Research
Autonomous coding agents, agentic organizations, and the patterns of AI-native software engineering.
A comprehensive, continuously-updated reference covering 25+ agent systems, 180+ infrastructure vendors, and the architecture driving the category — mirrored from automate.engineering.
Sponsored by 
Start here
Top PicksJust want recommendations?Opinionated, star-rated picks for each category — frontier models, coding agents, sandboxes, CI runners, observability, evals, memory, safety, and self-hosted inference. Skips the methodology; points at what to use.Read the picks →The reference
Deep dives on 30+ coding-agent systems — Stripe Minions, Claude Managed Agents, Vercel Open Agents, OpenAI Symphony, OpenHands, Hermes Agent, GStack, GBrain, AgentHub, the Steinberger ecosystem, and the 28-CLI harness comparison
The model layer underneath — closed-source frontier (Anthropic / Google / OpenAI / xAI), open-weights (DeepSeek / Qwen / Llama / Kimi / GLM / MiniMax / Mistral), and agent / coding specialists; with pricing and decision shortcuts
Cross-cutting architectural patterns — harness engineering, isolation strategies, orchestration models, context management, feedback loops, failure recovery, multi-agent coordination
The deep-dive page on what makes agents reliable — five-subsystem model, repo-as-system-of-record, WIP=1, three-layer verification, sprint contracts, clean-state exits
The named discipline of curating what's in the window — write / select / compress / isolate; attention budget; 95% / 85% compaction thresholds; the four context failure modes
How to write tools agents use well — consolidated actions, ResponseFormat compression, Tool Search Tool (-85% tokens), code-as-tool sandbox pattern (150K → 2K), Tool Use Examples (+18pp)
The cross-vendor primitive (Anthropic open standard, Dec 2025) for capability packaging — SKILL.md format, progressive disclosure, the 82% vs 9% lift, the ~12-skill ceiling
Persistent state across turns and sessions — three-axis taxonomy, episodic / procedural / semantic split, vendor map (Letta, Mem0, LangMem, LangGraph Store, Anthropic memory tool), filesystem-as-memory
How to measure agent quality (distinct from benchmarks) — pass@k vs pass^k, three silent invalidators (grading bugs, infra noise, eval awareness), tooling map (Inspect AI, LangSmith, Braintrust, Langfuse, Phoenix)
SWE-bench, SWE-bench Verified / Pro / Multimodal / Multilingual, Terminal Bench 2.0, τ-Bench, plus a 9-row "other benchmarks worth knowing" roundup; how to read the leaderboards and what they actually mean
Where does trust live? The three philosophical schools (Polosukhin / Chase / Ng) and the four operational schools (Stripe / Tan / Walking Labs / Steinberger)
29 named profiles of the people shaping the field — researchers, operators, chroniclers — with the single thing of theirs to read or watch first
How companies organize around agents — Stripe model, open-source model, agent-first development, infrastructure tiers
LLM inference solutions: direct API providers, platforms (Nebius, Together, Fireworks, Groq), routing gateways, self-hosted inference
The execution-environment layer — purpose-built agent sandboxes, Contree deep dive, CDEs, isolation tiers, integration patterns
150+ infrastructure vendors across 9 categories — turnkey platforms, agent-optimized hosting, orchestration, Cloud Mac, GPU clouds, VPS for agents, memory, observability, MCP, identity/auth
The agent's front-end story — Static, Declarative (A2UI), Open-ended patterns; CopilotKit, AG-UI, A2UI, MCP-UI; Vercel AI SDK; trade-offs between consistency and flexibility
Source-of-truth bibliography: structured digest of 100+ primary sources behind everything above — key claims, specific numbers, frameworks named, "which slot it fills"
Also in this reference
What's been added to the site, newest first. Content additions only — bug fixes, refactors, UX passes, and star refreshes are in git history but not tracked here.
Cost Economics for Coding AgentsDeployment for Coding AgentsEventsConferences, meetups, livestreams, and other in-person and virtual events relevant to agentic engineering.
Observability for Coding AgentsReading ListCurated entry points to follow the field — daily firehose (Simon Willison, TLDR AI, The Batch), weekly substantive (Import AI, Latent Space, Hamel.dev), deep long-form (Lil'Log, Cameron Wolfe, Chip Huyen), podcasts, courses, communities, conferences, reference repos. A practical weekly cadence at the bottom.
Safety for Coding AgentsTable of ContentsFull sitemap of this reference — every chapter and major section, with deep links.
Tenki — A Deep ReviewVendor review of Tenki — Sandbox, Runners, and Code Reviewer — benchmarked head-to-head against E2B / Daytona / Modal, Blacksmith / Depot / Namespace, and CodeRabbit / Greptile / Qodo / GitHub Copilot review. The bundle thesis, the strategic position, and where DevRel adds the most leverage.