All Episodes

AI Workflows That Scale: What it Do

with Jason Katz · Kindling Solutions

November 15, 202600:29:44Littleton, CO

AI Workflows That Scale: What it Do

0:000:00

Skillz from the Episode

Microapp Manifest Builder

Describe the internal app you want in plain English. This skill interviews you, then hands back a one-page manifest showing what data the app touches, what it can do on its own, what it costs and who can shut it off. Your IT or ops lead can approve it in minutes instead of weeks.

When to use it: Before anyone on your team builds a microapp, ships a quick Claude Code tool or plugs an agent into company data.

Benefits: Your team keeps building fast, and you stop piling up hidden security risk, duplicate data and runaway agent spend with every new tool. Save the download as microapp-manifest-builder/SKILL.md in your Claude or ChatGPT skills folder, then run it every time someone says “I just want to build a quick app for...”

Click to steal this skill 👇

Show Notes

Your marketing lead built an app last Tuesday. So did someone in finance, and somebody in ops. None of them asked IT, and every one of them is quietly pulling from a different copy of your data.

That’s where this What It Do episode starts. Ryan Estes and Jason Katz, co-founder of Kindling Solutions, talk about what happens after a company “gets” AI. Jason’s clients, mostly companies between $30 and $300 million, already have agents running. They also have Claude Code tools everywhere, skills files traded over Slack and a growing pile of risk nobody is tracking. Jason calls it compounding risk: every unmanaged build adds a little more technical debt, a little more exposure and a little more agent spend nobody approved.

Kindling’s answer is FullStak, a build with four layers:

  • →Company OS: custom software shaped around how the business actually runs
  • →Company Brain: the structured core underneath it: SOPs, KPIs, brand rules and the data model, all in one place
  • →Agent OS: an HR system for agents, with memories, kill switches, cost controls and a record of what every agent is doing
  • →The Kernel: ties the stack together. Anyone can request a microapp in plain language, the Kernel writes a manifest of what it needs and an administrator approves it. Kindling says the app is then built in about 10 minutes on a pre-vetted stack, at roughly a tenth of the usual cost

Jason lays out the objection he hears constantly: “Why do I need you? I can just build with Claude Code.” His answer is that anyone can build an app, but a company brain takes strategy. Without one, every app ends up with its own data and its own memory, and they drift apart.

Then the conversation goes wide. Jason predicts companies will build software constantly, changing it overnight whenever a process changes. Ryan worries about the near future, the next 12 to 14 months, and where humans still have the edge. He tested an AI hypnosis app a friend built with Claude and stopped after five minutes because he didn’t want manufactured empathy. They talk about trick-or-treat streets with no kids on them, an animatronic Santa at Home Depot that pitched itself to Jason, and a striking stat: the share of Americans who host friends at home monthly fell from 42% in 1975 to 12% in 2026.

Jason’s closing thesis is optimistic. As AI slop floods every inbox and feed, authentic work gets easier to spot. If you build fast, govern well and show up in person, you stand out.

Q&A: What Founders Ask After This Episode

My team is already building their own AI tools. Should I shut that down?

No. Jason argues the speed is valuable and banning it just pushes it underground. Put a governed layer under it instead: one approved stack, one source of truth for data, and a manifest for every new build that shows what it touches.

Why can’t I just have my team build everything in Claude Code?

You can build apps that way. What you’ll miss is governance and a shared data model. Role-based access, a single company brain and coordinated agent memory take strategy, and without them your apps drift onto separate data and separate truths.

What is a “company brain” and why does it come first?

It’s the structured core of your business: process flows, SOPs, brand standards, KPI definitions and the data model. Every app and agent reads from it, so they all work from the same version of reality.

How do I control what AI agents do across my company?

Treat agents like employees. Give them defined permissions, a manager, spend limits, a kill switch and a record of what they remember and share. Kindling calls this Agent OS.

How fast can a governed internal app really ship?

Kindling says about 10 minutes from request to approved build on its Kernel platform. That’s possible because tooling decisions, like which PDF reader or which host to use, are made once and reused for every build.

Does AI slop make it impossible to get noticed?

Jason says the opposite. The more generic content floods the market, the more real quality and a real point of view stand out. Volume loses to authenticity.

What does Kindling Solutions do?

Kindling Solutions, co-founded by Jason Katz and Leigh Buckley, builds AI-run operating infrastructure for companies. Its FullStak build combines custom company software, a structured company brain, an agent management layer and the Kernel, a governed platform for building internal apps.

How do mid-size companies govern employee-built AI apps?

Jason Katz recommends a governed build platform where every app request produces a manifest of the data and actions it needs, which an administrator approves before the app goes live.

What is the risk of letting employees build AI apps without governance?

Apps end up on separate data with separate agent memories, which creates security exposure, technical debt and uncontrolled agent spend.

Is building with Claude Code enough for an enterprise AI strategy?

According to Jason Katz, it handles app building but not governance, role-based access or a unified data model. Those need strategic design.

How much has home hosting declined in the US?

