Skip to specification

Who drew this, and from what.

This site is one person’s attempt to explain how agent harnesses are architected, built on talks given by other people. Both halves of that sentence are on this sheet.

Drawn by

Sunny Luthra

Sunny Luthra

Foundermrova.rocks·ohmydog.rocks

I built this sheet to keep what I have read about agent harnesses in one place — drawn rather than summarised, so the gaps stay visible.

Contact

If a layer of your own machine has run out of road and you want to talk about it, any of these reaches me.

Built from — 23 sources by other people

Every pillar on this site is a reading of someone else’s talk. The structure, the examples, and every external figure quoted anywhere here come from these, and each figure is attributed on the sheet to the person who gave it — no number is carried from one talk to a sheet credited to another. The synthesis, the diagrams, and any mistakes are mine.

Most feed a whole pillar. Where a talk goes deep on a single primitive it is credited on that sheet instead, and where one talk is large enough to raise a run of sheets it is credited on all of them — marked below with the sheets it stands under.

Sources read since — reports and talks that are examined here rather than built into a pillar — are kept in the reading room.

  1. The HarnessHarness Engineering MasterclassThe Carbon Layer
  2. The HarnessContext Management MasterclassThe Carbon LayerCredited on sheet 03 Context Management
  3. The HarnessSelf-Improving AI Agents: Evolving the Harness, Not the ModelThe Carbon LayerCredited on sheet 11 Evolution
  4. The HarnessWhy Agentic Systems Need OntologiesFrank Coyle, UC BerkeleyCredited on sheet 10 Verification & Observability
  5. The HarnessWhy This Is the Most Exciting Time to Be HumanKen Ono, Axiom MathCredited on sheet 10 Verification & Observability
  6. The HarnessAI Will Create New Wealth, But Not Where You ThinkPo-Shen Loh, Carnegie Mellon UniversityCredited on sheet 10 Verification & Observability
  7. The HarnessHow to Build a Software Factory for AI Coding AgentsDex Horthy, HumanLayer, and Vaibhav Gupta, BoundaryCredited on sheet 05 Execution Environment
  8. The HarnessPractical Loop EngineeringAddy OsmaniCredited on sheet 07 Orchestration
  9. The HarnessIntroducing System One Models & JevDiogo Almeida, TypeSafe AICredited on sheet 07 Orchestration
  10. The HarnessBuilding a Harness with JevSydney Runkle and Hunter Lovell, LangChainCredited on sheet 07 Orchestration
  11. The HarnessGraph Engineering ExplainedAishwarya SrinivasanCredited on sheet 08 Sub-agents
  12. Test BenchHarness-Bench: Measuring Harness Effects across Models in Realistic Agent WorkflowsYilun Yao et al.Credited on the Test Bench — the paper that made harness sensitivity a published finding
  13. MemoryAI Agent Memory MasterclassThe Carbon Layer
  14. MemoryMemory Harnesses for Long-Running Research AgentsStefania Druga, Sakana AICredited on sheet 05 Context Assembly
  15. Agentic SDLCAgentic SDLC ExplainedPavan Belagatti
  16. Agentic SDLCAI Can Code. That's Why You Should LearnChris Piech, StanfordCredited on sheet 04 Who Owns What
  17. Agentic SDLCHow to Build a Software Factory for AI Coding AgentsDex Horthy, HumanLayer, and Vaibhav Gupta, BoundaryCredited on sheet 05 The Agent Factory Floor
  18. Agentic SDLCIntroducing AI Driven Development Lifecycle (AI-DLC)Anupam Mishra and Raja SP, AWSCredited on sheets 01 From Waterfall to Agents, 02 Beyond Code Generation, 03 The ADLC Loop, 04 Who Owns What, 05 The Agent Factory Floor, 06 The Unified Context Layer
  19. Agentic SDLCIt Ain't Broke: Why Software Fundamentals Matter More Than EverMatt Pocock, AI HeroCredited on sheet 07 Code Is Not Cheap
  20. Forward DeployedForward Deployed Engineering 101Kevin Bai, Anthropic
  21. Forward DeployedTrust. Product. Impact.Colin Jarvis, OpenAI, with Apoorv Agrawal, AltimeterCredited on sheets 02 Selling the Outcome, 03 The Half-Million Gap, 04 The Platform of Primitives
  22. Forward DeployedForward Deployed Engineering at CursorPauline Brunet, CursorCredited on sheets 01 The 2×2, 02 Selling the Outcome, 04 The Platform of Primitives, 05 The Two Questions
  23. Forward DeployedAI tools for Forward Deployed EngineeringVasuman Moza, Varick AgentsCredited on sheets 02 Selling the Outcome, 05 The Two Questions, 06 When Every Platform Went Agentic

How this sheet was drawn

There are no case studies, no client names, no testimonials, and no usage numbers on this site, because none exist. The build is the argument: 30 deep-dive sheets and four maps you can open, pan, and inspect. If it reads as careful, that is the only claim being made.

SheetAbout
StackNext.js · React Flow · MDX
SetLiterata · Azeret Mono
Sources23 sources