Workflow Automation

Stems From:

Overview:

Automating repetitive workflows is more accessible than ever. Deterministic tasks can be automated with code quickly written with Agentic Coding tools and non-deterministic tasks can leverage Large Language Models (LLMs) to bring judgement into the workflow.

Teams should define their Ways of working so that automations work for the team rather than becoming another tool the team has to adopt. A successful workflow automation project introduces zero friction, it aligns with the existing ways of working and should not make users feel like they're learning a new tool

We should automate workflows to achieve at least one of three goals: Better, Faster, and/or Cheaper

A few ways of prioritizing automation projects:

What can we make 10x faster or cheaper
What processes can be made deterministic and don't require AI judgement?
What workflows require judgement and we have enough risk tolerance to accept errors in judgement?

Decision Points:

Where do you bring humans into the loop?
Where do you ask an LLM to use it's judgement?
Where do you create deterministic processes (e.g. code)

Key Learnings:

Getting agents to produce consistent responses is inherently difficult
When custom automations are better

Sub-Topics:

Thinking in Systems