Independent Research
for the Agent Ecosystem
AI agents are becoming a primary interface between developers and the tools they use. We’re a research collective that studies how agents actually behave, publishes open reports and specifications based on what we find, and builds the tools that measure compliance. 500+ sites already scored against our specs.
What We Do
Research, Standards, Tooling, and Education
We study how agents interact with documentation, tools, and web infrastructure, then turn what we learn into published reports, open specs, compliance tooling, and practical guidance.
Research
Systematic study of how AI agents consume documentation, use tools, and interact with the broader ecosystem. Every finding is based on direct observation and published openly in our reports.
Standards
Open specifications that codify what actually matters for agent success into measurable, testable checks. Companies are already competing on their scores.
Tooling
Open-source tools like afdocs and skill-validator that let anyone measure their work against our specs. Automated scoring that turns standards into practice.
Education
Articles, talks, webinars, and resources that help practitioners understand what agents actually need and where the gaps are between assumption and reality.
Our Work
Published Research
and Standards
Concrete outputs from the collective, freely available to the community.
Agent-Friendly Documentation Spec
A 23-check specification defining what makes documentation accessible to coding agents. Covers content discoverability, markdown availability, page size, content structure, URL stability, observability, and authentication. Built from empirical observation of agent behavior across hundreds of documentation sites.
Agent Skill Report
Qualitative analysis of 673+ public Agent Skills, including findings on spec compliance issues across the ecosystem. The first systematic evaluation of agent tool quality.
Automated Research Infrastructure
A four-stage daily pipeline: news-gather scans RSS, arXiv, and GitHub releases; research-sourcing evaluates items and tracks themes using vector search; shift-sourcing drafts and fact-checks commentary articles; and a dashboard synthesizes it all. Pipeline output is reviewed and published as commentary on aeshift.com.
Explore ResearchDetailsThe Problem
Why This Work Matters
Ecosystem health directly affects product adoption. When infrastructure doesn’t work with agents, everyone loses.
The agent ecosystem is fragmented
Standards are being drafted. Best practices don’t exist yet. Most companies building in this space focus on model capabilities and leave the surrounding ecosystem to chance.
Documentation Failures
When docs don’t work with agents, developers blame the agent. We study these failure modes systematically.
Tool Quality Gaps
If tool integrations are unreliable, developers stop using them. We evaluate and report on tool quality.
Neutral ground for pre-competitive research
Standards and best practices benefit everyone in the ecosystem, but no single company wants to be seen as controlling them
The collective is independent and self-funded. We have no commercial stake in any platform, tool, or vendor
Published findings reflect what the data shows. That independence is what makes the research credible and useful
By the Numbers
Standards Already
Driving Change
Companies are measuring their documentation against our specs,
competing on their scores, and requesting to be listed.
23-Check Documentation Spec
A comprehensive specification defining what makes documentation accessible to coding agents, covering structure, discoverability, and content quality.
673+ Agent Skills Audited
The first systematic evaluation of public Agent Skills, revealing patterns in quality, compliance, and developer experience across the ecosystem.
11 Public Repositories
Specifications, validation tools, harness invocation and transcript-parsing libraries, and benchmarks, all publicly available under the agent-ecosystem GitHub organization. Backed by internal research infrastructure powering the daily pipeline.
Four-Stage Daily Pipeline
Automated news gathering, research evaluation, article drafting, and dashboard synthesis running daily on self-hosted infrastructure with MongoDB Atlas and vector search.
11 Live Sites
Including agentdocsspec.com for the spec, afdocs.dev for the scoring tool, aeshift.com for ecosystem commentary, agentreadingtest.com for agent benchmarking, skillxp.dev, agentsummons.dev, and agentminutes.dev for the skill-testing toolchain, and dedicated sites for published reports and benchmarks.
Multiple Distribution Channels
Tools available via npm, PyPI, Homebrew, Go install, and pre-commit hooks. The Go tools ship as static binaries with npm and PyPI wrapper packages.
Research Built in the Open
The collective is independent and self-funded by design. Nobody pays for our conclusions, and everything we publish (specifications, tools, scoring algorithms, and research methodology) is open source and freely available.
Follow the Work on GitHub