About
How may we make sense of our reality so that we may think, believe, and act anew…
Activity
8K followers
-
Mark Spivey reposted thisMark Spivey reposted thisAgentic Systems: Shrinking Blast Radius at Shared Boundaries (Safe by Construction) Agentic systems are power tools. They don’t just recommend��they act. That’s the shift that matters. When an agent can read, write, and trigger workflows across real systems, the security question stops being “Is the model safe?” and becomes “What can it reach, and what happens when it’s wrong?” That’s where blast radius becomes the core design primitive. The trap: “Separate systems” doesn’t mean separate risk. A common early comfort is: “These agents don’t run in the same system, so they can’t leak to each other.” But agentic failures don’t require a shared runtime. They require a shared boundary. If two isolated agent deployments both connect to the same downstream tool—Jira, Slack, Google Drive, a data warehouse—then that downstream tool becomes the bridge. You’ve created a shared surface where data can mix, be discovered, or be amplified. In other words: Your real boundary isn’t where the agent runs. Your real boundary is what the agent can touch. Shared boundaries are where blast radius explodes. The blast radius of an agent isn’t theoretical. It’s the union of: • every credential it can use, • every tool endpoint it can call, • every resource those endpoints can reach, • and every side-effect those calls can create. Shared boundaries multiply risk because they turn “local” mistakes into “global” consequences: • An agent posts sensitive data into a shared Jira project. • Another agent indexes it. • A third user searches it. • Now your “separate systems” have quietly become a data-exchange bus. This is the agentic equivalent of leaving the doors locked but sharing the same ventilation system. Design rule: treat tool integrations as security perimeters In agentic architectures, integrations aren’t “connectors.” They’re perimeters. Agentic systems change the cost of mistakes. They turn “oops” into “oops at scale.” So frame the architecture around three ideas: 1. Blast radius is the unit of risk. 2. Shared boundaries are where blast radius grows silently. 3. Safe by construction means constraints, not conventions. If you build it right, you can still move fast. But your speed won’t be coupled to your worst day. Because in agentic systems, the question isn’t whether something will fail. It’s whether failure can cross the boundary you care about.
-
Mark Spivey posted this#Ontologies and #KnowledgeGraphs and any #SemanticTech (#SemanticLayer) doesn’t actually formalize #Semantics … it formalizes #Syntactics and #Pragmatics of which #Semantics is an emergent property of, situated and contextualized as such, which shouldn’t be assumed nor expected to be generalizable in the sense of merely thinking everyone can and should adopt the same ever upper ontology or ever specialized domain ontology .
-
Mark Spivey posted thisdeclarative #InfrastructureAsCode is where the #metadata for enterprise IT is … this needs to be represented as or at least decorated with distributed hypermedia affordances and #LinkedData …
-
Mark Spivey posted thisI’ve noticed that #Data people are co-opting and commandeering #Knowledge domain recently, likely as a result of emergent #GenAI opportunities from the #LLM and #RAG angles regarding #KnowledgeGraphs regarding a #DataMesh trajectory … and then the rediscovery of value of #Ontology (#SemanticWeb #LinkedData) driven approaches, especially regarding federated and decentralized and distributed approaches … this is likely catching traditional #Knowledge and #LearningAndDevelopment people off guard … which is ok, but!: everyone needs a reminder that #KnowledgeGraphs and #KnowledgeRepresentationAndReasoning and #KnowledgeManagement are all far different things for different purposes for different reasons … the interrogatives shouldn’t get confused nor conflated . everyone also forgets about #REST and #Hypermedia regarding #LinkedData . #DataProducts #DCAT #DPROD should include hypermedia affordances in representations, as that is where value is derived from human or machine perspective of behavior actualization . I’m working on #HYPRCAT which is a #DCAT profile which decorates #DataMarketplaces #DataCatalogs #DataProducts with distributed hypermedia affordances . I’m designing reference implementations first regarding #xAPI regarding the #LearningEngineering domain .
-
Mark Spivey reacted on this#UpperOntology and #DomainOntology are roles Ontologies play in context of a particular usage which emerge from one Ontology declaring sub/super class or set relations regarding the other . any particular Ontology is neither an “upper” or “domain” ontology independent of explicit assertions asserting such relationship . normally any #OntologyMapping or #OntologyAlignment gets complex quickly in regards to assuming one entire Ontology can magically be considered a true or general Upper Ontology or Domain Ontology regarding another … it’s far more messy and muddy and complex than that … meaning is usage, structure emerges from usage . usage in context dependent and situated . context doesn’t transfer . reification (assumption and expectation of context transfer) can and should be explicitly modeled .
