Show Notes
Every animal on earth runs on four drives. Humans have a fifth. The question is whether AI feeds it or buries it under a pile of chat windows.
That is the cold open Ryan Estes is bringing to a Denver AI panel next Wednesday, and it is the reason this week's What It Do with Jason Katz, co-founder of Kindling Solutions, turns into something bigger than a build-in-public check-in.
Ryan walks through his entire panel-hosting playbook live, and the metaphor he lands on is a DJ booth. Five panelists are five tracks. You mix them. You keep an eye on the room. You know when to loop somebody's best point and when to bring the fader down on a rambler. Jason's contribution to the craft is a bullhorn: one short honk to underline a great answer, a three-second honk to shut somebody up. It is a joke that is also, obviously, a real product.
Then Jason pushes back on the hook itself, which is the most useful sixty seconds in the episode. “Divine” is ambiguous, he says, and worse, it is divisive. People hear it and go straight back to Catholic school. Rewrite it as “does AI make us more or less human” and every single person in that room can answer it. Ryan takes the note on the spot. That is a masterclass in editing a hook down to the version that actually travels.
The heart of the episode is Jason's field report from two CEO meetings in one week, both at companies most founders would kill to run, and both at wildly different places on the AI maturity curve. One has visionary leadership, sharp instincts on the balance sheet and asset acquisition, and almost nothing underneath it operationally. The systems audit came back close to empty. The other has already built impressive end-to-end AI systems internally and now has the opposite problem: a pile of independent tools with no shared brain, and a CEO asking how to fuse them into a single operating system that learns as one thing and becomes the standard way the business runs.
Jason's read is that everyone lands in the same place eventually. Every company will run on a custom operating system with AI inside it. Right now the spread between companies is the widest it will ever be, and that spread is the entire market.
Then Ryan drops the question he has been asking every guest for months: what AI automation do you have running that you would be lost without? The answer is almost always the same. The morning brief. The thing that reads your Slack, your calls, your calendar, and your inbox and hands you five lines with your coffee. Half a dozen guests, all dependent on the same humble workflow. Jason's caveat is important: the morning brief is a beautiful entry point, and it is also the ceiling for any company that has not fixed its operational leadership first. Without executive buy-in to change how the business actually runs, you are just delivering a nicer-looking version of the same chaos.
From there the conversation goes where every serious operator conversation is going in 2026: model routing, tokenomics, compliance, and the quiet fear that the frontier labs will eat the tools built on top of them. Ryan uses Claude for Teachers as the case study. Anthropic watched an entire ecosystem build classroom tooling, then shipped its own. Jason's answer is not to panic about the model layer. It is to own the layer nobody can copy. Software is democratized. Code is democratized. Anyone can build a thing. The value is in understanding the engine: the workflows, the business logic, and the connection between all of it and what the company is actually trying to do. Route models correctly and you have to understand what each job is really doing and why. That understanding is the moat.
And then the best structural insight of the episode, which Jason pulled out of a conversation with a CEO that same day. An IT department is three buckets. Services, meaning hardware, software, permissions, users, laptops, printers, servers. Data, meaning governance, cleanliness, security, and a model anyone in the company can pull insight from. And applications, meaning custom software development. Go back to 2021 and bucket three meant six months from “we need a thing” to a signed proposal, and three to five hundred thousand dollars for an iPhone app. That timeline has collapsed to nearly zero. What has not collapsed is the reason IT owned that bucket in the first place. As Jason puts it, three years ago your guy running Claude Code would never in a million years have been near building software for your company. He is building it now. Governance and security did not stop mattering just because the build got fast.
The segment closes on a 1,500-pound buffalo launching a bearded prospector roughly fifteen feet into the Colorado air, which is either a metaphor for AI adoption or just a very good video. Proper buffalo. Well done.
Frameworks from This Episode
The Panelist DJ Method
Ryan's full structure for hosting an hour-long panel without it going off the rails.
- ›Cold open with a provocation, not a thank-you list. Hook the room in the first thirty seconds.
