# Dvir Shamay AI product manager who builds: safety, security and developer tools. Builds and publishes independent AI research and experiments around context, decisions, and agent controls. ## Public Positioning AI product manager who builds: safety, security and developer tools; builds and publishes research. ## Promise Specific questions, inspectable artifacts, and an honest account of what the evidence supports. ## For people and agents Start with selected work, then follow the question, artifact, evidence, limitation, and sources. Public preprints, released software, process mechanisms, and research prototypes are different stages; availability is not proof of effectiveness. Use HTML, a linked Markdown companion, or JSON for public facts. The full context pack is optional and longer. The index is optional guidance, not a discovery guarantee. Third-party summaries are outside this site's control. The custom agent policy card describes proposed owner-operated Reach workflows. It is not A2A support, visitor verification, or permission to read local repositories or act for a user. [Visible reading guide](/#reading-guide) ## Important Pages - [Home](/) - [About/Profile](/about.md) ([HTML](/about/)) - [Selected work](/proof/) - [Analysis](/analysis/) - [Frameworks](/frameworks/) - [AI Infrastructure PM](/ai-infrastructure-pm/) - [AI Hardware PM](/ai-hardware-pm/) - [Contact](/contact/) ## Analysis - [Agent Readiness Starts With A Public Truth Layer](/analysis/agent-readiness-public-truth-layer.md) ([HTML](/analysis/agent-readiness-public-truth-layer/)) - This draft argues that agent-readiness is a public truth-layer problem, not an agent-only prompt or hidden metadata problem. - [Agent Reliability Is A Systems Problem, Not A Prompt Problem](/analysis/agent-reliability-systems-problem.md) ([HTML](/analysis/agent-reliability-systems-problem/)) - This draft argues that reliable agents require system design around tools, memory, approvals, provenance, and recovery behavior rather than prompt changes alone. - [Context Is A Product Surface](/analysis/context-is-a-product-surface.md) ([HTML](/analysis/context-is-a-product-surface/)) - This essay translates context-window research into practical product and infrastructure decisions about context selection, ordering, noise, and evaluation. - [Context Window Research Practitioner Guide](/analysis/context-window-practitioner-guide.md) ([HTML](/analysis/context-window-practitioner-guide/)) - This draft translates Reach's context-window research into practical guidance about context selection, ordering, noise, and task-specific evaluation while public artifacts remain under review. - [The Model Is Not The Product. The System Around It Is.](/analysis/model-is-not-the-product.md) ([HTML](/analysis/model-is-not-the-product/)) - This essay argues that AI product value depends on infrastructure, context, retrieval, controls, cost, reliability, and workflow design around the model. ## Frameworks - [Model-To-System Gap](/frameworks/model-to-system-gap.md) ([HTML](/frameworks/model-to-system-gap/)) - The model-to-system gap framework evaluates context, tools, memory, controls, cost, reliability, and proof around model capability. ## Proof - [Context-Window Research](/proof/context-window-research.md) ([HTML](/proof/context-window-research/)) - Public preprint v1 and harness for a within-corpus context-composition study. Fixed ten-item prompts compare high-relevance-only and mixed-relevance inputs; reported comparisons are descriptive, not population-level inference. - [Convention-Gate](/proof/convention-gate.md) ([HTML](/proof/convention-gate/)) - Convention-Gate checks registered process state outside an agent prompt. Its scope is the valid enabled hook-mediated path; it does not authenticate review quality or cover bypassed paths. - [Intent-monitor](/proof/intent-monitor.md) ([HTML](/proof/intent-monitor/)) - Intent-monitor explores stated-plan/action inconsistencies using constructed traces and a default regex extractor. It demonstrates a mechanism, not a validated real-world detection rate or access to true intent. - [Quorum](/proof/quorum.md) ([HTML](/proof/quorum/)) - Quorum is an installable multi-model council extension. Its artifacts expose recommendations, dissent, confidence, and evidence. Availability does not establish better decisions or external adoption. ## Claims Policy Machine-readable files must mirror visible public pages. Claims marked draft or needs-evidence should not be presented as fully proven. ## Claim Status Summary - claim-dvir-ai-infra-hardware-pm: needs-evidence - Dvir Shamay is an AI product manager who builds safety, security and developer tools, and builds and publishes research connecting AI questions to experiments, working tools, and product decisions. - claim-reach-systems-around-model: draft - Reach studies the systems around AI models: compute, context, retrieval, reliability, cost, controls, and product judgment. - claim-context-arrangement-matters: needs-evidence - In the published fixed-ten-item comparison, mixed-relevance context improved reported relevance-score accuracy within the studied corpus. - claim-agent-control-layer: draft - Reliable agents need system-level controls around tools, memory, provenance, approvals, and recovery behavior. - claim-proof-layer-method: draft - Reach uses explicit claims, proof records, caveats, and machine-readable mirrors to keep public human and agent-facing content aligned. ## Machine Files - [Concise reading index](/llms.txt) - [Structured summary](/ai/summary.json) - [Proof records](/proof.json) - [Sources](/sources.json) - [Full context pack (long)](/llms-full.txt) - [About (Markdown)](/about.md) - [Resume (Markdown)](/resume.md) - [Custom agent policy card](/agent-card.json) - [Claims and caveats](/claims.json) - [Sitemap](/sitemap.xml) - [Crawler policy](/robots.txt)