Invisible Machines podcast by UX Magazine

There’s a particular kind of exhaustion that hits when you’re building with AI. You’ve got the license. You’ve got the pilot. You’ve got a chat window that generates whatever you ask for. But then comes the wall—the workflow that almost works, the dependency that loops back on itself, victories that only come when you quietly move the goalposts. In this episode, we talk about how to get around the wall.

Evan J. Schwartz, Chief Innovation Officer at AMCS Group and an adjunct professor teaching AI stewardship, argues the real shift isn’t “AI does the work.” It’s that humans stop owning the full vertical of doing and start owning outcomes—while AI takes more of the output. That sounds abstract until he gets specific: if you ask a model for a feature before you describe actors, containment envelopes, and dependency policy, you’ll get something that passes the demo and fights you every time you try to change it.

Josh Tyson and Robb Wilson push on the organizational version of the same mistake—automating fifty-five use cases exactly as humans do them today, then acting surprised when nothing compounds. Evan’s answer is a two-step path that respects how messy humans actually are: first compress low-value work into AI for a confidence-building lift; then rethink the whole iterative cycle for asymmetric returns.

Also, learn why “no change without pain” isn’t cynicism, why POCs love moving goalposts, why big enterprises optimize one segment and break the connective tissue upstream, and why the scarcest resource won’t be another prompt library—it’ll be people who can think about constraints.

We cover: AI stewardship and the inverse-pyramid skill set, outcome vs output, structure-before-feature development, the two-step adoption path, risk-averse vs innovator cultures, SMB speed vs mid-market discipline vs enterprise battleships, critical vs non-critical systems, and why cutting headcount first is how you donate your newly trained stewards to your competitor.

Guest: Evan J. Schwartz, Chief Innovation Officer, AMCS Group; Adjunct Professor, Jacksonville University

Hosts: Josh Tyson, Robb Wilson

--------- Support our show by supporting our sponsors

This episode is supported by OneReach.ai

Forged over a decade of RD and proven in 10,000+ deployments, OneReach.ai's GSX is the first complete AI agent runtime environment (circa 2019 — a hardened AI agent architecture for enterprise control and scale.

Backed by UC Berkeley, recognized by Gartner, and trusted across highly regulated industries, including healthcare, finance, government and telecommunications.

A complete system for accelerating AI adoption — design, train, test, deploy, monitor, and orchestrate AI agents.

Use any AI models
Build and deploy intelligent agents fast
Create guardrails for organizational alignment
Enterprise-grade security and governance

Get in touch:
https://onereach.ai/contact/?utm_source=soundcloud&utm_medium=social&utm_campaign=s7e14&utm_content=1

--------- The revised and significantly updated second edition of our bestselling book about succeeding with AI agents, Age of Invisible Machines, is available everywhere: Amazon — https://bit.ly/4hwX0a5

#InvisibleMachines
#Podcast
#TechPodcast
#AIPodcast
#AI
#AgenticAI
#AIAgents
#AIAdoption
#Leadership
#OrganizationalDesign
#CoCreation
#FutureOfWork

Most organizations, and most conversations about AI, default to triangles: who's on top and who's below? Are we the boss of the machines or is it the other way around? Matthew Barzun argues the comfortable shape is a trap and that bottom-up is still a pyramid. The alternative is the constellation: stand out as yourself, connect to other stars, and build something none of you could alone.

In this episode, Matthew Barzun joins Josh Tyson and Robb Wilson to connect diplomacy, design, and agentic AI through constellation thinking. Kentucky doesn't report to Washington. Berkeley doesn't report to Sacramento. Power isn't finite coal to lord, hoard, or divvy. It's something people make with and through one another when they stop ranking and start using differences as fuel.

Josh and Robb press the AI angle hard. If Karen Hao's Empire of AI is a warning about centralized power, Barzun's book reads like the antidote: the boss-or-overlord trap collapses into a design question of how much agency, when, and with what oversight? Individuals are already co-creating with these tools while companies drag their feet. The organizational job is integration, not compromise.

