Jon Harmeling automated 13 of his restaurant's 15 onboarding steps and freed up ten hours a week doing it. He owns and operates a Chick-fil-A in Miami Shores with about 90 people on his team, where only four or five of them ever touch a laptop, and that ten hours became a test case for a different question: what does freed-up time actually owe you?

In this conversation, Noah and Jon get into how a business built on shift work and thin margins decides what to automate first, the difference between using AI for management versus using it for leadership, and why the hardest part isn't building the tool. It's deciding, on purpose, where the time it saves actually goes.

Topics covered:

• Automating the repetitive admin work that eats a frontline operator's week 
• Using AI to surface leadership moments, not replace them.
• Where AI hits its limits: why scheduling a team well requires human context no model has been able to replicate

Chapter List
(00:00) What do you do with the time AI gives back?
(01:07) Using AI when your team doesn't sit at a desk
(02:23) From West Point systems engineering to running a Chick-fil-A
(07:40) The first win: AI takes over 13 of 15 onboarding steps
(13:26) What not to feed AI, and how much to trust it
(17:07) Catching bad reviews while you can still fix them
(19:03) The setup: one laptop, no servers, one step at a time
(23:45) From management to leadership: AI that finds who's being overlooked
(28:09) Keeping AI-prompted moments genuine
(32:53) Where AI failed: scheduling needs human context
(36:36) Humans caring for humans: the retention flywheel
(40:44) Jon's book, "Frontline AI Leadership," and where to find him

Resources
Valet — Episode Sponsor
Valet is the founding sponsor of the Serious People Podcast.

AI is getting very good at making useful things: dashboards, analyses, internal websites, and small tools. Making those things available to the rest of a team is a separate problem.

Valet lets you publish AI-created artifacts to a stable URL. New sites are private by default, and members of your organization can access them after signing in.

Try Valet at https://valet.dev

Connect with Jon on LinkedIn: https://www.linkedin.com/in/jonharmeling
For more info on Frontline AI Leadership: https://www.jonharmeling.com/book/

Connect with Noah on LinkedIn: https://www.linkedin.com/in/noahlevin/
Noah's Website: https://www.noahlevin.com/
Serious People Website: https://seriouspeople.ai/

• (00:00) - What do you do with the time AI gives back?

Christina Johnson has been shown a 33x return on ad spend and kept a straight face. She has sat with a retail media network selling hundreds of millions of dollars in advertising whose own sales team couldn't explain to buyers whether the money had worked. And she can point to a supplier that ran a flawless campaign in cities where the retailer has no stores.

Christina Johnson spent years inside Walmart's retail media business before founding Advisar, an intelligence tool for retail media buyers. In this conversation, she and Noah break down why so many brands are flying blind on their ad spend, why a metric like ROAS can look great while sales go nowhere, and what happens when AI gives you a confident, completely wrong answer.

If your business runs any kind of performance marketing, this episode will change how you look at your own numbers.

You'll hear:

• Why more ad spend rarely means better marketing
• How a "perfect" campaign can drive zero real sales
• What questions to ask before trusting a marketing metric
• Why AI can be confidently wrong, and how to catch it

Timestamps:
(00:00) More ad spend doesn't mean better marketing

(00:34) Meet Christina Johnson, founder of Advisar

(03:08) Why retail media is "wildly backwards"

(08:37) The meeting where the sales team couldn't explain their own numbers

(10:18) The question every retail marketer should ask

(11:49) The auto-bid campaign that spent $0 and sold nothing

(13:26) Why Christina built Advisar

(15:18) The "33x ROAS" that wasn't real

(18:11) ROAS explained, and why it can lie to you

(19:09) Using Claude for analysis, and where it breaks

(33:20) The AI mistake that changed how she builds

(39:31) One thing to do tomorrow with your ad spend data

Resources:

Valet — Episode Sponsor

Valet is the founding sponsor of the Serious People Podcast.

AI is getting very good at making useful things: dashboards, analyses, internal websites, and small tools. Making those things available to the rest of a team is a separate problem.

Valet lets you publish AI-created artifacts to a stable URL. New sites are private by default, and members of your organization can access them after signing in.

Try Valet at https://valet.dev

Christina Johnson operator field guide: https://seriouspeople.valet.run/spp-christina-operator-field-guide

Connect with Christina on LinkedIn: https://www.linkedin.com/in/christina-w-johnson

Advisars Website: https://www.advisar.io/

Connect with Noah on LinkedIn: https://www.linkedin.com/in/noahlevin/ 

Noah’s Website: https://www.noahlevin.com/ 

Serious People Website: https://seriouspeople.ai/ 

• (00:00) - More ad spend doesn't mean better marketing

Rebekah Stevens once spent months competing against an algorithm named Octopus, running weekly throwdowns to see who could staff a home care market better: her, with a spreadsheet and years of operating instinct, or a model built by her own data science team. She now leads Care Transformation at Honor, the senior home care platform, where that early contest became the blueprint for how the company decides what AI should touch, and what it shouldn't.

