Jeff Frick

I am not telling you something to sell you something. I am sharing something to entertain, educate, and get you to stop and think for a minute. | Top Voice | Friday Five | Turn the Lens | Work 20XX

With over 30 years of experience in the media industry, I am a podcast host and a media entrepreneur who is passionate about the future of work, leadership, and innovation. I have founded and led Menlo Creek Media, a company that helps leaders, marketers, and communications professionals navigate and leverage an AI-driven asynchronous world, by offering executive training for podcast hosts and guests, content creation and media library curation and amplification programs, and community development programs.

I also host two podcasts, Work 20XX and Turn the Lens, where I facilitate the spread of best practices, insights, learning, and wisdom on topics that matter to the modern workforce and society. I aim to be a trusted source of information and inspiration for my listeners, by featuring smart and diverse guests, engaging in lively and thoughtful conversations, and sharing nuggets of knowledge. My goal is to help people get more impact with their communications and content investments and to empower them to thrive in a time of rapid change and uncertainty.

Founder & Principal

Please visit www.MenloCreek.com for a more comprehensive list of services and resources. 

Helping Leaders, Marketers, and Communications professionals navigate and leverage an AI driver asynchronous world. 

Host of 'Work 20XX'  with Jeff Frick show on the Future of Work 
Host of 'Turn the Lens' with Jeff Frick on Leadership and of other topics of interest.

Manipulation is about negotiating with the world. Whereas a lot of robotics, there's not much negotiation. If you're in its way, you better hope you're stronger. - Jeremy Fishel

Compliance is critical because when you bump into something you want time to react to that information. - Jeremy Fishel
-----
Collisions. 
It's robot industry vocabulary for bumping into things.

Feeling isn't only about control, 
it's also about reacting to the world, in time to prevent harm.

Jeremy Fishel: Nature's Best Manipulator, Man's Best Controller: Hands | Turn the Lens with Jeff Frick

Episode page with extensive show notes, transcript, links and references 
https://lnkd.in/gN9spBJb 
YouTube - https://lnkd.in/gdgMY8ie 
Spotify - https://lnkd.in/g9G8gbkM 
Apple Podcast - https://lnkd.in/gNZrG7tV
or where ever you podcast 

For more on Humanoids Summit, including May 2026 Summit in Tokyo visit 
https://lnkd.in/ga487tGq 

For more from Humanoids Summit SV 
https://lnkd.in/gyxXZcVT 

#Robotics #HumanoidRobots #Compliance #RoboticManipulation #EmbodiedAI #FutureOfWork #RobotSafety #AIRobotics #Innovation #HumanoidsSummit #TurnTheLens #EdColgate #HANDERC #DeepTech #TurntheLens #RoboticsEngineering #WorkplaceTechnology #SanctuaryAI #HumanoidsSummit #Robots #Hands

When you remove someone's sense of touch, their capabilities drop off tremendously. - Joe Michaels

Bringing that back into the robot equation and giving the human the feeling of what that robot is doing and creating a closed loop. - Joe Michaels

It makes a very natural and powerful control system. - Joe Michaels
--
Humans and machines working together for an enhanced version of both.

Teleoperation, for literal operation as well as fine grain training.

Feedback is an essential component to control, whether teleoperating from a distance in dangerous places, or rehearsing high-risk situations in the safety of digital worlds and simulation environments.

And yes, Sigourney Weaver and her P-5000 Powered Work Loader Exoskeleton from Alien immediately jumps to mind.

Joe Michaels: Teleoperation, Controlling Complex Robots by Feel | Turn the Lens with Jeff Frick
Episode page with transcript, show notes, links, and references
https://lnkd.in/eYVnUAdT
YouTube - https://lnkd.in/g8TytYCb
Spotify - https://lnkd.in/gvfDZsnX
Apple Podcasts - https://lnkd.in/gA4UeFSy
or wherever you enjoy your podcasts.

