Edition 2026.08 · Reviewed 2026-08-09
Agentic Engineering Source Library
A curated primary-source ledger covering research, engineering practice, guidance, standards, protocols, and thought leadership.
OpenAI — Harness engineering: leveraging Codex in an agent-first world
A production case study in designing repositories, tools, feedback loops, and environments for agent-first delivery.
OpenAI — An open-source spec for Codex orchestration: Symphony
An orchestration specification that connects project work, isolated agent execution, and reviewable delivery state.
OpenAI — How we monitor internal coding agents for misalignment
A current account of monitoring powerful coding agents that can act inside consequential development environments.
OpenAI — Building self-improving tax agents with Codex
A feedback-loop pattern that turns practitioner corrections and production traces into evaluations and scoped engineering work.
Anthropic — Effective context engineering for AI agents
A working model for treating model-visible context as a finite resource that must be selected and maintained.
Anthropic — Effective harnesses for long-running agents
Patterns for incremental progress and durable handoffs across context windows and extended execution.
Anthropic — Harness design for long-running application development
A planner, generator, and evaluator architecture for multi-hour autonomous application development.
Anthropic — Scaling Managed Agents: Decoupling the brain from the hands
An architecture separating durable sessions and environment state from changing models and harnesses.
Anthropic — Demystifying evals for AI agents
A practical treatment of evaluating trajectories, outcomes, graders, tasks, and agent-environment interaction.
Anthropic — Quantifying infrastructure noise in agentic coding evals
An empirical demonstration that resource configuration can materially alter agent benchmark results and must be treated as part of the evaluation system.
GitHub — Spec-driven development with AI
A structured specification-to-plan-to-task workflow for making intent durable and actionable by coding agents.
GitHub Spec Kit — Spec persistence models
Distinguishes spec-first, spec-anchored, and spec-as-source approaches without asserting one universal model.
NIST NCCoE — Identity and Authority for Software Agents concept paper
A concept paper exploring how established identity and authorization practices can apply to software and AI agents.
NIST — AI Agent Standards Initiative
A standards initiative addressing secure, interoperable adoption of AI agents.
Microsoft Entra — What are agent identities?
A current enterprise identity model distinguishing agents from people and conventional applications, including ownership, sponsorship, delegated rights, and auditability.
NIST — Towards best practices for automated benchmark evaluations
Guidance on the integrity and limitations of automated benchmark evaluation practices.
NIST CAISI — Challenges in monitoring deployed AI systems
A current view of monitoring needs and limitations for deployed AI systems.
OWASP GenAI Security Project — OWASP Top 10 for Agentic Applications 2026
A threat-oriented reference for goal hijacking, tool misuse, identity abuse, memory poisoning, and cascading agent failures.
OWASP GenAI Security Project — Memory Is a Feature. It Is Also an Attack Surface
A focused account of why persistent agent memory requires integrity, provenance, and poisoning defenses.
Model Context Protocol — Enterprise-Managed Authorization for MCP
An emerging authorization pattern for enterprise-managed agent access to MCP servers.
Model Context Protocol — Model Context Protocol 2026-07-28 release candidate
A current protocol release candidate illustrating that agent-tool interoperability, authorization, and extension surfaces continue to evolve and require explicit version governance.
A2A Project — Agent2Agent Protocol v1.0
A protocol milestone for interoperable communication and collaboration between agents.
Google Cloud — Agent Sandbox on GKE and Agent Substrate
An agent-specific execution architecture emphasizing isolation, state, scheduling, and environment control.
OpenTelemetry — GenAI observability
Developing semantic conventions for model and agent telemetry across traces, metrics, and events.
SLSA — SLSA provenance specification v1.2
An established software-supply-chain foundation for attributable build provenance.
Andrej Karpathy — Software Is Changing (Again)
A thought-leadership lens on natural language, neural networks, and changing software interfaces—not a formal standard.
Google Research / DORA — DORA 2025 State of AI-assisted Software Development Report
Large-sample research framing AI as an amplifier of the capabilities and dysfunctions already present in an engineering organization.
Microsoft Agentic Center of Excellence — Define roles, responsibilities, and decision rights
An operating model that assigns one accountable role per decision and distinguishes domain, platform, risk, and runtime responsibilities.
Microsoft Agentic Center of Excellence — Govern agents by risk
Current enterprise guidance for matching review depth, autonomy limits, release gates, logging, and incident planning to agent risk.