In this conversation, Noah and Rebekah get into how a business built entirely on trust and vulnerable moments decides where to let AI in. They cover why unstructured data doesn't automatically mean you need a language model, what it actually takes to turn a fuzzy client complaint into a precise, AI-ready metric, and why the hardest part of deploying AI is writing the goal correctly, not building the model.

You'll hear:

• How Honor defined the outcome before automating staffing decisions
• The real story behind Rebekah vs. Octopus, the human-versus-algorithm contest that built the system running Honor's staffing today
• How her team decided a language model was the wrong tool for a problem everyone assumed needed one
• Where Honor draws the line between AI working in the background and AI on the phone with a client

Timestamps: 

(00:00) Introduction
(02:05) Inside Honor: senior care built like a tech platform
(10:06) Turning client churn into a math problem
(16:38) Rebekah vs. Octopus: human vs. algorithm
(23:53) Where large language models actually showed up in the product
(26:59) When not to use an LLM
(30:10) What's blowing her mind about agentic AI right now
(33:01) Directed Work: letting the system decide who does what
(45:01) How AI will reshape product teams next year

Resources:

Valet — Episode Sponsor

Valet is the founding sponsor of the Serious People Podcast.

AI is getting very good at making useful things: dashboards, analyses, internal websites, and small tools. Making those things available to the rest of a team is a separate problem.

Valet lets you publish AI-created artifacts to a stable URL. New sites are private by default, and members of your organization can access them after signing in.

Try Valet at https://valet.dev 

Rebekah Stevens operator field guide: https://seriouspeople.valet.run/spp-rebekah-operator-field-guide/

Connect with Rebekah on LinkedIn: https://www.linkedin.com/in/rebekah-stevens-6751b41a/ 

Honor’s Website: https://www.honorcare.com/

Connect with Noah on LinkedIn: https://www.linkedin.com/in/noahlevin/ 

Noah’s Website: https://www.noahlevin.com/ 

Serious People Website: https://seriouspeople.ai/ 

• (00:00) - Introduction

Nisha Balwani took over RCI Technologies, the government IT staffing and services firm her parents founded in the 1980s, and is now figuring out where AI actually fits inside a business built on decades-long contracts and highly regulated clients. Noah and Nisha get into what it took to get a skeptical team on board, how one employee's irreplaceable process got turned into an AI-assisted workflow without losing the human check, and why government adoption of AI looks nothing like the private sector. Nisha also shares how her internal experiments turned into a side consulting practice helping other small and mid-size businesses figure out where to start.

Topics covered:

• Turning a change-management standoff into team buy-in
• Encoding one employee's institutional knowledge into an AI workflow
• How government and regulated clients approach AI adoption differently
• Building a consulting practice out of solving your own AI problem first

Chapters:
(00:00) Introduction
(02:40) Will AI disrupt the staffing model itself?
(03:53) Why relationships beat RFPs — and the AI-pilled 60-something salesperson
(06:48) Finding your real bottleneck, and the candidate caught using AI mid-interview
(09:14) How to be an AI-pilled CEO without alienating your team
(11:22) Change management, inside champions, and clients who can't keep up
(16:09) Nisha's personal AI stack: Claude Enterprise and MCP connectors
(18:22) Killing busywork: automating the CRM and ditching the "Beautiful Mind" spreadsheet
(27:11) The invoicing build: from bus-factor-of-one to a working system
(34:02) How to actually start building your own AI tool
(40:29) Launching an AI consulting line: origin story and pricing
(45:39) Lightning round: strategy decks, dumbest AI moments, and where to find Nisha

Resources:

Valet — Episode Sponsor

Valet is the founding sponsor of the Serious People Podcast.

AI is getting very good at making useful things: dashboards, analyses, internal websites, and small tools. Making those things available to the rest of a team is a separate problem.

Valet lets you publish AI-created artifacts to a stable URL. New sites are private by default, and members of your organization can access them after signing in.

Try Valet at https://valet.dev 

Connect with Nisha on LinkedIn: https://www.linkedin.com/in/nishabalwani/

Follow RCI on LinkedIn: https://www.linkedin.com/company/rci-technologies-inc./

Connect with Noah on LinkedIn: https://www.linkedin.com/in/noahlevin/ 

Noah’s Website: https://www.noahlevin.com/ 

Serious People Website: https://seriouspeople.ai/ 

• (00:00) - Introduction

For the office manager at your doctor’s office, it’s not about having an "AI strategy." It’s about answering the phone during the 8am rush, or patients hanging up before anyone picks up.