The episode includes:
Mary Parker Follett's electrician parable (half the house burns, half is unlivable — unless you co-create a third plan);
Grandfather Jacques Barzun on fighting the mechanical;
Vint Cerf's "unreliable" network that became the most reliable ever built;
Dee Hock and Visa's constellation architecture;
The triangle of sadness (fight it out, hug it out, sit it out);
Southwest Airlines externalizing the constraint on a clipboard;
Jimmy Carr on jokes that need an audience to exist;
Ken Burns never sounding like a math teacher; and
Why kids say please and thank you to ChatGPT but never to Siri.

https://matthewbarzun.com/

--------- Support our show by supporting our sponsors

This episode is supported by OneReach.ai

Forged over a decade of RD and proven in 10,000+ deployments, OneReach.ai's GSX is the first complete AI agent runtime environment (circa 2019 — a hardened AI agent architecture for enterprise control and scale.

Backed by UC Berkeley, recognized by Gartner, and trusted across highly regulated industries, including healthcare, finance, government and telecommunications.

A complete system for accelerating AI adoption — design, train, test, deploy, monitor, and orchestrate AI agents.

Use any AI models
Build and deploy intelligent agents fast
Create guardrails for organizational alignment
Enterprise-grade security and governance

Get in touch:
https://onereach.ai/contact/?utm_source=youtube&utm_medium=social&utm_campaign=s7e14&utm_content=1

for SoundCloud: https://onereach.ai/contact/?utm_source=soundcloud&utm_medium=social&utm_campaign=s7e14&utm_content=1

--------- The revised and significantly updated second edition of our bestselling book about succeeding with AI agents, Age of Invisible Machines, is available everywhere: Amazon — https://bit.ly/4hwX0a5

#InvisibleMachines
#Podcast
#TechPodcast
#AIPodcast
#AI
#AgenticAI
#AIAgents
#AIAdoption
#Leadership
#OrganizationalDesign
#CoCreation
#FutureOfWork

Most enterprises are under board pressure to deploy AI agents. Sudhir Hasbe argues the harder shift is upstream: you cannot scale intelligence on missing context—and graph databases are how organizational data becomes knowledge agents can actually reason over.

In this episode, Sudhir joins Josh Tyson and Robb Wilson to map the pathway to organizational AGI (bounded expertise, not omniscient AGI), leaning into feature reduction for token sanity, and explaining why eighty-plus percent of enterprise AI projects fail before the model messes anything up. Graphs emphasize relationships over isolated rows; virtual and native storage let you meet latency where it lives; ontologies plus data plus memory form the backboard for self-learning systems.

Josh and Robb press on cost—when compute exceeds employee spend if agents spin without context—and on agent sprawl: without a shared semantic map, every bot maintains its own partial truth. Sudhir connects customer examples—Walmart's two-million-employee knowledge graph, Quarles & Brady turning unstructured legal corpora into navigable paths—and validates the season's through-line: knowledge before agents, humans included.

The demo: a live walkthrough of The Learning Machine—an agentic system that provides tailored instruction using the OneReach.ai orchestration platform and a Neo4j knowledge model of Roger Forsgren’s Lean Knowledge Management. The system assesses what a user knows and computes a personalized learning path through concepts. Instead of staring at an empty "ask me anything" box, agents can proactively educate from a source-of-truth. Growth Hub career journeys. Canonical ideas with temporal depth. Why vector similarity fails the three-little-pigs test—and why interconnected concepts beat similarity blobs.

Guest: Sudhir Hasbe—Neo4j
Hosts: Josh Tyson, Robb Wilson—Invisible Machines

---------- Support our show by supporting our sponsors!