Released in collaboration with Humanoids Summit. Humanoids Summit is sponsored and produced by ALM Ventures.
For more on Humanoids Summit, including the May 2026 Summit in Tokyo, visit
https://lnkd.in/ga487tGq

For more from Humanoids Summit SV 2025 -
https://lnkd.in/gA4XmZiE

#teleoperation #touch #feel #feedback #training #remoteoperation #finegrain #1HMX #humanoidssummit #turnthelens 1HMX #robots #robotics #Robotics #HumanRobot #HapticFeedback #Dexterity #FutureOfWork #Teleoperation #RoboticsEngineering #AI #Automation #PhysicalAI #HumanMachineInterface #Innovation #EmbodiedAI #Work20XX

Visiting Pakistan, I was seven, the first time I met someone missing a limb. She was living in poverty in the streets, missing her right leg and using a tree branch as a crutch. - Aadeel Akhtar, PhD

That's what inspired me to go into this field. - Aadeel Akhtar
--
Moment of impression.

Call it a calling, an obsession, a mission.
In a world that feel anything but certain or stable,
some experience a moment of clarity that sets a course that will not be denied. 

Very helpful if you want to be an entrepreneur. This was the ninth version of Aadeel's hand. Some lessons can only be learned by doing, not drawing.

Transforming Lives and Robotics: Jeff Frick talks with Aadeel Akhtar, PhD, PSYONIC - via Humanoids Summit
https://lnkd.in/gjw5GeSV
For more on Humanoids Summit
https://lnkd.in/ga487tGq
#Robotics #BionicLimbs #Prosthetics #FutureOfWork #HealthTech #MedTech #Innovation #SocialImpact #Entrepreneurship #Founders #PurposeDriven #HumanAugmentation #EmergingTech #TechForGood #Robots #Hands #InclusiveTech #Humanoids #HumanoidsSummit #TurnTheLens #PSYONIC

For robots, we have a term in the industry. My robot 'collided.' — Bernt Børnich

Everything we do through our life is a collision. When I pick up this bottle, I'm colliding with it. If I put my hand on this table, I'm colliding with the table. If I take a step, I'm colliding with the ground. — Bernt Børnich

Humans don't collide. We just interact with the world.
I can do this in a very stable manner, very low impact, not dangerous to myself or the environment. — Bernt Børnich
----
High energy with gears and ratios.
or
Low energy that moves slowly and has time to
react before something significant happens.

Bernt and the team worked 10 years on the motors to bring low-energy capabilities to their humanoids.

Safety first. That includes 'collisions,' or as we say in the human world, bumping into things.

How 1X is Changing the Game: Jeff Frick Chats with Bernt Børnich, 1X
Humanoids Summit YouTube Channel
https://lnkd.in/g8jYtFjb

Who's going to Tokyo? 
https://lnkd.in/ga487tGq

#Robotics #HumanoidRobots #RobotSafety #FutureOfWork #Automation #AIHardware #Cobots #RoboticsDevelopment #EmergingTech #EnterpriseRobotics #ManufacturingTech #HumanRobotInteraction #TechInnovation #IndustrialRobotics #SafetyFirst #1X #HumanoidsSummit #TurnTheLens #Robots #Humanoids

Pursuing vertical integration for its own sake is actually to be a bit misguided. Vertical integration should always be in service of either making the best thing or actually optimizing in the short or the long term for iteration speed. - Evan Wineland

We're able to control that much more of the end user experience. - Evan Wineland 

We only do it as much as is necessary to make the thing right. - Evan Wineland 
----------------------
Vertical Integration

Guess I shouldn't be surprised.
Both Evan and Kaan Dogrusoz, his co-Founder at Weave Robotics were most recently at Apple.

But the specific qualifier, 'in service of iteration speed' 
is notable and a little counter intuitive.

Seems logical that off the shelf components from multiple suppliers would develop faster as a group. And Evan qualified, they only do as much as necessary. And necessary is that which interacts with the objective, and the environment. The OS, the grippers, and more, they write and develop from scratch to increase the iteration speed to accomplish their list of task objectives. A large part of at this stage, folding laundry.

Evan Wineland: Deploying is the Point. Affordable Robots that Work | Turn the Lens with Jeff Frick 
Episode page with transcript, show notes, links, and references https://lnkd.in/gZnpAQSc 
YouTube - https://lnkd.in/gb-HbVxZ  
Spotify - https://lnkd.in/gHtgWHYw 
Apple Podcasts - https://lnkd.in/gk3Mnzja 
or wherever you enjoy your podcasts.

Release in collaboration with Humanoids Summit. Humanoids Summit is sponsored and produced by ALM Ventures. 