Alex Cohen used to run product at Carbon Health, where he first tried to solve that problem with call-center software. Now he's co-founder of Hello Patient, where AI agents answer the phone, book appointments, and handle insurance and billing calls for healthcare clinics. 

On this episode, he lays out the three rules his team won't break when they put AI in front of a patient. He also explains why the hardest part of any AI rollout isn't the AI; it's the 20% of the work that every clinic does differently. And he gives a direct answer to the question every operator actually has: what do I need to check before AI can touch my own systems?

No strategy decks. No jargon. Just what it takes to run AI inside a business that can't afford to get it wrong.

Topics covered:

• What Hello Patient's AI agents actually do inside a healthcare clinic
• The three non-negotiable rules for deploying AI in a clinical setting 
• How to actually get AI-ready (EMR access, APIs, compliance)

Chapters:

• (00:00) Introduction
• (01:00) Alex Cohen: From Carbon Health to Hello Patient's Origin Story
• (06:03) Finding the Pattern: Building the First Voice AI
• (08:05) From Manual IVRs to Agentic AI
• (12:14) What Hello Patient Actually Does
• (14:15) Rule #1 — No Clinical AI Without a Human in the Loop
• (20:49) Rule #2 — Never Let AI Pretend to Be Human
• (27:01) Rule #3 — Don't Try to Automate 100% of Calls
• (29:44) The Last-Mile Problem: Why the Final 20% Is 90% of the Work
• (33:25) Why EMRs Aren't Commoditized Like Your CRM 
• (34:58) How to Actually Get AI-Ready 
• (37:08) Why Some Systems Won't Give You API Access 
• (38:33) HIPAA, Compliance, and Where AI Actually Helps 
• (40:57) Where to Find Alex & Hello Patient

Resources:

Valet — Episode Sponsor

Valet is the founding sponsor of the Serious People Podcast.

AI is getting very good at making useful things: dashboards, analyses, internal websites, and small tools. Making those things available to the rest of a team is a separate problem.

Valet lets you publish AI-created artifacts to a stable URL. New sites are private by default, and members of your organization can access them after signing in.

Try Valet at https://valet.dev 

Connect with Alex on LinkedIn: https://www.linkedin.com/in/alexander-cohen/ 

Follow Hello Patient on LinkedIn: https://www.linkedin.com/company/hello-patient/

Connect with Noah on LinkedIn: https://www.linkedin.com/in/noahlevin/ 

Noah’s Website: https://www.noahlevin.com/ 

Serious People Website: https://seriouspeople.ai/ 

Matt Van Horn got rid of his IDE. Then he got rid of his text editor. Now he doesn't even read the plan his AI agent writes; he just asks for the TL;DR (Too long; didn’t read) and lets it run. He's the founder behind three open-source projects (Last 30 Days, Printing Press, and Agent Cookie), the former founder of June, the smart oven company that put an NVIDIA GPU on a countertop appliance years before anyone said "agentic," and he's now building a new company in what he calls the "after Claude" era.

In this conversation, Noah and Matt get into what it actually looks like to trust an AI agent with real work, where that trust should stop, and why the calculus changes completely once real customers, real revenue, or real hardware are on the line.

In this episode, you’ll hear:

• Why Matt stopped reading the plan, and when that's actually a bad idea
• How building a company after Claude 4.5 changes headcount and strategy
• The simplest way to start learning AI coding by contributing to open source

Chapters:
(00:00) Introduction

(02:15) The case for “dangerously skip permissions”

(03:31) From “don’t read the code” to “don’t read the plan”

(07:45) Why vibe coding died last Thanksgiving

(09:11) Matching your risk tolerance to the stakes

(11:13) Building a company after Claude 4.5

(15:04) Personal productivity with Granola and Claude Code

(16:05) The missing agentic email product

(18:07) How the open source projects feed the business

(20:35) The best way to start contributing to open source

(25:06) There's more than one right way to build now

Resources:

Sponsor: Valet — https://valet.dev — the secure place to share what AI creates.
Follow Valet on LinkedIn: https://www.linkedin.com/company/valetdotdev/ 

Connect with Noah on LinkedIn: https://www.linkedin.com/in/noahlevin/
Serious People Website: https://seriouspeople.ai/ 