Microsoft Agentic Center of Excellence — Manage the agent lifecycle
A product-lifecycle model spanning intake, deployment, monitoring, improvement, ownership, evaluation, and deliberate retirement.
Microsoft Entra ID Governance — Governing Agent Identities
An identity-governance model for agent sponsors, access packages, ownership continuity, lifecycle workflows, and enable or disable decisions.
Microsoft Entra Agent ID — Administrative relationships in Microsoft Entra Agent ID
Separates technical ownership from business sponsorship and assigns sponsors lifecycle, access-review, and incident-response decisions.
OWASP GenAI Security Project — State of Agentic AI Security and Governance 2.01
A current threat and governance synthesis covering instruction-data confusion, containment gaps, autonomous action, and agentic incident readiness.
Open Policy Agent — OPA Management APIs and Architecture
An established policy-engine architecture for distributed enforcement, centrally managed policy bundles, decision logs, and enforcement telemetry.
NIST — Artificial Intelligence Risk Management Framework 1.0
A consensus-based foundation for governing, mapping, measuring, and managing AI risk according to context, impact, and organizational priorities.
Anthropic — Building effective agents
A foundational distinction between fixed, code-orchestrated workflows and systems in which a model dynamically directs its process and tool use.
Artificial Analysis — Artificial Analysis Intelligence Benchmarking Methodology
A transparent methodology for comparing model quality, price, latency, throughput, and cost per task. Its composite indices are useful evidence, not a substitute for use-case-specific evaluation.
Matt Dailey · AI Engineer World's Fair — Velocity Sickness: What Happens When Your Whole Team Gets 10x Faster
A practitioner framing for the coordination pain that appears when individual generation accelerates faster than shared decisions, review, and integration. It is an attributed field term, not a clinical or validated organizational diagnosis.
DORA — Work in process limits
Research-backed guidance for making work visible, limiting concurrent work to real capacity, and improving the most consequential constraint in the delivery system.
DORA — Working in small batches
Guidance that connects smaller changes with faster feedback, easier review, and safer integration, including in AI-assisted delivery.
Microsoft Research / ACM Queue — The SPACE of Developer Productivity: There's more to it than you think
A multidimensional productivity framework spanning satisfaction, performance, activity, communication, and efficiency rather than reducing engineering value to one activity metric.
GitHub — Agent pull requests are everywhere. Here's how to review them
A current practitioner account of agent-generated pull requests saturating review bandwidth and the continued need for contextual human judgment.
METR — Measuring the Impact of Early-2025 AI on Experienced Open-Source Developer Productivity
A randomized study in one narrow setting that found experienced contributors took longer with early-2025 tools. The authors explicitly caution against generalizing the result to all developers, tools, repositories, or later model generations.
Anthropic — Trustworthy agents in practice
Describes a self-directed plan–act–observe–adjust loop and the engineering layers needed to make longer-horizon agents more trustworthy.
OpenAI — A practical guide to building agents
Defines a practical agent around a model, tools, instructions, and a run loop that continues until an exit condition is reached.
Peter Yang — 5 Rules for Building AI Agents That Work in Production | Nan Yu & Jacob Shumway
A Linear engineering discussion of production agents, tool design, context loading, feedback, evaluation, and simple model-in-a-loop explanations.
DeepLearning.AI — Learn to build effective Agentic AI systems
A short introduction to an Andrew Ng course covering reflection, tool use, planning, multi-agent patterns, and evaluation.
DeepLearning.AI — Agentic AI
A practitioner course organizing agentic systems around reflection, tool use, planning, multi-agent workflows, evaluation, and optimization.
IndyDevDan — My Super Simple Software Factory (For Agentic Engineers)
A practitioner demonstration of an agentic software-factory workflow built from reusable agents, commands, hooks, and feedback loops.
AI Engineer — Harness Engineering: How to Build Software When Humans Steer, Agents Execute — Ryan Lopopolo, OpenAI
A conference presentation about engineering repositories, feedback loops, tools, and constraints around coding agents.
arXiv / NeurIPS — Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks
Introduces retrieval-augmented generation as a combination of learned generation and explicit retrieved non-parametric memory.
arXiv / ICLR — ReAct: Synergizing Reasoning and Acting in Language Models
Studies interleaved reasoning traces and environment actions, a foundational pattern for model-driven agent loops.
arXiv / NeurIPS — Reflexion: Language Agents with Verbal Reinforcement Learning
Introduces a feedback pattern in which an agent converts signals into verbal reflections retained for later attempts.
arXiv — Judging the Judges: A Systematic Study of Position Bias in LLM-as-a-Judge
Documents position-related bias in model-based comparative judgment and motivates calibration and bias controls.