-
Mark Spivey posted this#UpperOntology and #DomainOntology are roles Ontologies play in context of a particular usage which emerge from one Ontology declaring sub/super class or set relations regarding the other . any particular Ontology is neither an “upper” or “domain” ontology independent of explicit assertions asserting such relationship . normally any #OntologyMapping or #OntologyAlignment gets complex quickly in regards to assuming one entire Ontology can magically be considered a true or general Upper Ontology or Domain Ontology regarding another … it’s far more messy and muddy and complex than that … meaning is usage, structure emerges from usage . usage in context dependent and situated . context doesn’t transfer . reification (assumption and expectation of context transfer) can and should be explicitly modeled .
-
Mark Spivey reacted on thisare people who are curious about the phrase Knowledge Graph Construction unaware of the historical phrases of Ontology Learning or Automated Ontology Engineering ... ?
-
Mark Spivey posted thisare people who are curious about the phrase Knowledge Graph Construction unaware of the historical phrases of Ontology Learning or Automated Ontology Engineering ... ?
Mark Spivey commented on a post
1mo
the way I’d put it is we don’t manage meaning, we manage for meaning … in a complexity facilitation sense .
Mark Spivey commented on a post
8mo
definitely, and when I think through this type of stuff, I tend to do “semiotics of data” and then explore, describe, and explain that .
“semantics” is just one part ppl like to focus on, the others being syntactics and pragmatics …
ppl will never focus on pragmatics enough which I would say is what teases out the “aboutness” .
I tend to wrap all this line of think up into “emergent semiotics” .
-
Mark Spivey liked thisMark Spivey liked thisThe failure modes breaking enterprise AI today have names. Ontologists coined them 25 years ago. → Four conflicting definitions of "revenue": the open vs. closed world assumption → Stale org charts presented as current: the reification problem → RAG systems answering from a world that no longer exists: the context graph problem → Provenance and multi-tenancy feeling intractable: the named graph problem Same problems. Different decade. Most teams building today have never seen the original map. Kurt Cagle wrote the map in the latest Context and Chaos. "The semantic web failed to scale. The arguments it generated did not." Full piece on Context & Chaos 👇Four Arguments Ontologists Never Finished (And Why AI Teams Will Have Them Again)Four Arguments Ontologists Never Finished (And Why AI Teams Will Have Them Again)
-
Mark Spivey liked thisMark Spivey liked thisWhen Steve Jobs came back to Apple and revealed the "Think Different" campaign, he talked internally to his team about Nike's logo, and how "Just Do It" was so successful because it planted a flag for what the company stood for. That flag stated their identity: who they were, and who they were not. He challenged the room to think of what they stood for, and why someone should choose to be part of Apple's identity. "Current LLMs are unreliable delegates: they introduce sparse but severe errors that silently corrupt documents, compounding over long interaction." Read that again, it's from Microsoft Research. They tested 19 LLMs on long delegated editing tasks: even frontier models corrupted ~25% of a document's content by the end of the workflow. Another important paper: A NeurIPS 2025 spotlight from Berkeley found most (about three-quarters of) multi-agent system failures were system-design and inter-agent coordination problems, not model limitations. (both paper links in comment section). I believe we're heading toward preventable economic harm driven by misplaced assumptions about LLMs and context architectures. I shared this with a VC yesterday, who used to work at Apple, and he commented that for some reason, everyone seems to be okay that "whoops, it just made stuff up", assuming it will be fixed at some point later. A larger textbook isn't how you help students learn. You want surgical precision to find & fill their knowledge gaps. Here's a flag I'm planting: not addressing context problems, meaning letting these silent corruption events propagate across decisions and documents, leads to preventable economic risks. I'd say it was a belief, it's not. Non-technical people deserve an even playing field when they make decisions about their company's future, their employees, and AI. Tony Seale said it brilliantly: "Learning happens in networks ... and the lessons have to live somewhere that is yours." Context is the knowledge you don't have that is required to understand the knowledge you do have. Every decision depends on context. All of this requires networks of systems communicating, each with their own context. The "just add more tokens strategy" is going to hit a limit, and good luck arguing with math. These risks are barely beginning to be understood, let alone model, price and manage them. The good news: a lot of this is fixable. We're building an open context specification at the W3C Context Graph Community Group, a way to precisely measure context and distribute detection sensors at scale, so teams can catch and prevent these silent corruption events before they propagate. There's a lot of math and intent, please be patient, and please reach out! Intelligence scales. But so do errors from a lack of context infrastructure. Protect your teams, your people, and your organization.
