CHAPTERS
- 0:00 – 1:55
Why n8n stands out: workflows + AI agents in one tool
Aakash introduces Pawel Huryn and frames n8n as a uniquely powerful automation platform that supports both classic workflow automation and modern AI-agent systems. Pawel sets expectations: you can automate a huge range of business tasks without heavy coding.
- 1:55 – 3:14
Competitor monitoring workflow: trigger types and pulling competitors from Google Sheets
They begin building a real workflow to send a weekly competitor update email. Pawel explains triggers (manual, schedule, webhook) and shows how to read competitor names from a Google Sheet.
- 3:14 – 8:44
Development speed hack: pinning node data to avoid repeated API calls
A key n8n tactic is introduced: pin output data so you can iterate on downstream steps without re-querying external services. This reduces cost and speeds up workflow development.
- 8:44 – 12:09
Perplexity search per competitor: prompt design and low-cost API usage
The workflow calls Perplexity for each competitor to gather recent market intel. Pawel demonstrates variable injection from each sheet row and explains cost considerations (API key + pay-per-use).
- 12:09 – 13:53
Compressing context with a Code node to cut tokens and improve reliability
Perplexity returns a lot of extra metadata that would waste tokens if forwarded to an LLM. Pawel uses a Code node to keep only the essentials (content + citations), describing this as “context compression.”
- 13:53 – 19:57
No manual loops needed: n8n’s item-based execution model
Pawel clarifies a core n8n concept: nodes automatically run once per input item, so you often don’t build explicit loops. They then prepare to unify multiple competitor outputs into a single object for reporting.
- 19:57 – 23:13
Generate the final report with OpenAI: aggregation, formatting rules, and JSON conversion
They aggregate the per-competitor results into one payload and prompt OpenAI to produce a clean competitor monitoring report. Pawel highlights prompt structure, link formatting rules, and the crucial JSON-to-string conversion for structured data.
- 23:13 – 31:50
Markdown-to-HTML conversion and sending the Gmail report end-to-end
To avoid spending tokens generating HTML, they generate Markdown and then convert it to HTML using a dedicated node. The final step sends the report via Gmail as an HTML email, validating the workflow end-to-end.
- 31:50 – 34:11
Workflow vs agentic workflow: rebuilding with an AI Agent node (trade-offs)
They rebuild the same competitor workflow using an AI Agent node with tools (Google Sheets + Perplexity). This increases flexibility but costs more tokens and time; logs show tool calls and parallelism behavior.
- 34:11 – 40:36
“True agent” version: minimal instructions, full tool autonomy, big token cost
They push to a higher-agency approach: give the agent the goal and the tools (including Gmail), and let it decide the steps and formatting. Output quality improves, but token usage explodes compared to earlier designs.
- 40:36 – 45:35
n8n best practices for production: error workflows, retries, tool descriptions, max iterations
Pawel shares operational practices for making automations reliable in production. Emphasis is on handling failures, avoiding premature agent stops, and improving tool-call success with better descriptions.
- 45:35 – 49:04
Multi-agent research system: orchestrator + sub-agents + report generation pipeline
They showcase a complex multi-agent research architecture inspired by Anthropic’s patterns. A lead agent decomposes tasks, sub-agents research in parallel (search + scrape + compression), and a copywriter agent produces a final report stored in Google Drive.
- 49:04 – 51:12
PM automation use cases: inbox triage, PRDs, Slack/Drive search, data imports
They translate capabilities into everyday product and ops work. Examples include email summarization/drafting, competitor research, PRD generation with internal context, and routine data syncing for products.
- 51:12 – 57:57
Free plan hacks & self-hosting: history retention, global variables via tables, workflow version backups
Pawel explains practical ways to overcome free-tier limitations: short execution history, lack of global variables, and limited version history. He recommends self-hosting for low cost and demonstrates using n8n to back up n8n workflows to Google Drive.
- 57:57 – 58:53
Learning AIPM skills with n8n: prompting, context engineering, RAG, evals, and intuition
They close by positioning n8n as a practical learning environment for AI product skills. By building real automations, you internalize prompting, context management, RAG building blocks, and evaluation/guardrail concepts—without heavy coding.
Get more out of YouTube videos.
High quality summaries for YouTube videos. Accurate transcripts to search & find moments. Powered by ChatGPT & Claude AI.
Add to Chrome