Google Research / ICLR — Outrageously Large Neural Networks: The Sparsely-Gated Mixture-of-Experts Layer
Introduces a sparsely gated neural layer that routes each example to a subset of learned expert subnetworks.
Model Context Protocol — Model Context Protocol architecture overview
Documents the host, client, server, capability-negotiation, and message layers used to connect AI applications with context and tools.
NIST — Zero Trust Architecture (NIST SP 800-207)
Defines zero trust around explicit, continuously evaluated access decisions rather than implicit trust from network location.
Addison-Wesley, by special arrangement with Pragmatic Bookshelf — The Pragmatic Programmer, 20th Anniversary Edition
The official title record for Thomas and Hunt’s revised edition identifies the authors, publication date, and canonical Tracer Bullets topic, which uses real feedback from a thin implementation path to guide development toward an evolving target.
AI Engineer — "Software Fundamentals Matter More Than Ever" — Matt Pocock
A conference talk arguing that shared language, deliberate software design, deep modules, and fast test feedback become more important as coding agents increase implementation throughput.
Matt Pocock — mattpocock/skills: A complete AI Coding workflow, end-to-end
Demonstrates a skills-based coding workflow that develops shared domain context, writes a specification, decomposes work into tickets, implements it, and reviews against the specification and standards.
Matt Pocock — /wayfinder: Nothing is too big to plan anymore
Demonstrates planning a large body of work through an explicit destination, known dependencies, uncertainty, a current frontier, and decision tickets that can be resolved over time.
AI Hero — Claude Code for Real Engineers
The official course page frames AI-assisted engineering around planning, decomposition, steering, feedback loops, tracer bullets, sandboxing, and maintaining navigable production codebases.
Matt Pocock — Skills for Real Engineers
An open repository of composable, user-invoked agent skills covering discovery, specifications, tickets, implementation, review, testing, debugging, teaching, and codebase design.
AI Hero — The /teach Skill
Documents a stateful, source-led learning workflow organized around a learner mission, a persistent learning record, primary-source research, short lessons, retrieval practice, references, and a glossary.
GitHub — GitHub Spec Kit
Documents an intent-driven workflow that turns a specification into a plan, tasks, and implementation artifacts while supporting multiple coding agents and replaceable process extensions.
Anthropic — Agentic coding and persistent returns to expertise
Reports a privacy-preserving observational analysis of roughly 400,000 Claude Code sessions, including task composition, planning-versus-execution decisions, domain expertise, and operationalized success.
Ollama — Ollama FAQ
Documents local prompt handling, local-only mode, networking, storage, and runtime configuration for operating models through Ollama on user-controlled hardware.
ggml.org — llama.cpp
Documents a C and C++ runtime for large-model inference across a wide range of local and cloud hardware, including GGUF models, quantization support, and an API server.
vLLM — OpenAI-Compatible Server
Documents serving supported models through an HTTP API, including runtime arguments, model configuration, parallelism, request options, and deployment-oriented server behavior.
Hugging Face — Fine-tuning
Defines fine-tuning as continued training of a pretrained model on a smaller task- or domain-specific dataset and provides an implementation path through Transformers.
Hugging Face — Parameter efficient fine-tuning methods
Catalogs parameter-efficient adaptation methods, including soft prompting, selective layer tuning, adapters, and Low-Rank Adaptation variants supported by the PEFT library.
Hugging Face — Quantization overview
Surveys lower-precision model representations and the quantization methods integrated with Transformers for reducing model memory and supporting constrained inference environments.
OpenAI — Introducing gpt-oss
Introduces two Apache 2.0 open-weight reasoning models intended for customizable deployment, including consumer-hardware and local-inference scenarios.
OpenAI — gpt-oss-120b & gpt-oss-20b Model Card
Documents the gpt-oss model family, release conditions, evaluations, safety analysis, capabilities, and limitations alongside downloadable model weights.
Stanford University — A Philosophy of Software Design, Second Edition
The author’s official book page identifies deep, general-purpose modules and separation of important from unimportant information as central software-design concerns.
arXiv / ICLR — LoRA: Low-Rank Adaptation of Large Language Models
Introduces Low-Rank Adaptation, which freezes pretrained weights and injects trainable low-rank matrices to reduce the number of parameters updated for downstream adaptation.