Connect with Matt on X: https://x.com/mvanhorn 

Find Matt on GitHub: https://github.com/mvanhorn 

June’s Website: https://juneoven.com/ 

Lyft’s Website: https://www.lyft.com/ 

Matt's projects

Last 30 Days: https://github.com/mvanhorn/last30days-skill 

CLI Printing Press: https://github.com/mvanhorn/cli-printing-press 

Printing Press Library: https://github.com/mvanhorn/printing-press-library 

Agent Cookie: https://github.com/mvanhorn/agentcookie 

Agentic engineering

Compound Engineering: https://github.com/EveryInc/compound-engineering-plugin 

CE plan: https://github.com/EveryInc/compound-engineering-plugin/blob/main/docs/skills/ce-plan.md 

Claude Code: https://claude.com/product/claude-code 

OpenAI Codex: https://openai.com/codex/ 

Products and tools mentioned

Howie: https://howie.com/ 

Granola: https://www.granola.ai/ 

Superhuman: https://superhuman.com/ 

Clicky: https://github.com/farzaa/clicky 

Wispr Flow: https://wisprflow.ai/ 

Digg: https://digg.com/ 

OpenClaw: https://github.com/openclaw/openclaw 

Hermes Agent: https://github.com/NousResearch/hermes-agent 

cmux: https://github.com/manaflow-ai/cmux 

Ghostty: https://github.com/ghostty-org/ghostty 

Podcast reference

Lenny's Podcast with Cat Wu: https://open.spotify.com/episode/7wTqD5zwl8ZitUeplapvBG 

Cat Wu: https://www.linkedin.com/in/catwu/ 

Michael Ducker and Noah Levin get into the practical mechanics of bringing AI into a real business. Michael shares the story behind his work building open source emoji at Twitter, then digs into why AI is creating real disillusionment even as it transforms how work gets done.

They walk through what it actually takes to get IT and a skeptical team on board, and the step-by-step process for turning a single AI session into a reusable skill through a technique called hill climbing.

The conversation lands on the real shift underway: agents moving from single-player tools into something teams build and share together, changing what it means to be a worker at all.

You’ll hear:

• Three reasons AI adoption is creating disillusionment right now
• How to get IT comfortable with AI tools through proxies and access controls
• What a skill really is, and how to build one using the hill climbing technique
• Moving from single player AI use to building agents with a team
• Where to start automating real business processes, from knowledge bases to full agent workflows

Chapters:

(00:00) Introduction
(01:06) Michael's role in bringing open source emoji to the web
(04:10) Sponsor: Valet
(05:39) Three reasons AI is causing disillusionment
(09:33) The trust question: giving AI access to email and bank accounts
(12:07) The founding story of Valet
(15:43) What actually sells AI adoption to a team
(20:31) Building a shared folder as the foundation for team AI adoption
(22:45) What a skill is and how to build one through hill climbing
(32:34) Single-player vs. multiplayer mode
(33:39) What an agent is, and graduating to agent builder
(36:54) Where to start: high-value agent use cases
(42:37) Startup and model risk: playing the long game
(43:48) Where to find Michael, and wrap-up

Resources:
Valet — Episode Sponsor

Valet turns repeatable business processes into agents that work where your team already does—across Slack, email, calendars, and the tools you use every day. Try the Calendar Research agent, which researches who you’re meeting and sends the context you need directly to Slack before the conversation.

Set up Calendar Research with Valet

https://dashboard.valet.dev/deploy/calendar-research

Valet — own the business, not the busy-ness.

Follow Valet on LinkedIn: https://www.linkedin.com/company/valetdotdev/ 

Valet Website: https://valet.dev/

Connect with Michael on LinkedIn: https://www.linkedin.com/in/miradu/

Connect with Michael in person at Agent Builders Breakfast: https://agentbuildersbreakfast.com/ 

Connect with Noah on LinkedIn: https://www.linkedin.com/in/noahlevin/ 

Serious People Website: https://seriouspeople.ai/ 

Most AI content is tech people talking to tech people. Serious People is different.

This is for the operators actually running things — selling supplements, managing caregivers, coordinating field crews. Real businesses with bad data, compliance headaches, and employees who've heard it all before.

Host Noah Levin spent almost twenty years inside complex operations at Amazon, Whole Foods, and healthcare tech. Now he sits down with operators who've done the work: what broke, what they tried, what it cost, and whether it actually paid off.

No hype. No magic prompts. No pretending the hard part is picking a model. Just AI, inside a real business, from people who run one.

New episodes start this month. Subscribe now.

Resources:

Connect with Noah on LinkedIn: https://www.linkedin.com/in/noahlevin/ 

Noah’s Website: https://www.noahlevin.com/ 

Serious People Website: https://seriouspeople.ai/