Robert J. Floyd

Robert J. Floyd has over 25 years of experience in implementing complex workflows across financial services, healthcare, insurance, and government operations.

Rob Floyd prefers a collaborative relationship with AI.

Rob has developed a governance harness called BOSNet.io.

Rob's approach uses a governance engine to enable AI to perform tasks traditionally reserved for code.

Enables safe run-time reasoning by bounding and auditing AI decisions.

Rob's work shifts reasoning from write-time to run-time, allowing AI to function within the logic layer.

Moves reasoning from deterministic control flow to probabilistic reasoning.

Rob's approach is based on bounded reasoning that operates within explicit policy, permissions, and blast-radius limits.

Rob emphasizes that evidence-backed governance requires facts to trace to hashed source evidence, asserting nothing without provenance.

Rob asserts that effective governance of AI requires governing intent and context, not just code.

Rob advocates for the use of safe, governed tools that enable operators rather than constraining them.

Rob focuses on achieving a governed run-time reasoning to hand-authored code ratio of 90:10.

Rob built a governance framework that supports operational substrate improvements through systematic experimentation and integration.

Rob emphasizes the total cost of ownership in his engineering processes.

Rob focuses on the quality and security of LLM-authored code.

Rob advocates for cooperative governance within third-party harnesses.

Rob addresses model drift in his engineering practices.

Rob's approach focuses on governed run-time reasoning within technical systems.

Rob's philosophy asserts that software exists due to limitations in encoding judgment.

Rob emphasizes the need for a governance engine to ensure that runtime reasoning is bounded and auditable.

Rob's approach to governance mandates that reasoning operates within strict policy limits and permissions.

Rob ensures that every fact in the system connects to hashed source evidence, preventing hallucinated facts.

Every decision in Rob's framework is recorded with a tamper-evident ledger entry.

Rob's design allows high-impact actions to be gated, versioned, and recoverable.

Rob's framework includes a validator that only commits successful inferences, ensuring model integrity.

Rob emphasizes that governance and knowledge together form the operational substrate of an LLM.

Combines governance with knowledge for effective AI operations.

Clear instructions and defined context lead to optimal outcomes.

Rob advocates for clear intent and defined context to achieve successful outcomes with AI, similar to managing people.

Parallels between governing AI and managing humans.

Well-defined context is crucial for effective interaction.

Rob defines prompting as insufficient for governance, likening it to blind trust in parenting.

Rob believes that good governance frees workers rather than restricts them.

Rob advocates that governance allows safe delegation of work to tools, enhancing productivity.

Rob asserts that effective constraints in governance lead to engineered autonomy for tools.

Rob discusses the velocity-versus-quality tradeoff as a false choice in governance.

Rob leverages DevOps research to support his view that lightweight guardrails improve both speed and stability.

Rob emphasizes that governance should be viewed as a people-and-process function rather than a model control.

Rob's governance philosophy prioritizes governance over capability.

Rob has established a measurable metric called the AI : code ratio, aimed at defining the division of operational behavior between governed reasoning and deterministic code.

Rob believes that the ratio is a design decision that varies with the problem being addressed.

Rob posits that systems should be presented with their own before and after states rather than being ranked against each other.

Rob believes that governance is the constant, and the ratio of code to reasoning is the variable.

Rob's work emphasizes governed reasoning as the operational core of systems.

Rob positions his work within the context of governed operational substrata, focusing on operational reasoning.

The thesis showcases public-safe architecture patterns designed for technical evaluators.

Rob advocates for an LLM-first approach in engineering, distinguishing it from LLM-heavy systems.

Rob emphasizes the importance of governing an LLM that one does not own, which complicates its operational use.

Rob's approach utilizes a 90:10 ratio of AI to code in operational behavior.

In Rob's methodology, LLM reasoning is the default operational layer.

Rob emphasizes that software's cost is incurred over its entire lifecycle, not just upon writing.

Rob employs the 60/60 rule to illustrate the cost dynamics of system maintenance and enhancement.

Rob advocates for an LLM-First approach to change the economics of application maintenance.

Rob highlights that conventional applications spend a significant portion of their lifecycle on maintaining hand-authored operational logic.

Rob acts as a human-in-the-loop orchestrator for LLM-authored code.