For more on Humanoids Summit, including May 2026 Summit in Tokyo visit 
https://lnkd.in/ga487tGq 

For more from Humanoids Summit SV 2025 - 
https://lnkd.in/gHDGNhdX 

#VerticalIntegration #ProductStrategy #TechLeadership #Innovation #FutureOfWork #IterationSpeed #UserExperience #StartupStrategy #EnterpriseTech #WeaveRobotics #HumanoidsSummit #WorkTwentyXX #TurnTheLens #ProductDevelopment #TechInnovation #BuildingProduct #FounderMindset

People have been working on robotic hands gosh, for 40, 50 years, really as long as we've been thinking about robots. What's new? AI. - Ed Colgate

The ability to take something so complicated as this hand and actually do something useful with it. That wouldn't be possible without AI. - Ed Colgate
--
AI, GenAI, Foundation Models, Transformers, and the relentless pace of better, faster, cheaper hardware (from GPU to mechanical parts) means robotic hands finally have the potential to mirror the performance of the most dexterous manipulator in nature, the human hand.

Ed Colgate and the extended HAND ERC community are building the foundational technology for the next decade's technological advances in robotic hands, including sensing, actuation, design, simulation, learning, control, social, legal and industrial studies.

Ed Colgate: Soft Hands, Dexterous Robots | Turn the Lens with Jeff Frick
Transcript and Show Notes https://lnkd.in/grjhZf2y 
YouTube - https://lnkd.in/gsv4UKKs
Spotify - https://lnkd.in/gTCHeBx2 
Apple Podcast - https://lnkd.in/ghkjsVgY 
Substack - 
or wherever you podcast 

HAND ERC 
The National Science Foundation 'Human AugmentatioN via Dexterity' Engineering Research Center (HAND ERC)
https://hand-erc.org/

Released in collaboration with Humanoids Summit. Humanoids Summit is sponsored and produced by ALM Ventures. 
For more on Humanoids Summit, including May 2026 Summit in Tokyo visit 
https://lnkd.in/ga487tGq 

For more from Humanoids Summit SV 2025 - 
https://lnkd.in/gPcYu2Xt 

#Robotics #Robots #PhysicalAI #Hand #Hands #RoboticHand #HANDERC #ProstheticHand #Humanoids #HumanoidsSummit #TurntheLens #AI #GenAI #EmbodiedAI #RoboticHands #AIInnovation #FutureOfWork #ManufacturingTech #Automation #HumanoidRobots #MachineLearning #RoboticsEngineering #DeepTech #AIApplications #IndustrialRobotics #EmergingTech #TechLeadership

Business Data Products | Data Product Storytelling | Data Product Pyramid | Data Product Marketplace

Ask me about delivering e2e AI native business processes using the Nebulyx AI platform and our A.D.O.P.T. framework - AI + Data Object Graphs and Data Product Thinking. Using our Data Product Pyramid framework on our AI Digital Twin solution and generative AI, AI Agents, and advanced analytics with Data Object Graphs to deliver the “Composable Enterprise”: a low friction, adaptive eco-system of re-configurable data products that runs at the speed of the business.

This includes 
1- Using LLMs/AI to create advanced analytical and Data Products in reconfigurable Data Object Graph based eco-system
2- Creating a Lingua-Franca that allows business teams to communicate analytics asks with data and tech teams.
3- Creating an Adaptive Business Strategy - The business strategy can start small and evolve and adapt as the business changes.
4- Implementing as a rapid delivery, highly valuable data and analytic use-cases as Data Products
5- Creating A portfolio of actual and specific real world Data Products that can used and re-used across the business.
6- Creating a Data Product Marketplace where internal and external customers can create, prototype, search and use data products, including: Datasets, ML Models, BI Dashboards, Graph analytics
7- Creating hybrid cloud polyglot data infrastructure to deploy Data Products

Jon Cooke

AI Shortcuts the Ontology Tax

Saturday reflections. 🤔

Ontologies and knowledge graphs are so hot right now and I'm always keen to read good posts about the realities of implementing them.

As such, I was intrigued to read Alexander Shereshevsky’s piece -“The Ontology Tax: What Nobody Tells You About the Real Cost of Knowledge Graphs” - (link in comments).

It's operner is about the tax most don’t see coming.

His headline: Typical cost, between $500K-$1.5M+ over five years for a single domain ontology. 

Months of expert work before usage with 15-25% annual maintenance. 

With governance adding  30-50% 

He says only 27% of orgs get knowledge graphs to production despite a $6.9B market.