This episode is supported by OneReach.ai
Forged over a decade of R&D and proven in 10,000+ deployments, OneReach.ai’s GSX is the first complete AI agent runtime environment (circa 2019) — a hardened AI agent architecture for enterprise control and scale.
Backed by UC Berkeley, recognized by Gartner, and trusted across highly regulated industries, including healthcare, finance, government and telecommunications.
A complete system for accelerating AI adoption — design, train, test, deploy, monitor, and orchestrate AI agents.
Use any AI models
Build and deploy intelligent agents fast
Create guardrails for organizational alignment
Enterprise-grade security and governance

Get in touch:
https://onereach.ai/contact/?utm_source=youtube&utm_medium=social&utm_campaign=s7e12&utm_content=1

for SoundCloud:
https://onereach.ai/contact/?utm_source=soundcloud&utm_medium=social&utm_campaign=s7e12&utm_content=1

---------- The revised and significantly updated second edition of our bestselling book about succeeding with AI agents, Age of Invisible Machines, is available everywhere: Amazon — https://bit.ly/4hwX0a5

#AgenticAI #KnowledgeManagement #KnowledgeGraph #Neo4j #EnterpriseAI #AIAgents #OrganizationalAGI #GraphDatabase #InvisibleMachines #AI #FutureOfWork

Everyone wants to talk about agents and models. Jeff McMillan, starts where almost nobody else does: the foundation.

In this episode, Jeff McMillan, founder of McMillanAI, former Head of Firmwide AI at Morgan Stanley, and advisor on enterprise AI, maps AI as a stack: high-quality accessible data → semantic layer (knowledge graphs, RAG) → control and governance → models → orchestration → applications. The heavy lifting is in the bottom layers. Organizations that skip them can fake it for a handful of agents, but at 150 or 15,000 agents, you need near-100% accessibility and 99%-plus quality, or you’re monitoring chaos you can’t see.

Josh and Robb press him on why knowledge management feels unfundable, why tribal institutional knowledge breaks when machines execute without judgment, and why evaluation (golden datasets, custom org evals, regression when models upgrade) is the work builders hate and operators can’t skip. Robb names the trap CTOs are falling into: grinding tokens on feature backlogs that never reach production or revenue. Jeff agrees on the strategic gap — after controlled experimentation, leaders should ask what destroys the business in ten years, not what demo ships next quarter.

The trio also discuss:
Embedded ethics and monitoring, including independent models asking, “Does something smell right?”
Capacity vs. value (30% freed time spent golfing is not ROI)
Process mapping in high-end knowledge businesses that can’t articulate how work moves
Use case zero — knowledge that maintains and teaches itself
Agent-in-the-loop and humans with something to lose in the accountability chain
Jeff’s Board of Advisors experiment at MacmillanAI
AI can make you incredibly smart or comfortably dumb. The choice is cultural, not technical.

Learn more about McMillanAI: https://mcmillanai.com/

---------- Support our show by supporting our sponsors!

This episode is supported by OneReach.ai
Forged over a decade of R&D and proven in 10,000+ deployments, OneReach.ai’s GSX is the first complete AI agent runtime environment (circa 2019) — a hardened AI agent architecture for enterprise control and scale.
Backed by UC Berkeley, recognized by Gartner, and trusted across highly regulated industries, including healthcare, finance, government and telecommunications.
A complete system for accelerating AI adoption — design, train, test, deploy, monitor, and orchestrate AI agents.
Use any AI models
Build and deploy intelligent agents fast
Create guardrails for organizational alignment
Enterprise-grade security and governance
Get in touch:
https://onereach.ai/contact/?utm_source=soundcloud&utm_medium=social&utm_campaign=s7e12&utm_content=1
---------- The revised and significantly updated second edition of our bestselling book about succeeding with AI agents, Age of Invisible Machines, is available everywhere: Amazon — https://bit.ly/4hwX0a5

#InvisibleMachines
#Podcast
#TechPodcast
#AIPodcast
#AI
#AgenticAI
#AIAgents
#EnterpriseAI
#KnowledgeManagement
#AITransformation

Most organizations treat a bigger context window like a cheat code: dump every document in, skip the data work, ship. Jonathan Frankle, Chief AI Scientist at Databricks, says that's still wrong.