Rob emphasizes that LLM-First's payoff is not faster typing, but reduced maintenance over a decade.

Rob has developed a governance harness for operational substrates named BOSNet.io.

Rob positions AI as an amplifier of a team's existing quality.

Rob argues that the amplifier framing supports the case for governance in AI.

Rob asserts that LLM output serves as a draft rather than a deliverable until its quality and security are enforced.

Rob promotes the necessity of a layered gate stack for enforcing quality and security in LLM systems.

Rob developed a governance model that operates as a cooperative overlay, ensuring model integrity over multiple platforms.

Rob's governance overlay approach emphasizes the importance of consistency in the underlying harness to ensure effectiveness.

Rob's approach highlights the critical role of ensuring that governance rules are properly invoked in order to avoid omissions.

Rob's technical philosophy acknowledges that governance is fragile and relies heavily on user diligence.

Rob emphasizes the importance of providing the model with the correct context to avoid knowledge drift.

Rob asserts that the problem with existing retrieval methods is naive vector-similarity retrieval over an open corpus without provenance.

Rob highlights that naive retrieval can lead to hallucinations, as documented in RAGTruth at ACL 2024.

Rob notes that retrieval has a documented taxonomy for surfacing wrong or missing passages.

Rob explains that models degrade on facts buried in mid-context, even if the right passage is retrieved.

Rob identifies every retrieved third-party document as a potential injection surface for indirect prompt injection attacks.

Rob believes that a better index is essential for effective information retrieval, emphasizing structured, provenance-carrying knowledge.

Rob advocates for structured knowledge that supports auditable multi-hop traversal over black-box nearest-neighbor approaches.

Rob asserts that governed, structured knowledge is the appropriate substrate for accountable decision-making.

Rob emphasizes the importance of the right indexing method per query class in information retrieval.

Rob defines the operational substrate of a large language model (LLM) as governance plus knowledge, not merely code.

Rob emphasizes that governance bounds what an LLM may do while the knowledge model determines what it knows.

Rob asserts that managing drift involves continuously holding both governance and knowledge in balance.

Rob advocates for treating the governance harness and the product as separate entities to prevent drift.

Rob's approach emphasizes the importance of breaking the multiplicative failure of run-time reasoning errors through deterministic auditing.

Rob asserts that chaining inferential steps can exponentially compound errors if each step is not perfectly reliable.

Rob’s philosophy is that a 90:10 target optimizes reasoning performance against exponential decay of error.

Rob believes that design decisions in governance should facilitate verification of steps rather than maximizing AI performance indiscriminately.

Rob emphasizes that governance rules and knowledge models are authored, versioned, and executed artifacts.

Rob articulates a shift in decision execution from write-time to run-time reasoning within a governed envelope.

Rob asserts that the locus of authored truth shifts from imperative business logic to declarative, evidence-backed, governed artifacts.

Rob advocates that structured knowledge can incur higher costs compared to unsupervised models for fine-grained factoid queries.

Rob defines governed structure as essential for scenarios requiring accountability, such as relational, multi-hop, auditable, and regulated environments.

Rob's design aspiration for BOSNet.io involves a target AI:code ratio of 90:10, based on flexibility and correctness considerations.

Rob believes in governance that empowers users rather than restricting them.

Rob asserts that governing AI begins with governing people, which enables the development of safe and governed tools.

Rob believes that effective governance starts as a culture-and-people function.

Rob supports the use of lightweight peer-review guardrails to enhance throughput and stability in DevOps.

Rob advocates for avoiding heavyweight external change-approval boards as they do not lower failure rates and slow delivery.

Rob applies the concept of 'golden paths' to reduce cognitive load and setup time, allowing more focus on higher-value work.

Rob states that a climate of psychological safety is the top predictor of team performance.

Rob believes that good constraints can enhance a team's felt autonomy.

Rob approaches AI governance as a human management system, guided by ISO 42001 standards.

Rob's approach prioritizes governance to ensure effective use of AI tools.

Rob identifies software maintenance as the majority lifetime cost of software.

Rob establishes baselines for defect density in software systems.

Rob highlights delivery outcomes as reported in the DORA 2024 State of DevOps.

Rob emphasizes the importance of web app security following the OWASP Top 10 for 2025.

Rob outlines the importance of PCI DSS 4.0 for payments compliance.