-
Mark Spivey liked thisMark Spivey liked thisAn LMS database is a filing cabinet inside one room. An LRS is a centralised records system that every room in the building can write to. That one shift in thinking changes a lot. The next article in the Beyond the LMS series breaks it down in plain language, no technical background required. #LRS #xAPI #DigitalLearning #LearningData #BeyondTheLMS #LearningDesign #EdTech #eLearning #CapabilityEvidence #LandD #LMSYour LMS Has a Database. An LRS Is Something Else Entirely.Your LMS Has a Database. An LRS Is Something Else Entirely.
-
Mark Spivey liked thisMark Spivey liked thisEvery C-suite is arriving at the same uncomfortable realisation right now. We spent a decade building it. The warehouse. The lakehouse. The dashboards. It was expensive. It was impressive. And - most importantly - it worked. But here's the part no one wants to say out loud - it is also the wrong shape for AI. 🔵 Bad context in, bad decisions out. You can buy the best model on the market. Give it tools, memory, a beautiful interface. Point it at fractured, siloed, contradictory enterprise data and it will reason flawlessly straight to the wrong answer. In the pre-AI era, success looked like getting 1% of your data estate into a warehouse. That ain't gonna cut it for what's already here - and it sure ain't gonna cut it for what's coming next. This is the elephant in every enterprise: the data integration problem we have been postponing for twenty years. AI did not create it. AI just made it impossible to keep ignoring. 🔵 BI was built for a box. Intelligence needs a network. The analytics stack is shaped like a box. Rows, columns, tables. Clean rectangles, optimised for one job: aggregating the past into a report. It did that job well. But an agent does not ask for a report. It asks what relates to what. It follows chains of meaning across products, customers, risks, processes and policies - crossing the boundaries the warehouse kept apart. And a box cannot hold a chain. The moment intelligence needs relationships, the rectangle breaks. The age of AI is not box-shaped. It is network-shaped. 🔵 Learning happens in networks. This is not a metaphor. It is the deepest pattern we have. Neural networks - the creative, generative half. The part that imagines. Data networks - the grounded, factual half. The part that knows what is actually true. ⚡ The neural network imagines. The data network grounds. ⚡ You already have the first half. Everyone does - it is a commodity now, the same frontier intelligence your rivals get within months. What almost nobody has built is the second half: a connected model of their own reality, solid enough for intelligence to stand on. 🔵 And the lessons have to live somewhere that is yours. Here is the part that should keep directors awake. In a network, the structure IS the knowledge. Every relationship your business learns - what a good customer looks like, why that deal failed - is a new edge in the graph. More than that, accumulated conceptual abstractions form the ontological core of your organisation. It is the most valuable asset you will build this decade. So do not hand it away. The moment you let a third party hold the model of your meaning, you have outsourced your moat. The intelligence is rented. The structure has to be owned. Connect the data. Model the meaning. Keep it yours. Because the box served the last era well. It is just the wrong shape for this one. 🔗 Want help getting your data in shape for AI? Let's talk: https://lnkd.in/ezHU2amU
-
Mark Spivey liked thisMark Spivey liked thisA JSON Schema is not an ontology. Everyone building LLM extraction pipelines is converging towards a JSON Schema (YAML same thing). It works really well, because function calling (say, Pydantic AI) and structured outputs all speaks JSON natively and you get typed, validated data out of your model. But there's a quiet assumption baked into that choice. A JSON Schema describes the shape of data, it has no concept of what the data means. Here are 4 things it simply cannot do and why they matter once your knowledge graph grows past a few hundred nodes. 1. Two extractions, two valid JSON objects, same real-world person. JSON Schema sees them as unrelated. OWL's functional properties and `owl:sameAs` let a reasoner merge them automatically. Without this, your graph accumulates ghost duplicates indefinitely. 