This is Jonathan's return visit to Invisible Machines — a conversation recorded last summer, released ahead of Databricks Data + AI Summit. His first appearance (season 2) was the MosaicML-era craft conversation: lottery tickets, mixology, mini-cupcakes. This one is the enterprise engineering thread: be a scientist, curate before you scale, and treat specification (what you actually want the system to do) as the bottleneck between raw model power and useful AI.

Robb and Josh press him on the myths that still seduce enterprise teams: million-token windows as a substitute for real data work, hyperscaler résumés as a proxy for talent, and the fantasy that unlocking every PDF in the org automatically makes knowledge useful. Jonathan's answer is consistent: measure success, test your use case, climb the ladder of techniques, and accept that multimodal is where long context actually earns its keep, not as a universal bypass for curation.

Along the way: the nuclear fusion vs. power lines metaphor; why building a benchmark is a cop-out compared to describing intent; prompts as parameters; chat-only UIs vs. a generation that never wanted buttons; LLM-oriented publishing and static FAQ pages; unlocking PDF at scale when curation gets skipped; early-adopter mistakes we'll laugh at in ten years; and why separating knowledge from reasoning is the north star, even if we aren't there yet.

---------- Support our show by supporting our sponsors!

This episode is supported by OneReach.ai
Forged over a decade of R&D and proven in 10,000+ deployments, OneReach.ai’s GSX is the first complete AI agent runtime environment (circa 2019) — a hardened AI agent architecture for enterprise control and scale.
Backed by UC Berkeley, recognized by Gartner, and trusted across highly regulated industries, including healthcare, finance, government and telecommunications.
A complete system for accelerating AI adoption — design, train, test, deploy, monitor, and orchestrate AI agents.
Use any AI models
Build and deploy intelligent agents fast
Create guardrails for organizational alignment
Enterprise-grade security and governance
Get in touch:
https://onereach.ai/contact/?utm_source=youtube&utm_medium=social&utm_campaign=s7e11&utm_content=1

for SoundCloud:
https://onereach.ai/contact/?utm_source=soundcloud&utm_medium=social&utm_campaign=s7e11&utm_content=1
---------- The revised and significantly updated second edition of our bestselling book about succeeding with AI agents, Age of Invisible Machines, is available everywhere: Amazon — https://bit.ly/4hwX0a5

#InvisibleMachines
#Podcast
#TechPodcast
#AIPodcast
#AI
#AgenticAI
#EnterpriseAI
#Databricks
#RAG
#MachineLearning
#DataEngineering
#EnterpriseEngineering
#AIStrategy
#AIEngineering0:00 Jonathan Frankle Returns | Databricks Chief AI Scientist · Invisible Machines
1:47 We Remember the Plants | Returning Guest Jonathan Frankle
2:22 Million-Token Context Windows: Do You Still Need to Train LLMs?
3:40 Be a Scientist | Measure AI Success Before You Scale
5:54 Hyperscaler Résumés Are Not Proof of AI Expertise
10:01 Maximize Impact | MosaicML, Databricks & Enterprise AI
13:02 Lottery Ticket Hypothesis vs. Real-World AI Impact
14:12 Nuclear Fusion but No Power Lines | Jonathan Frankle
16:08 AI Specification & Evals: Why "Build a Benchmark" Is a Cop-Out
17:59 The Smoothie Problem | From Model Power to Useful AI
18:53 Prompts as Parameters | Fine-Tuning Without Model Weights
22:46 It's Computing | Specification, Testing & Agent Design
24:44 LLM SEO, PDFs & Enterprise Data for AI Ingestion
27:35 Static FAQs, Curation & LLM-Oriented Publishing
30:26 Unlocking PDFs Scales Your Mistakes | Enterprise RAG
33:25 Knowledge vs. Reasoning | Brand Control in AI Search
34:50 Thanks for Listening | Invisible Machines