Medium: How product managers can use AI agents to automate everyday work without spending anything
Most guidance on AI for product managers assumes commercial tools and paid APIs. This March 2026 article takes a different approach: building working automation using only open-source or free-tier software, with no vendor lock-in. It is written for PMs who want to experiment with agent-based workflows but are not in a position to commit budget or who want to understand what is technically possible before evaluating commercial options.
What it covers
The author identifies five PM tasks worth automating first: converting meeting notes into structured summaries with decisions and action items, grouping customer feedback by theme, drafting PRDs from rough notes, generating stakeholder updates, and digesting competitive research. These are described as “operational drag” — text-heavy, repetitive, and low-risk enough that automated drafts with human review are a practical workflow rather than a risk.
The recommended tool stack for building these automations:
- n8n as the self-hosted automation backbone
- Flowise for designing LLM workflows with built-in human-in-the-loop checkpoints
- Ollama to run language models locally via API, keeping sensitive data off commercial infrastructure
- Qdrant for retrieval and semantic search across documents
A detailed example walks through a meeting notes agent that ingests transcripts, classifies comments by theme, extracts action items, and exports into document templates. This workflow is recommended as the first thing to build because it generates measurable time savings immediately without touching any decision-making processes.
The core principles
Two principles run through the article. First, start with one specific, annoying workflow rather than designing a system upfront — early automation work produces returns faster and teaches more than planning does. Second, always include a human review step before any output is shared or committed, particularly for PRDs and stakeholder communications. The measure of success is practical: time saved per task and the percentage of drafts that require significant editing, not technical sophistication.
Who it is useful for
PMs at early-stage companies or teams without established AI tooling budgets who want to begin automating before getting formal approval. Also useful for those evaluating commercial solutions, since building a working version first clarifies exactly what the commercial tool needs to do and what it costs to replicate with free alternatives.