2. A JSON Schema composes structure but it carries no semantic inheritance. `OWL's rdfs:subClassOf` means a reasoner knows a Researcher is a Person without you re-stating it everywhere. Add property chains and you get implicit relationships inferred from existing ones for free. 3. SHACL validates relationships between nodes — every Contract must have exactly one signatory who is a verified Person. That's a constraint you cannot express in JSON Schema at all. 4. Not the least...the open-world assumption. A JSON Schema treats missing information as an error. OWL treats it as unknown. In a knowledge graph built from thousands of documents, most entities are partially described. That's not a bug, rather the nature of real-world knowledge. The closed-world assumption is correct for API contracts. It's wrong for knowledge graphs. Now, of course I use YAML and JSON for POCs and MVPs. It's easy and fun. My remarks above relate to production, scale and enterprise needs. The world is a complex place and the variety of approaches available reflects the variety of challenges and business needs. #KnowledgeGraphs #Ontology
-
Mark Spivey liked thisMark Spivey liked thisThere is no avoiding it. Data catalog is a bad name for a data catalog. link to my thoughts in comment
-
Mark Spivey liked thisMark Spivey liked thisLet me make one thing clear straight away. This post isn’t going to claim that SCORM is dead or that LMSs are on their way out. I don’t believe that, and above all, those in the industry don’t want that. So, even though some people try to suggest it every now and then, let’s put that topic out of the way straight away. I’ve created this post, this meme, simply because I love cinema, because I think it offers us visual metaphors that help us express ourselves. Anyone who knows me knows that for years I’ve been using stills from famous films to illustrate the world of e-learning standards. This is the latest one. The aim of the image is to show that today, more than ever, we have the opportunity to fulfil the promises of XAPI: to be able to imagine experiences that exist wherever they’re needed, allowing us to gather data and their intrinsic meaning. Today, we can use the concepts of headless training and vibe authoring to gain complete freedom to decide not only where the experience takes place, but also what information is used to generate the XAPI statements that will help us understand the meaning of that experience. I certainly don’t have the hair of the late Patrick Swayze, but I have always fought to offer the best opportunities for those who want to learn and improve. And if you’re young and haven’t seen the film, well, I feel sorry for you. #xapi #scorm #vibeauthoring #vibe #authoring #lms #elearning #training #experienceapi #dirtylearning
Experience & Education
-
Foxxi Mediums Inc (dba XWISEE)
View Mark’s full experience
See their title, tenure and more.
or
Licenses & Certifications
-
-
-
-
-
HFI Certified Usability Analyst (User Experience Design)
Human Factors International
IssuedCredential ID 2012-3881 -
Cloudera Certified Developer for Apache Hadoop (Data Science & Analytics)
Cloudera Foundation
Issued -
Courses
-
BLOC.io - Full-Stack Web Development
-
-
COGNITIVE EDGE: #Complexity Informed #Leadership
-
-
COGNITIVE EDGE: SenseMaker Signification Design (Qualitative + Quantitative #Research Design)
-
-
COOPER U: Interaction Design (#ExperienceDesign #DesignThinking)
-
-
LEARNING ANALYTICS: Learning and Knowledge Analytics
-
-
OREILLY - Implementing Hypermedia APIs (RESTful)
-
-
SALT BOX: TinCanAPI (Experience API, #xAPI) Workshop
-
-
SEMANTIC ARTS: Designing and Building Business Ontologies (#KnowledgeGraph)
-
Projects
-
Interactive Narrative
-
For Burroughs and Chapin and their H.L. Hunley interactive museum, we produced digital elements for an interactive narrative between on-screen characters and physical animatronic tour guide.
-
Jungle Jim's Nutty Puzzles
-
For a national contest "Hidden Agenda Stealth Education", we produced "Jungle Jim's Nutty Puzzles", a fun game of constructing simple machines from basic shapes in hopes of using the laws of physics to guide a ball to a specific end position. The judges included Richard Garriott (aka Lord British, of Ultima fame), and we were awarded national finalist as runner up.
Honors & Awards
-
National Finalist
Hidden Agenda Stealth Education
Edutainment competition... "Jungle Jim's Nutty Puzzles"...
-
NC Representative
Youth Technology Summit - Austin, Texas
-
Blue Diamond Award
Charlotte, NC Chamber of Commerce
Outstanding Contributor for the TechConnect program at local High Schools.
Other similar profiles
Explore more posts
Explore top content on LinkedIn
Find curated posts and insights for relevant topics all in one place.
View top contentOthers named Mark Spivey in United States
65 others named Mark Spivey in United States are on LinkedIn
See others named Mark Spivey