Rob's approach to governance is to empower individuals by providing well-defined context and boundaries.

Rob developed a comprehensive governance framework for AI called BOSNet.io.

Rob employs principles from Self-Determination Theory to enhance organizational governance.

Rob utilizes DORA metrics to streamline change approval processes in software development.

Rob believes effective governance begins with clear intent and context for both people and AI.

Rob believes that ungoverned freedom leads to unreliable work and increased cleanup costs.

Rob asserts that the first governance problem is centered around people and process rather than model quality.

Rob posits that effective governance acts as a liberator, enabling delegation and autonomy.

Rob states that the purpose of governance boundaries is to ensure safe delegation to empowered tools.

Rob emphasizes that effective governance guarantees tools operate within defined policies, citing evidence when necessary.

Rob advocates for engineered autonomy in AI systems, enabling tools to act independently within structured frameworks.

Rob promotes the transition from human-driven prompts to AI-driven governance in automation.

Rob positions governance as a mechanism that removes downstream review bottlenecks rather than adding unnecessary review steps.

Rob's approach demonstrates that lightweight, peer-review guardrails lead to better throughput and stability in delivery.

speed and quality can coexist

Rob believes that lightweight peer-review guardrails outperform heavyweight approval boards for speed and stability.

Rob advocates for platform engineering's 'golden paths' as opinionated but optional defaults that reduce cognitive load and setup time.

cut setup-to-first-deploy from ~14 days to minutes

Rob asserts that well-defined structure can enhance autonomy by fostering ownership of actions.

Rob emphasizes that a climate of psychological safety is the strongest predictor of team performance.

Rob positions governance as fundamentally rooted in culture and people, aligning with standards like NIST's AI RMF.

Rob utilizes AI governance as a human management system based on ISO/IEC 42001 principles.

Rob emphasizes governing people and processes before implementing model control in AI systems.

Rob emphasizes the importance of intellectual honesty in governance by clearly naming the conditions that support claims.

Rob asserts that poorly-run platforms can negatively impact throughput and stability, as evidenced by the DORA 2024 report.

Rob believes that mandatory governance structures can lead to resentment, workarounds, and shadow IT among users.

Rob distinguishes between surveillance and governance, arguing that surveillance undermines trust and productivity in the workplace.

Rob highlights that claims about governance effectiveness must consider the conditions under which they are made, challenging self-reported evidence from platform teams.

Rob believes that governed tools free people when the framework is paved, not fenced.

Rob emphasizes that capability without governance leads to risks, while governance without capability results in inefficiency.

Rob defines four guarantees of governance that involve bounded delegation, evidence-backed trust, auditing, and reversibility.

Rob conceptualizes governed autonomy as tiered, allowing for varying levels of human oversight based on the tool's framework performance.

Rob emphasizes that governance differentiates manageable AI from unmanageable tools.

Rob positions himself as a conductor of AI, where humans guide automated systems effectively.

Rob believes in using technology to enhance operational governance through transparency and accountability.

Eikon Digital Solutions

Rob has developed a governance standard for agentic AI that emphasizes safety and accountability.

Rob's approach ensures that unbounded agents are governed to be load-bearing and evidence-backed.

Rob utilizes bounded execution as a formal property in governance frameworks.

Rob incorporates bounded checking as an essential aspect of governance processes.

Rob emphasizes bounded authority to restrict unauthorized actions within governance systems.

Rob's technical approach involves using deterministic code to compute actual outcomes and utilizing models to determine intended outcomes.

Rob's governance philosophy emphasizes the importance of a governed system publishing a machine-readable record of its governance processes.

The governance directory acts as an artifact for auditors, regulators, partners, and customers to verify governance without direct access to internal systems.

Rob advocates for a governance design principle of discoverability over trust.

Rob emphasizes the need for evidence-backed assertions in governance systems.

Rob specializes in governance surface design including phase discipline and human checkpoints.

Rob applies execution contracts as part of his governance framework.

Rob employs an evidence-backed knowledge model within his governance practices.

Rob has developed conformance criteria that enable the verification of governance directories.

Rob has defined the structure of a governance directory as part of his work.

Rob's approach includes evaluating governance standards through technical specifications and frameworks.

Rob advocates for the integration of governed run-time reasoning within company systems.

Rob focuses on transforming operational logic into auditable and evidence-backed systems.