The scale is staggering. The SNOMED CT initiative- has approx. 350,000+ clinical concepts and requires an entire non-profit organisation just to maintain it with monthly releases. 

Semantics are super important for accurate and robust information based decisioning systems, especially if one is using AI.

But with traditional manual approaches, typically there is this massive effort needed before any use-cases can be even started and also to get up-to-date. 
 
It’s interesting in my work building these types of systems I have found the same thing. 

Spending weeks trying to classify domain knowledge, build taxonomies and align definitions. 

Every time I added to the graph it became more brittle. 

A compliance process step isn’t an obligation or prohibition - the ontology had no box for it, so I spent ages trying to shoehorn into compromise approaches.

This is what led me to creating a new type of small language model (now Nebulyx Cognition) that discovers semantic structure directly from language. 

It learns that “shall report all transactions” and “clinicians should offer CBT” share the same shape - someone must do something w/ a degree of authority. 

A really cool part is that business users can adjust, add and maintain through simple user interfaces. I.e. we don't require armies of knowledge specialists.

But the grunt work is done by the model i.e. encode 70-80% automatically and directly  from source material.

When one queries e.g. “what are the medical procedures for x” - the actual text and location form the medical book is returned, every time. 

We can constantly and quickly update as new information comes in. A change in procedure or policy a new regulation.

Plus, link and do differences between text by querying the AI - which is very tough to do with traditional methods.

All within minutes. 

Knowledge compounds through this agile flywheel effect rather than drifting despite it.

Semantic graphs (including Ontologies and Knowledge Graphs) are awesome and needed in an AI world, but we shouldn’t have to pay a large tax to adopt them 

Happy Saturday.

Nebulyx AI

Nebulyx AI Studio - AI Digital Twins in minutes, Nebulyx Cognition - 100% Deterministic Small Language Model, Nebulyx Agent DOG - Agents For Regulatory Sectors | Data Product Workshop Podcast host

AI Closes the Intent -> Knowledge -> Execution Gap
 
There's a gap between what an organisation intends & what it actually executes.

In that gap sits knowledge.

How we handle it determines whether we move fast or grind to a halt.

The traditional approach treats knowledge as something you define upfront.

Build the database schema. Define the service contracts. Model the Knowledge Graph.

Then use it.

But this often breaks on first contact with reality.

Business can't always articulate what they need - they ask for what they want. 

Even with robust upfront capture (requirements, questions etc), the thinking breaks on usage.

Proved in the start-up world.

Encoding for change is brutally hard.

Database schemas evolve slowly - migrations, versioning, careful modelling.

Service contracts version too - you can run v1 and v2 simultaneously, but the change cycle is long.

Knowledge graphs become brittle the moment the information/usage shifts.

Meanwhile, the actual knowledge is constantly moving.

- A new regulation.
- A legal judgement changes interpretation.
- An email clarifies intent.
- A decision (human or agent) creates new context.
- Document versions override previous guidance.

Isn't edge-case complexity, it's the normal cycle of organisational knowledge.

Add to this: business silos operate at different speeds with different perspectives.

What compliance sees isn't what operations sees.

What's urgent for legal is invisible to product.

Organisations are networks of discrete processes, not unified data models.

What's the alternative?

- Knowledge as a flywheel, not a foundation.
- Intent flows forward into execution.
- Execution feeds back into knowledge.
- Knowledge refines intent and the cycle continues.

The flywheel spins in both directions - encoding what we learn, informing what we do next.

Nebulyx Cognition does this. It's our small language model that discovers semantic structure directly from source material.

- 70-80% encoded automatically. 
- SMEs tune the 20% that needs human judgement.
- No months of upfront work before value appears.

But encoding is only half the story.

AgentDOG - our agent architecture - tracks execution state as a Data Object Graph.

Every decision, every handoff, every outcome gets captured. Context accumulates.

The graph grows through usage, not in isolation from it.

When an agent executes a process, the results feed back. New patterns emerge and knowledge compounds.

This means:

- Evolving knowledge structures through usage
- Temporality built in - versions, overrides, effective dates
- Speed matches the domain - compliance moves monthly, operations moves hourly
- Capture at the edges - decisions, context, outcomes - not just source documents

A living system that moves at the speed of the business.

Intent → Capture → Learn → Encode → Execute → Intent ...

The flywheel keeps spinning.

Happy Friday.

