 | 3 | Step 03. Add the language-model step. | | 8 | 3 | Step 04. Connect the model. | | 9 | 3 | Step 05. Run and inspect. | | 10 | 3 | Step 06. Compare a second input. | | 11 | 4 | Make the working version. | | 12 | 4 | Ask for a review. Then check it. | | 13 | 6 | Prove your result works. | | 14 | 4 | When it does not work yet. | | 15 | 5 | Choose your next challenge. | | 16 | 3 | Keep the result. Keep the evidence. | | 17 | 2 | Sources you can check. | Total: 60 minutes. ## 1. Put one AI step in a workflow. Outcome: Turn a fictional request into a short reviewable summary. This is a 60-minute in-person laptop session after preflight. No setup-duration guarantee. Core first, extensions second. ## 2. Before you start. Local inference uses your hardware. Hosted API alternatives may incur charges. The facilitator verifies accounts, installed tools and model/host connections before the session. Pair up if setup is incomplete. Do not create accounts or incur costs without understanding the provider terms. ## 3. Your first small win. Quick win: Run a prepared model connection on one fictional sentence. Aim for a visible result in the first 15 minutes after preflight. Explain that generated code/output is untested until the learner checks it. If tools are unavailable, inspect the prepared example with a partner. Copyable prompt: [Open prompt](index.html#08-quick-win) ```text Summarise this fictional request in one sentence. Use only its facts: I want the community room for a reading circle on Saturday. Time and permission are not confirmed. Mark missing facts as unknown. ``` ## 4. What happens between input and output? Ask a learner to explain each stage in their own words. This is a teaching model, not a promise that all tools behave identically. ## 5. Step 01. Check the prepared model. Perform these numbered actions on the laptop. Pause after the step and inspect the result. Keep credentials private and use the prepared workshop instance. Source: [n8n: Basic LLM Chain](https://docs.n8n.io/integrations/builtin/cluster-nodes/root-nodes/n8n-nodes-langchain.chainllm/) Source: [n8n: Ollama credentials](https://docs.n8n.io/integrations/builtin/credentials/ollama/) ## 6. Step 02. Create the input item. Perform these numbered actions on the laptop. Pause after the step and inspect the result. Keep credentials private and use the prepared workshop instance. Source: [n8n: Basic LLM Chain](https://docs.n8n.io/integrations/builtin/cluster-nodes/root-nodes/n8n-nodes-langchain.chainllm/) Source: [n8n: Ollama credentials](https://docs.n8n.io/integrations/builtin/credentials/ollama/) ## 7. Step 03. Add the language-model step. Perform these numbered actions on the laptop. Pause after the step and inspect the result. Keep credentials private and use the prepared workshop instance. Source: [n8n: Basic LLM Chain](https://docs.n8n.io/integrations/builtin/cluster-nodes/root-nodes/n8n-nodes-langchain.chainllm/) Source: [n8n: Ollama credentials](https://docs.n8n.io/integrations/builtin/credentials/ollama/) ## 8. Step 04. Connect the model. Perform these numbered actions on the laptop. Pause after the step and inspect the result. Keep credentials private and use the prepared workshop instance. Source: [n8n: Basic LLM Chain](https://docs.n8n.io/integrations/builtin/cluster-nodes/root-nodes/n8n-nodes-langchain.chainllm/) Source: [n8n: Ollama credentials](https://docs.n8n.io/integrations/builtin/credentials/ollama/) ## 9. Step 05. Run and inspect. Perform these numbered actions on the laptop. Pause after the step and inspect the result. Keep credentials private and use the prepared workshop instance. Source: [n8n: Basic LLM Chain](https://docs.n8n.io/integrations/builtin/cluster-nodes/root-nodes/n8n-nodes-langchain.chainllm/) Source: [n8n: Ollama credentials](https://docs.n8n.io/integrations/builtin/credentials/ollama/) ## 10. Step 06. Compare a second input. Perform these numbered actions on the laptop. Pause after the step and inspect the result. Keep credentials private and use the prepared workshop instance. Source: [n8n: Basic LLM Chain](https://docs.n8n.io/integrations/builtin/cluster-nodes/root-nodes/n8n-nodes-langchain.chainllm/) Source: [n8n: Ollama credentials](https://docs.n8n.io/integrations/builtin/credentials/ollama/) ## 11. Make the working version. Use fictional practice inputs. Ask learners to show their work, not just the final answer. Copyable prompt: [Open prompt](index.html#08-build) ```text Summarise this request in one sentence, then list missing facts. Use only the input, no promises or invented dates. Request: {{$json.request}} ``` ## 12. Ask for a review. Then check it. Use fictional practice inputs. Ask learners to show their work, not just the final answer. Copyable prompt: [Open prompt](index.html#08-review) ```text Compare the summary with the original request. Identify unsupported facts and missing caveats. Give a corrected summary. Original: [paste]. Summary: [paste]. ``` ## 13. Prove your result works. These are manual acceptance tests for the learner to perform. No test is claimed to have run merely because it appears here. Ask pairs to demonstrate one pass and one corrected failure. ## 14. When it does not work yet. Choose the observed symptom. Change one thing and rerun the smallest test. Avoid restarting the whole project or adding tools before diagnosing the current step. ## 15. Choose your next challenge. Experienced learners use the extensions while beginners complete the same core outcome. An extension is optional, not a prerequisite for participating. ## 16. Keep the result. Keep the evidence. Save the working version before leaving. Add a short note: what worked, what needed correction and what you would try next. The next module builds on the saved output. ## 17. Sources you can check. Primary sources checked during course creation on 2026-10-08. Provider interfaces and availability can change. The examples, prompts and activities are original community teaching designs. Source links explain capabilities; they do not prove these fictional exercises were executed. Source: [n8n: Basic LLM Chain](https://docs.n8n.io/integrations/builtin/cluster-nodes/root-nodes/n8n-nodes-langchain.chainllm/) Source: [n8n: Ollama credentials](https://docs.n8n.io/integrations/builtin/credentials/ollama/) Source: [Ollama: Quickstart](https://docs.ollama.com/quickstart) ## Extensions - Use a structured output parser after text works. - Add an explicit approval step before any external action. ## Delivery notes Arrow keys, Space, Home and End navigate slides. N opens notes. Read all slides creates a scrolling lesson. Every prompt has a real Copy prompt button with an honest manual fallback if the browser blocks clipboard access. Keep the whole curriculum folder together: PPTX prompt links open the companion index.html at the same prompt anchor. PowerPoint does not copy directly. For sharing online, rebuild with --companion-base-url set to the published curriculum URL. Motion illustrates input, agent, tool, human review and output. Pause flow stops it; system reduced-motion is respected. PDF contains a static diagram. PPTX uses an animated GIF where available; playback varies by viewer. Original teaching content, synthesized from learning objectives. Public source decks remain separate in library/. No one-to-one rebranding or permission to republish is implied.