← @bgSponge

how to build agentic systems for knowledge work

Summary

Heinrich argues that knowledge work engineering requires building domain-specific harnesses that bundle knowledge, tools, and methods into reusable agent repositories.

  1. Knowledge systems require domain-specific harnesses.

    Instead of generic tools, users must build software around agents that provides specific context, tools, and methods for particular professional workflows.

    become a domain-specific harness

  2. Agent repos bundle knowledge with capabilities.

    An agent repo contains typed markdown for domain objects, agent skills for extraction, and interfaces like citation graphs to explore material.

    an agent repo can contain: - knowledge in typed markdown

  3. Knowledge should behave like a codebase.

    Systems should implement custom type systems, diagnostics for broken references, package imports, and version history to manage evolving thinking.

    we want similar patterns for knowledge

  4. Type engines enable cross-repo reasoning.

    A type engine resolves wikilinks across repository boundaries, allowing a project to import a shared methods library while keeping private interpretations.

    our type engine lets wikilinks cross those repo boundaries

  5. Agents automate method application.

    By making workflows explicit through types and skills, agents can execute research protocols, check claims against sources, and surface missing information.

    organizing domain knowledge and working methods in a similar way

Highlights

  • agents can help you make a working method explicit and write the software needed to apply it

  • i want to make the case for replacing them with custom apps you and your agent build on top of your own data, all within

  • a researcher needs to follow citations and compare the evidence behind claims

Read the original ↗Save to your Sponge

Summarized with Sponge. Quotes are the author’s own words.

More from @bg

All readings →