Anansi Labs — Independent Applied AI Research Lab

The Lab

Research that survives contact with reality.

We study how intelligent systems make consequential decisions when the problem gets complicated — through rigorous computational experimentation. We build enough to find out.

We are not primarily a software company.

We are not a federal contracting shop.

We are not a consulting company searching for billable hours.

We are a research institution that builds.

The opportunity finds the research.

We don't wait for a solicitation to decide what matters — we find an important problem, ask a difficult question, build the experiment, and follow the evidence.

Question→ Research→ Experiment→ Evidence→ Publication→ Reputation→ Opportunity

We pursue funding for our research more often than we invent research for available funding.

Intelligence must be tested in systems ◆ Architecture changes behavior ◆ Complexity creates emergence ◆ Decisions matter more than predictions ◆ Evidence outranks narrative ◆ Building is a form of research ◆ Independent inquiry has value ◆ Intelligence must be tested in systems ◆ Architecture changes behavior ◆ Complexity creates emergence ◆ Decisions matter more than predictions ◆ Evidence outranks narrative ◆ Building is a form of research ◆ Independent inquiry has value ◆

Annual research cadence

We create evidence of expertise.

Substantial publications

Approximately one major publication per quarter.

6–12

Research notes

Published when findings warrant them.

2–4

Demonstrations

Public experimental systems tied directly to research.

Presentations

Conferences, universities, and technical forums.

Quality outranks volume. One paper that becomes part of the conversation is worth more than twelve nobody remembers.

Five questions worth a laboratory.

Decision Science & Computational Strategy

How can intelligent systems improve consequential decisions under uncertainty?

Game-theoretic AIImperfect-information reasoningCounterfactual reasoningCourse-of-action analysisAdversarial decision-makingSequential decision problemsComputational wargamingStrategic optimizationResource allocationDecision confidenceAlternative futures

Agentic Systems

How do autonomous AI agents behave individually and collectively?

Multi-agent systemsAgent coordinationAgent managementDelegationMemory architecturesFailure propagationAgent organizationsEmergent behaviorHuman–agent teamsAI workforce architectures

We are particularly interested in the proposition that management architecture can materially change AI-system performance.

III

AI Reliability, Governance & Control

What happens when intelligent systems receive meaningful authority?

Authority boundariesFailure containmentHuman oversightVerification & validationObservabilityExplainabilityAuditabilityEscalation mechanismsFail-safe behavior

A decision system that cannot be trusted cannot reliably create decision advantage.

Autonomous Decision Systems

How can intelligent systems decide effectively with limited human intervention?

Mission autonomyDistributed autonomyHuman–machine teamingAutonomous planningResilient autonomous behaviorDynamic task allocationAutonomous coordination

Physical platforms may provide experimental environments. The research subject remains intelligence and decision-making, not platform engineering.

Complex Intelligent Systems

What happens when many intelligent components interact?

Complex adaptive systemsDistributed decision-makingHuman–AI organizationsCascading failuresCollective intelligenceInformation propagationDecision hierarchiesCompetitive intelligent systems

The central research thesis

Decision advantage through applied AI.

Decision advantage is the ability to make better-informed, more timely, more robust, or more strategically effective decisions under complexity, uncertainty, incomplete information, competition, or rapidly changing circumstances.

How can artificial intelligence improve the quality, speed, robustness, and understanding of consequential decisions under complexity and uncertainty?

The central research question

Anansi may investigate many technologies. But technology itself is rarely the research objective. The decision is.

Recognizable by how we study, not just what.

Observe Question Hypothesize Build Experiment Measure Challenge Reproduce Publish

We seek evidence that could prove us wrong.

A negative result is still a result. A failed architecture can be more informative than a successful demonstration. Unexpected behavior becomes the next research question.

The objective is not to prove Anansi correct. The objective is to discover what is true.

We do not hide inconvenient results. We distinguish observation from interpretation. We do not claim generality from narrow evidence. If the evidence does not support the desired conclusion, the conclusion changes — the evidence does not.

Every program leaves an evidence chain.

Research question — worth answering
Hypothesis — falsifiable
Experimental design — specific enough to challenge
Implementation — the system actually built
Data — raw or processed results
Analysis — disciplined interpretation
Limitations — what this does not establish
Reproducibility — artifacts for independent replication
Publication — a durable research artifact

A publish-first laboratory.

Papers, research notes, experimental reports, demonstrators, and Anansi Quarterly — the lab's flagship publication.

Anansi Quarterly · Vol. 01 · July 2026 · Conceptual Working Paper

The Agentic Conway Effect

A theory of source-code dispersion under massively parallel AI development — what happens to source code when a hundred machines write it at once, and whether the boundaries they carve will earn their keep. Eight propositions, and the experiment designed to break them: from a 72-run pilot to 1,040 runs.

Armond E. Sinclair, PhD

Download the paper · PDF

What the lab produces

Research papersFormal publications presenting original research.

Research notesShorter findings that don't warrant full papers.

Experimental reportsDocumentation of controlled investigations.

DemonstratorsWorking systems that test or communicate research.

Research artifactsCode, datasets, evaluation frameworks, reproduction packages.

BriefingsPresentations for conferences, universities, industry, and government.

The Anansi standard

Before our name goes on it.

What did we ask?

Why does it matter?

What did we build?

What did we test?

What happened?

What evidence supports that conclusion?

What doesn't the evidence prove?

Can someone challenge or reproduce it?

What should be tested next?

If those questions cannot be answered, the work is probably not finished.

The team

Small, independent, intellectually formidable.

Anansi takes its name from the Akan storytelling tradition — intelligence, strategy, and navigating situations where brute force alone is insufficient. The deeper metaphor is the web: ideas connect, systems interact, decisions create consequences. Anansi studies the web.

"We don't know yet. Here is what we would need to test next."

A sentence that should increase a laboratory's credibility — and does.

Armond E. Sinclair, PhD

Co-founder

Game theory, decision science, and mission engineering. Author of The Agentic Conway Effect.

Policarpio Soberanis, PhD

Co-founder

Optimization, applied AI, and computational experimentation.

Work with the lab

A research network, not a payroll.

Universities, laboratories, companies, foundations, and government research organizations engage Anansi where our accumulated research provides differentiated insight.

Research partnerships

Joint investigations with universities, laboratories, companies, and government research organizations.

Technology licensing

License intellectual property emerging from internal research programs.

Strategic research advisory

Selective engagements where our evidence base provides differentiated insight.