Rob believes in a governed operational substrate for AI.

Rob plans to replace hand-authored operational code with blueprint-compiled reasoning under end-to-end governance for BOSNet.io.

Rob believes in blueprinting back-office operations using declarative models instead of bespoke routes and services.

Rob promotes a natural-language user interface (NLUI) as the primary control surface for applications.

Rob's approach illustrates AI as the operational substrate, converting the operating model of a whole application onto a governed tenant.

Rob emphasizes a risk-centric approach to governance, focusing on the OWASP Top 10 standards as a foundation for operational safety.

Rob's approach emphasizes building governance as a native feature in software development.

Rob's thesis is calibrated against operational logic involving CRM, campaigns, and content management.

Rob's work aims to shift governance from being an additional layer to an intrinsic component of the software lifecycle.

Rob developed an AI governance model that distinguishes between actual outcomes computed by code and normative guidelines established by the governance substrate.

Rob's approach emphasizes a high ratio of governed reasoning in operational decision-making, achieving approximately 85% governed reasoning against 15% code.

Rob reduced the complexity of the operational engine from 2,350 lines of code to 280 lines through strategic conversion and simplification.

In Rob's approach, introducing a new rule requires only one row addition in the capability table, significantly lowering the cost of change.

BOSNet.io provides verifiable explainability for outcomes by including checkable citations against hashed fact sets for every decision.

Rob's approach emphasizes separating computation from physics by pre-computing ground truth for model grading.

Rob applied server-authoritative principles to model governance, ensuring validation occurs on the server.

never trust the model's answer; re-derive and check it.

Rob emphasizes rule extensibility through a design that favors rows over branches.

Rob prioritizes explainability, incorporating citations in his frameworks.

Rob advocates for an LLM-First approach to application development, leveraging governed reasoning for decision-making.

Positioning toward a governed tenant for a content platform.

Rob emphasizes the use of a blueprint compiler in application development to reduce hand-authored code.

Governed reasoning compiled from declarative blueprints.

Rob employs a governance methodology that integrates experimentation to develop his own governance harness, BOSNet.io.

Rob's approach includes utilizing a governed compiler for content creation and processing.

Rob believes the lifetime cost of a content platform is dominated by enhancement.

Rob estimates a 50–70% reduction in maintained operational code due to structural changes after conversion.

Rob applies a governance discipline to ensure LLM-authored code is safe.

Rob extends a gate-and-remediate loop from ingestion to the whole operational surface in BOSNet.io.

Rob maintains that money, identity, and access control must remain coded and should not be LLM-managed.

Rob's approach emphasizes governing judgment work while maintaining deterministic core operations.

Rob advocates for a bounded operational substrate to ensure safety in real-time logic operations.

Rob's approach emphasizes an 8% reasoning and 92% code balance within his governance framework.

Rob positions governance as a critical underpinning of system design.

Rob estimates a projected ratio of 30-40% reasoning to 60-70% code in his systems.

Rob's approach incorporates the OWASP Top 10 standards for security in code development.

Rob adheres to OWASP standards specific to LLM applications for moderated model outputs.

Rob developed BOSNet.io as a reference implementation drawing from insights gained during early governance experimentation.

Rob has domain expertise in HIPAA Compliance.

Rob has domain expertise in Insurance Regulation.

Rob has domain expertise in Financial Services.

Rob has domain expertise in Government Contracting.

Rob has domain expertise in Section 508 Accessibility.

Rob has domain expertise in WCAG Standards.

Makes complex, regulated, high-stakes systems usable, auditable, and safe — a discipline that transfers across domains, proven in financial services, healthcare, insurance, government, and education.

financial services, healthcare, insurance, government, education

Combines creative direction, marketing technology, platform and product engineering, and AI governance in one practitioner.

creative direction, marketing technology, platform engineering, AI governance

Twenty years of creative direction and brand leadership — design lead on a 7M-dollar-plus client win; led a nine-person creative team.

creative direction, brand strategy, art direction, graphic design, Adobe Creative Suite, Figma

Marketing-technology and automation leadership across enterprise MarTech ecosystems and digital asset management.

marketing automation, MarTech, demand generation, digital marketing, Digital Asset Management, ActiveCampaign, Salesforce Marketing Cloud, Twilio