Nebulyx AI

AI Closes the Intent - Knowledge - Execution Gap


AI Native Semantics: A Tail of Two Journeys

Let me tell you about two companies.

Company A signs a $1M plus semantic layer initiative.

18 months to model their domains, mapping relationships, defining contracts. 

Governance committees reviewing every entity.

By Q3 next year, they’ll have a foundation for AI.

Then they spin up another initiative to build the AI solution to query the semantics.

Company B is in the same space.

They point an AI native semantics system at Harrison’s Principles of Internal Medicine. 

5,113 pages. 

One of the most comprehensive clinical references ever published.

30 minutes later:

∙ 122,312 sentences encoded
∙ 7,916 clinical concepts identified
∙ 22,224 domain terms discovered

All in a determinstic AI model that can be queried immediately.

Doctors look over, check and tweak the semantics.

Then they ask: “How should a clinician manage chest pain?”

It returned relevant guidance - cited, traceable, grounded in source.

Not because someone modelled “chest pain” into a hierarchy. Because the language itself carries the semantics.

The questions Company A should be asking:

What will Company B look like in three years?

When we finish our ontology programme, will we have to unwind it to compete?

Is this one transformation - or two?

Is this a multi-year initiative costing millions, or a immediate value approach with a fraction of the cost?

Two paths to semantics:

Model upfront. Define before you use. Years/Months and million plus in cost.

Or

Use AI to discover through usage. Let it find and encode the structure. Hours and iteration.

Both get you semantics.

Only one gets you there at the speed your market is about to demand.

Happy Saturday.

Nebulyx AI

AI Native Semantics: A Tail of Two Journeys

Your Business Works in Silos. Your AI Should Reason Across Them.

Your compliance team calls it an "obligation."
Your operations team calls it a "procedure."
Your legal team calls it a "requirement."

All talking about the same thing.

This is the knowledge problem all organisations grapple with.

The issue isn't that the information doesn't exist. It's that every domain has encoded it in its own language, its own structure, its own context.

And when you need to reason across those boundaries - "are we actually compliant?" - you're relying on someone who knows both worlds well enough to join the dots manually.

That person is usually expensive, overworked & about to retire.

In regulated industries:

Compliance builds a register of obligations mapped to frameworks like MiFID or SOX. Operations writes procedures that implement those obligations. 

Legal interprets the regulation differently again, with caveats & jurisdictional nuances. 

And finance tracks the cost of all of it using a completely different taxonomy.

Four domains but same underlying reality but four different ways of describing their view on it. 

And those view are different i.e. the semantics are not the same across them.  

The traditional answer: build an ontology, map everything to a common model & get everyone to agree on the vocabulary.

We all know what happens next.

Eighteen months of workshops. A beautiful model that's out of date before it's finished. And a business still asking: "If this regulation changes tomorrow, what actually breaks?"

This keeps failing because we're trying to solve a reasoning problem with a modelling exercise.

What if instead of forcing every domain into one shared schema, you had an AI layer that understood each domain in its own terms & reasoned across them at query time?

That's what we've built with Nebulyx Cognition.

Cognition doesn't just store triples. It encodes the actual semantics of language - the full meaning, not just subject-predicate-object.

When compliance says "obligation" & operations says "procedure" & legal says "requirement," the model understands the relationship between them without anyone manually mapping it.

You ingest the regulation, the operating procedures, the legal interpretations.

Each stays in its own domain context with the reasoning layer sitting across all of them.

So when you ask "what's my exposure if this regulation changes?" - it doesn't search one silo. 

It reasons across compliance, operations & legal simultaneously.

Full traceability back to source document, paragraph & sentence.

No six-month alignment project. No master ontology that breaks when someone adds a new framework. No reliance on the one person who holds it all in their head.

The knowledge already exists in your organisation - policies, standards, contracts, procedures & emails.

The problem was never capturing it or even modelling it. 

The problem was reasoning across it.

Happy Thursday

Nebulyx AI

AI Driven Semantics: Real World Results

Following my post on Saturday about the Ontology Tax, I had some push-back (some constructive - love these, but some just wanted to rant - which I don't have time for).

A lot of the genuine challenges were around whether the claims hold up in practice.

So I did a use-case this weekend.

I pointed the system at Harrison's Principles of Internal Medicine. 5,113 pages. 

One of the most comprehensive clinical textbooks ever published.

Not a curated dataset, a real medical reference that clinicians train on.

