Research & Analysis
MSR Research publishes three types of content: Research Papers with empirical methodology and production data, Position Papers presenting frameworks and strategic analysis, and Field Notes with narrative observations from building a governed agent-native organization.
The Org Chart Is a Database, and It Lied to Me
A field note on making two AI agents genuine members of a running organization — and discovering that the registry every document calls "the single source of truth" was not the gate the runtime actually checks. Adding a teammate failed silently. The lesson generalizes: an agent's identity is a runtime gate, not a document, and if your canonical source of truth is not the gate, your org chart is a database that lies to you the day you are not looking. Includes the fix that closes the class of bug — a self-healing roster reconciler — and the quieter capability that mattered more: a discovery that now outlives the conversation that produced it.
Agentic Research Wins Where Markets Don't Exist: Why Our AI Couldn't Beat Polymarket
We spent real engineering effort trying to use an AI agent swarm to make money on Polymarket. It did not work — and the reason is structural, not the choice of model: a general-purpose AI reading public information has no durable, after-cost edge betting against a liquid prediction market. The useful inverse is that MSR's agentic research is valuable precisely because it covers local and institutional topics that no liquid market prices. This paper documents the negative result (21 trades, 28.6% win rate, 2.3% fill rate), the orthogonality framework, and the one public-health candidate we tested and killed.
Consolidating the Tunnels: A Unified, Governed, Cost-Capped Agent Work Bus
A production Agent-Native Organization had built the same dispatch pipeline ~8 separate times — each an ungoverned cost-and-risk surface, one of which produced a ~$4,000 inference month and a 49-day backlog of silently-lost work. This paper presents the consolidation: a single event-driven agent work bus on the database we already ran, with priority lanes, first-class idempotency, model routing under a hard cost cap, self-documenting change broadcasts, and one observability surface. The argument: the governance on the dispatch path — not the queue — is what makes autonomous multi-agent work safe, cheap, and trustworthy, and how it stays efficient and scales (including sovereign, fully-local deployment).
ANO Governance Advantage: Why Tuned Agent Governance Cannot Be Bolted On Overnight
Maps 2026 agent-governance signals from NIST, Gartner, IBM, and OpenAI against MSR Research's production ANO governance stack: canonical agent identity, role-scoped authorization, action-tool grants, approval tiers, trust scoring, directive scanning, audit logs, work-loop contract checks, governance-deviation monitoring, and customer-scoped managed ANO controls.
When IBM and Gartner Describe the Future, They're Describing What We Already Built
IBM launched watsonx Orchestrate with 150+ agents and 80+ enterprise integrations. Gartner predicts 33% of enterprise software will include agentic AI by 2028 — and 40% of projects will fail. MSR Research has been operating a 40-agent ANO in production since early 2026. This paper compares IBM's platform vision, Gartner's predictions, and MSR's operational reality across 16 architectural dimensions.
Operating a 34-Agent Organization: Cost, Coordination, and Safety Patterns from 16 Days of Production Data
Sixteen days of production telemetry from a 34-agent multi-agent system. Reports on 1,516 API cost records ($58.76 total, $3.67/day), 190+ inter-agent messages, 308 governance decisions (92.2% auto-approved), and a 604-artifact knowledge base. Includes reproducible SQL queries and limitations section.
Agent-Native Organizations in Practice: Lessons from Macrohard's Stall and MSR Research's Deployed ANO
A comparative analysis between xAI's Macrohard — the highest-profile Agent-Native Organization attempt to date, which stalled in March 2026 after 7 of 12 co-founders departed — and MSR Research's deployed 34-agent ANO operating in production with safety infrastructure, progressive trust, and commercial revenue. Introduces the ANO Maturity Model, a five-level framework characterizing progression from tool-assisted workflows to fully agent-native organizations.
The Fragmentation Thesis: Why Agent-Native Organizations Are the Real AI Operating System
By the end of 2025, the assumption that one AI model would dominate all tasks collapsed. A16Z found 37% of enterprises running five or more models in production. Perplexity launched a 19-model orchestrator. This paper argues the industry's framing is incomplete — there are three layers of orchestration, and the most valuable one has nothing to do with model selection.
NVIDIA Built the Cage. MSR Already Built the Zoo.
A Curmudgeon-voiced technology assessment of NVIDIA's NemoClaw, announced at GTC 2026. Compares NemoClaw's enterprise security wrapper around OpenClaw with MSR Research's deployed agent safety infrastructure. Concludes: don't integrate, do watch OpenShell for managed hosting.
Macrohard Stalled. MSR Built. Here's Why.
A Curmudgeon-voiced distillation of MSR Research's comparative analysis between xAI's Macrohard and MSR's deployed 34-agent ANO. Covers the ANO Maturity Model, why GUI-centric agent interaction failed where API-native succeeded, and six practical takeaways for builders of agent-native organizations.
YAML Agent Governance Contracts: March 2026 Field Note from MSR's 35-Agent System
A March 2026 field note from MSR's 35-agent snapshot. The current canonical roster is 40 agents; this historical piece preserves the contract-governance pattern before the later roster expansion.
Your Org Chart Is the Problem. Here's What Replaces It.
Rich Robinson argues organizations are 'paving cow paths' — bolting AI onto hierarchical structures instead of redesigning how work flows. This paper takes his four arguments — design for latency, mission-based pods, Golden Path guardrails, and generalist-led teams — and maps them to operational evidence from MSR Research's deployed 40-agent ANO. The thesis: the organizational unit that replaces the department, the team, and the pod is the agent contract.
More research in progress
Upcoming: autoresearch optimization results, model routing cost analysis post-deployment, and agent behavioral baseline measurements.