- ›Frame the conversation twice: once for the panelists so they know the arc, once for the audience so they know what they walk out with.
- ›Self-intros capped at thirty seconds each. Enforce it.
- ›Questions as a narrative arc, not a checklist. Never ask the same question to every panelist.
- ›Deliberately create friction. Call on the quiet panelist with "you buying this?" New ideas live in disagreement.
- ›Audience Q&A with a loaded backup question ready for the silence.
- ›Closers framed as action: what does someone do Thursday morning?
- ›Thanks and acknowledgements last, not first.
The Hook Edit
How Jason improved Ryan's cold open in under a minute.
- ›Test the hook for ambiguity. Does the key word carry one meaning or five?
- ›Test the hook for division. Does it activate a fight the audience is already having?
- ›Rewrite until every person in the room can answer it without a definition.
- ›"Does AI feed the divine" became "does AI make us more or less human." Same idea, zero friction.
The Operational Maturity Split
Jason's diagnostic for where an AI engagement actually starts.
- ›End A: strong vision, strong finance, no operational spine. Systems audit comes back empty. AI work here is rudimentary and necessary.
- ›End B: dozens of homegrown AI systems, no unified layer. The ask is consolidation into one operating system that learns centrally.
- ›Both are real customers. The intervention is completely different.
- ›Diagnose before you sell. Prescribing the wrong end of the curve burns the engagement.
The Trust Ladder, Starting With the Morning Brief
Jason's ceiling on what a single automation can do without executive buy-in.
- ›Rung one: a morning brief that reads Slack, calls, calendar, and inbox and returns five lines.
- ›Rung two: expand from the thing they already refuse to live without.
- ›Hard prerequisite: executive buy-in to change operations. Without it, rung one is the whole ladder.
- ›Diagnostic question to ask any prospect: what automation would you be lost without? If the answer is nothing, you are selling change management, not AI.
Model Routing as the DJ of the Stack
Why routing models is business logic work, not model work.
- ›Route by job, not by default. Drafting an email does not need a frontier model.
- ›Three forces driving it: token cost, compliance requirements, and fear of the frontier labs absorbing your product surface.
- ›Local and open source models for the jobs that cannot leave the building.
- ›Routing correctly requires understanding what each task actually is. That is business logic work, not model work.
Understand the Engine
Why the moat is not what you built.
- ›Software is democratized. Code is democratized. Anyone can build a thing.
- ›The durable value is workflows, business logic, and their connection to strategy.
- ›Corollary: the moat is not what you built, it is why you built it that way.
The Three Buckets of IT
Jason's structural map of what an IT department actually owns.
- ›Bucket one, services: hardware, software, permissions, user management, infrastructure, servers.
- ›Bucket two, data: governance, cleanliness, security, and a shared model anyone can pull insight from.
- ›Bucket three, applications: custom software and application development.
- ›The 2021 to 2026 change: bucket three's timeline and cost went to near zero, while its governance and security requirements did not change at all.
- ›The risk: everyone can write the app now, so nobody is asking who owns it.
The Handle, Not Just the Card
Jason's networking swag doctrine.
- ›The business card is the floor. Give them a second object that makes a story.
- ›Western Union style telegram postcards. 1950s matchbooks. A brand that reads like NASA crossed with mid-century office life.
- ›An NFC dot: tap phones, load a personal page, save straight to the contact card.
- ›Success condition for the interaction: they remember you, and you got the tap.
Win, Loss, or Buffalo
The weekly ritual that closes the segment.
- ›One win, one loss, or one buffalo, meaning something wild and unclassifiable that happened this week.
- ›Cheap format, high retention, endlessly repeatable.