The results:
→ 16.9 million characters extracted with page-level tracking 
→ 122,312 sentences encoded and structurally understood 
→ 7,916 clinical concepts identified 
→ 22,224 domain terms discovered automatically

Processing: 20-30 minutes, end-to-end.

Then I asked it: "How should a clinician manage a patient presenting with chest pain?"

The relevant information returned from Harrison's, including:

"In patients with acute chest discomfort, the clinician must first assess the patient's respiratory and hemodynamic status"

Plus guidance on differentiating life-threatening conditions from chronic presentations, all cited and all traceable to specific sections. 

Real semantic results, all without hallucination and no hand built ontology.

That last point is what I think people misunderstood from Saturday.

We're not building ontologies faster, or at all. 

We're building AI native semantics.

The system discovers semantic structure directly from the language itself. 

It understands that a sentence carries an obligation, a recommendation, or a prohibition. 

Not because someone mapped those concepts into a taxonomy, but because the language tells you.

"Shall report all transactions" and "clinicians should offer CBT" share the same linguistic shape. 

Someone must do something, with a degree of authority. 

The system recognises that natively and indexes it with the source, page, section and context.

SMEs then go in and adjust. We are building business-friendly UIs for this now.

Feed it documents, SME tweaks then ask questions, that give accurate results with citations and traceable answers. 

All in hours, not months.

That's not a cheaper path to an ontology. It's a different path entirely.

And when source material changes? Process it again the knowledge stays current without a maintenance team.

Is it perfect i.e. zero-shot? No. The system gets sharper with more training and SME input.

But organisations can now use AI to get rich semantic knowledge bases in days, not months.

You don't pay a huge Ontology Tax. You, all but, remove it.

Happy Monday

Nebulyx AI

Becoming AI-Native: Scale - The AI Factory!

𝗣𝗿𝗼𝘃𝗲 ✓ | 𝗖𝗮𝗽𝘁𝘂𝗿𝗲 ✓ | 𝗗𝗲𝗽𝗹𝗼𝘆 ✓ | 𝗟𝗲𝗮𝗿𝗻 ✓ | 𝗦𝗰𝗮𝗹𝗲 ← today

This week: Prove. Capture. Deploy. Learn. Today: what happens when you do it again and again.

The first use-case is the investment. 
You prove it works with real people. 
You capture the knowledge. 
You deploy governed agents. 
The flywheel starts turning.

Now you point it at a second use-case.

The models carry over. 
The agent framework is reusable. 
The governance architecture is already in place. 
The proving methodology - real UIs, real people, real edge cases - it's a repeatable process now, not a one-off.

Second use-case is faster, cheaper and with less friction. 

Because you're not starting from scratch. You're extending a platform that already knows how to learn.

Third use-case scales even faster.

This is what compound value actually looks like. 

Not just AI getting smarter within one use case, the entire capability of deploying AI across your organisation accelerating with each iteration.

The proving stage that took 4 weeks the first time? 
Takes 2 the second time. 

The knowledge capture that felt like heavy lifting? 

The system already knows the semantic structure - it just needs the new documents. 

The agents? Same governance framework, new decision space.

And the economics work differently too.

Sam Jacobs wrote about why agentic AI economics are broken, every query costs real money, margins compress, vendor lock-in deepens. 
(Link in comments)

He's right, for most architectures.

Nebulyx Cognition runs on your infrastructure or ours. 

- Small Language Models (Sub-1GB). 
- Fixed cost. 
- No API bills. 
- No per-query metering. 
- The marginal cost of the next query is effectively zero. 
- Data sovereignty by default.
- Fraction of a second response time (even on CPU)

The cost of your second domain isn't "first domain plus more API bills." 

It's a fraction of the first and the third is a fraction of that.

This is the full platform: 
- Nebulyx AI Studio to prove it. 
- Nebulyx Cognition to reason over it. 
- Nebulyx AgentDOG to execute. 

All governed the right way for regulated industries.

All compounding.

Prove → Capture → Deploy → Learn → Scale.

Not a transformation programme that ends with a report. 

An organisation that teaches itself, continuously, at predictable cost, with full provenance on every decision.

This week, I laid out the path from sandbox to becoming AI-native. 

If you've been following along and thinking "this is the gap we're stuck in"  that's exactly the conversation we love to have with you.

One use case with working solution in weeks. Everything compounds from there.