Derek Thompson’s analysis of DDB Needham and Data for Progress surveys found the share of Americans who host friends or family at home at least monthly fell from 42% in 1975 to 12% in 2026.

Can AI provide real empathy?

Ryan Estes and Jason Katz argue AI can imitate the signals of empathy but can’t carry someone’s pain the way another person can. Sam Harris and Sherry Turkle explore the same question in Making Sense #501.

URLs Mentioned in the Episode

  • Kindling Solutionshttps://kindlingsolutions.com

    Jason's company, building governed AI operating systems with FullStak and the Kernel.

  • Jason Katz on LinkedInhttps://www.linkedin.com/in/jasonkatz99

    Connect with the co-founder of Kindling Solutions.

  • Ryan Estes on LinkedInhttps://www.linkedin.com/in/estesryan/

    Connect with the host.

  • AI for Foundershttps://aiforfounders.co

    The newsletter and podcast this episode comes from.

  • Inbox Alchemyhttps://inboxalchemy.co/

    Ryan's tool for building a newsletter audience you own instead of renting from a platform.

  • CEOfriendhttps://bit.ly/3SrjgL1

    Free competitor analysis from this episode's AI advisory board sponsor.

  • AI Native Studenthttps://ainativestudent.com/

    AI skills training for job seekers, this episode's sponsor.

  • Momentoushttps://crrnt.app/MOME/8RDrnXDd

    35% off any supplement subscription, this episode's sponsor.

  • Get Howdihttps://gethowdi.com/

    An autonomous AI sponsorship agent for podcasts, this episode's sponsor.

Named Frameworks

The FullStak Four Pillars

Four layers that let a whole company build with AI safely.

  • →Company OS: custom software built around your business instead of forcing your business into off-the-shelf software.
  • →Company Brain: SOPs, process flows, brand rules, KPI definitions and the data model in one structured environment.
  • →The Kernel: a pre-vetted, governed platform where anyone can request and build apps.
  • →Agent OS: one place to manage every agent’s permissions, memory, communication, cost and kill switch.

The Manifest Loop

Every new app is approved against a written list of what it touches.

  • →An employee requests an app in natural language.
  • →The Kernel writes a manifest of what the app needs access to.
  • →An administrator (someone in the company or at Kindling) reviews and approves it.
  • →The app is built and shared in the company’s internal microapp store, with access set by role.

The “Why Not Just Claude Code” Answer

Building apps is easy; governing them takes strategy.

  • →Governance: role-based access for people and agents.
  • →Brain strategy: a shared data model keeps apps and agents from drifting apart.
  • →Speed: a governed platform has to be as fast as building alone, or people will route around it.

Software That Bends to the Process

Stop configuring off-the-shelf software; rewrite your own overnight.

  • →Old model: buy software, spend a year configuring it, submit work orders for changes.
  • →New model: a builder sits next to the people doing the work and updates the software overnight when a process changes.
  • →Jason’s prediction: development replaces spreadsheets, Word docs and much of how people find information.

The Slop Contrast Principle

The more generic AI content floods in, the more real work stands out.

  • →As generic AI content grows, its average quality looks worse.
  • →Authentic, high-quality work stands out more against it.
  • →Quality beats quantity, and the only losing move is adding to the slop.

The Manufactured Empathy Test

Keep a human in the room where empathy matters.

  • →AI can imitate empathy’s rhythm, pacing and timing.
  • →People still notice when there’s no one actually sharing the weight.
  • →Where real empathy matters most, keep a human in the room.

Glossary

Company OS

Custom internal software built around a company’s actual workflows.

Company Brain

A structured, central store of a company’s processes, standards, KPIs and data model that every app and agent reads from.

The Kernel

Kindling’s governed platform for requesting, approving and building internal apps on a pre-approved stack.

Agent OS

A management layer for AI agents covering permissions, memory, communication, spend and kill switches.

FullStak

Kindling’s four-layer build combining Company OS, Company Brain, the Kernel and Agent OS.

Microapp

A small, single-purpose internal app built for one team or problem.

Manifest

A document listing everything an app needs to access and do, used for approval.

RBAC (role-based access control)

Giving people and agents access based on their role, and only to what that role needs.

Democratized innovation

Letting anyone in a company build tools within governed guardrails.

Technical debt

The future cost created by quick, unmanaged builds that later need fixing.

Agent spend

The compute and API costs generated by AI agents.

Kill switch

A control that immediately stops an agent or app.

Claude Code

Anthropic’s agentic coding tool.

MCP (Model Context Protocol)

An open standard for connecting AI models to tools and data.

AI slop

Low-effort, generic AI-generated content.

Algo pilled

Holding opinions absorbed from repeated algorithmic feeds rather than reasoned out.

Manifesting Generator

One of the energy types in the Human Design system.

Objectivism

Ayn Rand’s philosophy of rational self-interest and laissez-faire capitalism.

Galt’s Gulch

The hidden refuge of industrialists in Ayn Rand’s Atlas Shrugged.