Founder Experiment
Run the “Lost Without It” Audit, Then Build the Brief
Time required: one week. Cost: near zero. Day one: ask every person on your leadership team one question in writing, what automation or AI system are you running right now that you would be genuinely lost without, without leading them or offering examples. Day two: sort the answers into three piles, real dependencies, nice-to-haves, and nothing; the size of the third pile is your operational maturity score. Day three: build one morning brief for yourself only, reading your calendar, your unread priority inbox, your team's channel activity from yesterday, and your open tasks, output capped at five lines. Days four through six: run it every morning, change nothing else, and note every time you open a source app anyway, because that is the brief failing. Day seven: decide two things. First, would you be lost without it? If yes, roll it to three more people. If no, your brief is summarizing the wrong sources, so fix inputs before you touch the model. Second, pick your single most expensive recurring AI task and test whether a cheaper model could do it at the same quality. Success metric: by day seven, at least one person other than you asks when they get theirs.
Key Terms
Q&A
What is the best first AI workflow for a founder to automate?
A morning brief. It pulls your calendar, inbox, team messages, recent calls, and open tasks into a short executive summary delivered before the workday starts. Ryan Estes notes that when he asks guests which AI system they would be lost without, the morning brief is the near-universal answer, with roughly half a dozen guests describing real dependency on it.
How do I know if my company is ready to scale with AI?
Run a systems audit first. Jason Katz describes two companies at similar revenue where one had strong vision and finance but almost nothing operationally documented, while the other had already built end-to-end AI systems and needed consolidation. AI amplifies whatever operational spine already exists. If there are no SOPs and no operational leadership, the constraint is management, not tooling.
What is model routing and why does it matter for AI cost?
Model routing sends each task to the cheapest model capable of handling it, rather than defaulting everything to a frontier model. It matters for three reasons: token cost, compliance requirements that may demand local or open source models, and reducing dependency on any single provider. Routing well requires understanding what each task actually does, which is business logic work.
Should founders worry about frontier AI labs competing with their product?
It is a real dynamic. Anthropic launched Claude for Teachers in July 2026 after an ecosystem of classroom AI tools had already formed, though notably it launched with connectors into nine existing ed tech platforms rather than purely displacing them. Jason Katz's answer is that defensibility sits below the model layer, in workflows, business logic, and the connection to company strategy, because code itself is now democratized.
What are the three functions of an IT department in the AI era?
Services, covering hardware, software, permissions, user management, and infrastructure. Data, covering governance, cleanliness, security, and a shared data model. Applications, covering custom software development. The third function's cost and timeline collapsed from roughly six months and three to five hundred thousand dollars for a custom app down to near zero, while its governance and security responsibilities did not change.
Who should own AI application development inside a company?
IT should still own governance and security even though anyone with an agentic coding tool can now ship an app. Jason Katz's framing: three years ago the person now writing internal apps with Claude Code would never have been near company software development. The speed changed. The reason for oversight did not.
How do you host an AI panel discussion that does not go off the rails?
Open with a provocation instead of acknowledgements. Frame the takeaway for the audience up front. Cap self-intros at thirty seconds. Build questions as a narrative arc rather than a checklist. Deliberately create friction instead of letting panelists agree with each other. Keep a backup question loaded for audience Q&A silence. Close by asking each panelist what a listener should do the next morning.
What happens to software businesses when AI intelligence becomes free?
Ryan Estes raises the scenario where token cost and energy cost trend toward zero and intelligence is effectively unlimited and continuous. The consequence is that scarcity moves away from building capability and toward judgment, taste, and understanding the underlying business engine, since building the thing stops being the hard part.
Does AI make people less capable?
Jason Katz argues AI has more potential than any prior tool to make some people less capable, but that the cause is the choice to accept model output as gospel rather than anything inherent to the technology. His position is that users carry a responsibility to understand how the system works and combine its output with their own judgment.
How do you stand out at a networking event as a founder?
Give people a second object beyond the business card. Jason Katz brings Western Union style telegram postcards and 1950s style matchbooks that match Kindling Solutions' mid-century and NASA-inspired brand, plus an NFC dot that loads a personal page and saves directly to a phone contact. The goal is memorability plus a captured contact.
Links from This Episode
- Kindling Solutionshttps://kindlingsolutions.com
- Jason Katz on LinkedInhttps://www.linkedin.com/in/jasonkatz99/
- Ryan Estes on LinkedInhttps://www.linkedin.com/in/estesryan/
- AI for Foundershttps://aiforfounders.co