🔗 Nebulyx AI

Becoming AI Native: Scale - The AI Factory

Becoming AI-Native: Learn With Each New Decision

Built on yesterday's business. Already out of date → The org learns by doing.
𝗣𝗿𝗼𝘃𝗲 ✓ | 𝗖𝗮𝗽𝘁𝘂𝗿𝗲 ✓ | 𝗗𝗲𝗽𝗹𝗼𝘆 ✓ | 𝗟𝗲𝗮𝗿𝗻 ← today | 𝗦𝗰𝗮𝗹𝗲

Monday: proved it. 
Tuesday: captured the knowledge. 
Wednesday: deployed governed agents. 
Today: the flywheel.

Most organisations get here and call it done. Workflows automated, agents deployed. 

Transformation complete.

Except the model was trained on yesterday's business.

It doesn't know about last quarter's policy change. The new product line. 

The regulation that updated three weeks ago. 

The edge case your most experienced person handles on instinct every Tuesday morning.

And the traditional fix? A knowledge management project. 

Months of interviews. Documentation that's out of date before it's published. 

Expertise that nobody can articulate because it lives in reflexes, not rules.

This is the problem nobody talks about: static AI decays from the moment it goes live.

We built a different loop.

Every time an expert overrides an AgentDOG agent, every edge case caught, every judgment call made, it's captured with full context. 

Not filed away as a correction. 

Fed back into the system as structured knowledge.

The agent's original reasoning. The state it was in, the override, the outcome. 

All connected.

The system doesn't just record what happened. It learns what should happen next time.

Tacit knowledge, the stuff that only lives in people's heads, enters the system through action, not interviews.

Policy versions are tracked, you can compare what the agent would have done last month versus today. 

You can audit every change. If a new version performs worse, you roll back.

This is the flywheel: knowledge goes in, agents execute, experts refine, the system improves, continuously. 

Not a project with a start and end date. An operating characteristic of the business.

Your best people's expertise doesn't walk out the door any more. It compounds.

Value unlocked: an organisation that gets smarter with every decision made, continuously, not in annual reviews.

Tomorrow: Becoming AI-Native, Scale It. And Why the Economics Actually Work.

🔗 Want AI that learns from your people without losing control? Nebulyx AI

Becoming AI Native: Learn With Each New Decision

AI Native Semantics: Avoiding Ontology Loss

Ontologies can be lossy artefacts, as I described in a thread in my post yesterday, which outlined a different approach to create an AI native semantic reasoning knowledge base.

In response, Anna Bergevin asked a great question: what are you giving up by skipping the [formal] ontology step?

The answer: less than you think.

The problem with traditional knowledge engineering

It happens at two levels.

First, the ontology - the schema layer. Experts define classes, properties, hierarchies upfront. 

This can take months and can't anticipate every future question (even with rigour e.g. pre-interviewing).

Second, with the knowledge graph - the instance layer, engineers extract facts and encode them as triples: subject-predicate-object.

Both steps can lose information.

When a regulation says "shall report transactions where value exceeds €1M, except intraday transactions"

A knowledge engineer might capture "must report transactions" and separately note "intraday exempt.".

But this loses the structure. One is an exception to the other, qualified by a condition.

Both are interpretations and model transformations based on human interpretation, with what it known at the time.

Which can result in these failure modes:

Incomplete - relationships lost in human interpretation

Brittle - document changes require manual re-curation

Expensive - time and effort to assess and hand-craft every update

But the  thing is the language already encodes the relationships.

The text captures relationships through syntax and grammar.

"Unless" creates exceptions. "Subject to" introduces conditions. "Amended as follows" signals modification.

These patterns aren't domain-specific, they work the same in regulations, clinical protocols, or recipes.

A different approach

We preserve the full linguistic structure and learn how meaning flows through it.

Sentences retain complete structure - not reduced to triples.

Relationships detected from language itself - "if", "except", "pursuant to" signal how sentences connect.

Cross-document links emerge automatically.

At query time

Ask about reporting obligations and you get the core requirement plus its conditions, exceptions and consequences.

When source documents change, relationships update automatically, without heavy re-curation process.

The ontology and knowledge graph aren't totally gone. They're reimagined - where deep semantics are discovered and encoded from the language itself.

It's the difference in agility, depth and months/years of effort for change.
 
Happy Sunday.

