Skip to content

Research Papers

In-depth technical analysis to help engineers evaluate frameworks, tools, and practices.


Harness Engineering: How Claude Code and Codex Became Long-Running Agentic-Engineering Systems

June 2026

A survey of "harness engineering" as an emerging discipline: the pre-wired agentic loop, a borrowed nine-component harness anatomy, three distinct separation patterns for long-running work, a CoALA-grounded memory mapping, and the copilot-to-operator shift, with evidence-status labels on every vendor and practitioner claim.

Topics: Harness engineering, agent harness anatomy, separation patterns, CoALA memory mapping, multi-agent orchestration

Read →


When the Loop Never Stops: How Long-Running Agents Broke Seat-Based Pricing and Created the AI Value Problem

June 2026

A causal account of how long-running coding agents mechanically multiplied token consumption, forced the seat-to-token pricing pivot, exhausted enterprise AI budgets, and triggered the late-May 2026 cost reckoning, with every link in the chain traced to a dated, publicly verifiable proof point.

Topics: Token economics, usage-based billing, agentic cost mechanism, enterprise AI budgets, the AI value problem

Read →


Always-On Enterprise Agents: Persistent Architecture, Delegated Identity, and the Productivity Hypothesis

April 2026

A research taxonomy of persistent agent patterns for enterprise deployments, grounded in primary vendor documentation. Covers session-anchored architecture, identity-bound governance, and the cycle-time productivity hypothesis.

Topics: Persistent agents, identity-bound delegation, tool-boundary governance, reference architecture, cycle-time productivity

Read →


Autonomous AI Agents: Execution Loops vs Interactive Assistance

February 2026

Evidence synthesis comparing autonomous AI agent execution loops against interactive human-in-the-loop assistance, covering SWE-bench benchmarks, the METR RCT, industry telemetry, and adjacent domain evidence.

Topics: Autonomous agents, execution loops, SWE-bench, developer productivity, multi-agent systems, scaling laws

Read →


Agentic Development Tools and Execution Architectures

January 2026

Architectural comparison of Claude Code, Goose, Cursor, and GitHub Copilot - how they differ in execution model, context management, and governance.

Topics: Execution architectures, multi-context systems, tool comparison, failure modes

Read →


Spec-Driven Development Framework Patterns

January 2026

Analysis of BMAD, SpecKit, and OpenSpec frameworks - when to use each, how they integrate with existing workflows, and practical adoption guidance.

Topics: Specification-driven development, BDD patterns, contract testing, CI/CD integration

Read →


Companion Articles

Each paper is supported by practitioner articles in the Articles section.

About

These papers are:

  • Practically focused - real-world implementation over theory
  • Based on primary sources - official docs, repos, and hands-on experience
  • Open - free to use, share, and contribute to

Contributing

Found an error or have a suggestion? Open an issue or submit a PR.

Updated at:

Released under the MIT License.