Scaling Organizational AI Learning Through Public Workflows
Nate B Jonesgo watch the original →
the gist
Companies fail to scale AI because individual usage remains private, creating an 'apprenticeship gap.' The fix is forcing AI interactions into public, declared channels where senior staff model their judgment, context-loading, and revision processes for the rest of the team.
The Apprenticeship Gap
Most organizations suffer from a hidden AI problem: employees use AI tools in private windows, meaning individual gains in productivity do not translate into organizational knowledge. This creates an 'apprenticeship gap' where junior employees never observe how senior operators frame problems, load context, or refine outputs. When senior staff perform critical tasks like critiquing roadmaps or investigating bugs in private, the team loses the opportunity to develop shared taste and judgment.
Implementing Public AI Workflows
To solve this, organizations should adopt 'declared spaces' where AI interactions are visible to the team. Shopify’s internal agent, River, serves as a model by requiring all interactions to occur in public Slack channels rather than private direct messages. This design choice allows other engineers to audit the entire lifecycle of an AI-assisted task, including the initial prompt, the context provided, the human's pushback, and the final review.
To make this work, teams should focus on making four specific parts of the AI workflow visible:
- The Task: Clearly define what the human is trying to achieve.
- The Context: Document the specific data, constraints, and background information fed to the model.
- The Interaction: Show the iterative prompting process, including how the human pushes back on initial model outputs.
- The Review: Explicitly state what the human accepted, rejected, or manually corrected, and why.
Managing Constraints and Privacy
Public AI work does not require dumping raw logs into a channel. Instead, teams should establish 'declared channels' with pinned guidelines that define what is appropriate for public view. Regulated teams can maintain compliance by using anonymized data or stripping PII before running workflows in public channels. The goal is to create a safe surface for learning while maintaining necessary security boundaries. By mandating that agents only operate in public channels, leadership uses a binding constraint to force collaboration and prevent the duplication of effort that occurs when individuals solve the same problems in isolation.