Nebulyx AI

AI Native Semantics: Avoiding the Ontology Loss

Libraries to Use Cases: Four Stages of Knowledge to Value

Every organisation has more knowledge than it can use - API documentation, operating procedures, maintenance schedules, regulatory frameworks, and a huge amount locked in the heads of people who've been running things for decades.

But the question isn't: "how to curate knowledge" - the librarian mindset. 

It's: "what's the path to drive the actual business?" - the value mindset.

Here's how we think of that journey.

𝗦𝘁𝗮𝗴𝗲 𝟭: 𝗞𝗻𝗼𝘄𝗹𝗲𝗱𝗴𝗲 𝗖𝗮𝗽𝘁𝘂𝗿𝗲

A rail operator has system documentation describing every asset, route & depot, maintenance procedures, and engineers who know the schedule says one thing but wet Tuesdays in November actually work differently. 

All of that is knowledge - structured & unstructured, formal & tacit.

We encode it into our semantic AI (Nebulyx Cognition): system specs, procedures, interviews - extracting concepts, relationships & structure. 

A working model that experts check, refine & build upon. 

We're not capturing questions, just what the organisation knows.

𝗦𝘁𝗮𝗴𝗲 𝟮: 𝗨𝘀𝗲 𝗖𝗮𝘀𝗲𝘀 𝗘𝗺𝗲𝗿𝗴𝗲

Once captured, we query the knowledge with the real questions they've been trying to answer for years. 

"Which track sections maintained by contractor X are on routes affected by this timetable change?" - that spans asset, procurement & operations domains.

𝗦𝘁𝗮𝗴𝗲 𝟯: 𝗠𝗼𝗱𝗲𝗹 𝗮𝘀 𝗗𝗮𝘁𝗮 𝗢𝗯𝗷𝗲𝗰𝘁 𝗚𝗿𝗮𝗽𝗵𝘀

Next we extract and model use cases as executable processes (Data Object Graphs in Nebulyx AI Studio), bringing in live data - signalling feeds, track sensor readings, maintenance logs.

The knowledge model told you what an asset is & how it relates to a route. 

The Data Object Graph shows how it actually works in your business. 

The operational data tells you three sections of track are showing accelerated wear on the same corridor & nobody connected it to a drainage change six months ago.

Knowledge becomes insight - the model stops being a reference library & starts being a lens on what's actually happening.

𝗦𝘁𝗮𝗴𝗲 𝟰: 𝗦𝗰𝗲𝗻𝗮𝗿𝗶𝗼 𝗔𝗻𝗮𝗹𝘆𝘀𝗶𝘀 & 𝗣𝗹𝗮𝗻𝗻𝗶𝗻𝗴

- What happens if we close that section for six months?
- Which services reroute & what's the knock-on?
- If we accelerate renewal on corridor B, what maintenance can we defer on corridor C?

Not describing what you know - simulating what could happen. 

The Data Object Graph gives you this: testing decisions before you make them.

That's the shift from knowledge as an abstract library to knowledge as an operational capability.

𝘛𝘩𝘪𝘴 𝘪𝘴𝘯'𝘵 𝘴𝘦𝘲𝘶𝘦𝘯𝘵𝘪𝘢𝘭. 𝘐𝘵'𝘴 𝘪𝘵𝘦𝘳𝘢𝘵𝘪𝘷𝘦.

This is Nebulyx AI:

- Knowledge drives the use case.
- Data Object Graphs create the process.
- Operational data turns insight into action.
- Scenario analysis turns action into strategy.

Knowledge is only valuable when it drives the business. 

Getting out of the library mindset to creating real business impact.

Happy Monday.

Nebulyx AI

Libraries to Use Cases: Four Stages of Knowledge To Value

taplio

The Battle

Jeff Frick

Jon Cooke

Followers

Posts

Engagements

Likes

Performance ScoreJeff Frick's LinkedIn Performance

Performance ScoreJon Cooke's LinkedIn Performance

Ready to grow your LinkedIn presence?

taplio

The Battle

Jeff Frick

Jon Cooke

Followers

Posts

Engagements

Likes

Performance Score.css-goub6s{font-size:var(--chakra-fontSizes-md);font-weight:500;color:var(--chakra-colors-gray-600);display:block;margin-top:var(--chakra-space-1);}Jeff Frick's LinkedIn Performance

Performance ScoreJon Cooke's LinkedIn Performance

Ready to grow your LinkedIn presence?

Performance ScoreJeff Frick's LinkedIn Performance