Building Your AI Work Factory
Atlassian introduces the concept of a 'work factory' to standardize AI-assisted knowledge work, replacing ad-hoc AI usage with structured, repeatable processes that encode expertise and institutional memory.
Software engineering’s 'software factory' model—where structured pipelines turn raw inputs into reliable outputs—is being adapted for knowledge work. Atlassian proposes a 'work factory' that combines standard tools, curated configurations, and encoded expertise to process inputs like customer data or market signals into consistent, high-quality outputs. This approach aims to reduce repetitive manual tasks such as decision-making checklists or report generation, which currently rely on human cognitive effort and are prone to inconsistency.
The work factory divides labor between humans and AI: humans define strategy, frameworks, and judgment rules, while AI handles execution tasks like data retrieval, synthesis, and formatting. Modeled after software design patterns, these factories provide a shared vocabulary and structure, enabling teams to review and improve processes collaboratively. Unlike ad-hoc AI interactions, a work factory accumulates institutional memory through run logs, research summaries, and decisions, refining its knowledge base over time by querying connected systems like Jira, Confluence, or Salesforce via AI tools using the Model Context Protocol (MCP).
A key feature of work factories is selective data access, where AI queries only relevant systems to avoid noise and confabulations. Knowledge files—either run-scoped or evergreen—serve as the AI’s grounding material, while directories like frameworks/, templates/, and workflows/ store structured instructions. The AI can also write to these files, creating a growing repository of institutional knowledge that captures past findings, changes, and reusable context, ensuring continuity across tasks.
Work factories differ from ad-hoc AI setups by enforcing intentional scoping and systematic feedback loops. Skills—structured instruction documents—define precise AI tasks, such as assessing opportunities by retrieving specific data and applying scoring frameworks. This structure ensures consistent outputs and simplifies troubleshooting, as issues like incorrect scoring or output formats can be traced to specific components for targeted updates. The result is a repeatable, improvable system that standardizes expertise and reduces maintenance overhead.