- Needs To Exist
- Posts
- Your industry has a Medvi-shaped hole in it. Here's how to find it.
Your industry has a Medvi-shaped hole in it. Here's how to find it.
Needs To Exist
May 17, 2026
In partnership with
Yesterday I told you about Medvi. Two-person company, $401M in year one, 16.2% margin, run out of an LA house with a dozen AI tools and one brother on the payroll.
A good story. The problem with good stories is they're inspiring for about 12 hours, then the feeling fades and the inbox wins.
So this morning I'm going to fix that.
What follows is the five-step check I'd run on myself before lunch today if I were trying to find a Medvi-shaped business inside an industry I already know.
A notebook. A pen. Thirty minutes. Most readers will skim this and feel productive. The handful who actually pick up the pen are who I'm writing for.
Step 1 — List the saturated category you already have access to
Don't pick the category that sounds exciting on Twitter. Pick the category you have actual access to. A job. A family member. A former career. Customers you can call today.
Gallagher's access was direct-response consumer marketing. He'd run a watch subscription business to $300M. He knew funnels, ad buying, conversion, and creator-driven attention. Telehealth was the product wrapper. The DTC muscle was the wedge.
Write down one to three categories where you already know the customer, the seasonality, the unit economics, and where the spend actually goes. Not categories you'd like to enter. Categories you're already in.
Specificity wins. "Marketing" is too broad. "Lower-funnel paid social for women's apparel under $100" is a category.
Step 2 — Find the fat-margin incumbent
For each category from Step 1, name the public or near-public incumbent. The biggest one. Pull up their LinkedIn and their last 10-K if they're public.
Note three numbers. Headcount. Revenue. Net margin.
Hims has 2,442 employees, $2.4B in revenue, 5.5% net margin. That's the target shape.
Look for categories where the public incumbent has thousands of employees and single-digit margins. The big employee count and the thin margin are signals that the org chart got built for a different decade. There's air in there.
If your category's biggest incumbent has 80 employees and 35% margins, pick a different category. You're not going to out-margin an already-lean operator. Move on.
Step 3 — Map their workflow stack
Write down everything the incumbent's headcount actually does. Not the org chart. The verbs.
Buys ads. Answers tickets. Writes copy. Builds landing pages. Edits creative. Runs A/B tests. Generates reports. Talks to prescribers. Processes refunds. Manages the affiliate program. Updates the FAQ.
Now star the verbs an AI stack can do with one human in the loop. Be honest, not aspirational. "Writes copy" gets a star. "Talks to a doctor about a complex case" does not.
Gallagher's star list, inferred from the NYT piece, was code, copy, creative, support, and analytics. Almost everything except the parts that legally required a licensed prescriber or the founder himself. That's why the headcount stayed at two.
How Marketers Are Scaling With AI in 2026
61% of marketers say this is the biggest marketing shift in decades.
Get the data and trends shaping growth in 2026 with this groundbreaking state of marketing report.
Inside you’ll discover:
Results from over 1,500 marketers centered around results, goals and priorities in the age of AI
Stand out content and growth trends in a world full of noise
How to scale with AI without losing humanity
Where to invest for the best return in 2026
Download your 2026 state of marketing report today.
Step 4 — Pick your Replit, your Claude, your Grok
Don't pick six tools. Pick three. The fewer the better.
The tool list isn't the point. The commitment is. Founders who switch tools every two weeks are not running an AI-leveraged company. They're running a procurement function. Gallagher didn't change his stack every Tuesday. Neither should you.
Write the three tools down. Date the page. Hold the stack for 90 days minimum.
Step 5 — Decide your AI failure policy before you launch
This is the step most readers will skip. It's also the line between a Medvi-shaped business and a Medvi-shaped lawsuit.
Gallagher's customer service bot hallucinated drug prices. He honored every wrong price the bot quoted. The bot also invented product lines that didn't exist, and he hand-corrected each one. The FDA sent a warning letter about claims on the site.
That is the cost structure of a two-person AI company. The founder is the only human backstop when the AI does something stupid.
Write down the worst thing an AI in your stack could plausibly do inside your category. Then write the policy you'll hold when it happens.
Honor the price. Refund the customer. Pull the page. Call the regulator. Whatever it is, decide today, not the day the email lands in your inbox at 11pm on a Saturday.
That decision is the moat. Not the model. Not the prompt. The reliability of your response when the AI fails.
What you should have on paper before lunch
If you ran the five steps, you have:
One saturated category with insider access
One incumbent with thousands of employees and single-digit margins
A starred list of verbs an AI stack can do with one human in the loop
Three tools you're going to hold for 90 days
One AI failure policy committed to before launch
That's more concrete progress than 90% of the people who replied to yesterday's email will make this week. Cost: 30 minutes and a pen.
The reason NTE Pro is structured the way it is, 7,000+ ideas organized by industry and shape, is because Step 3 is where most founders freeze. They have access. They have an incumbent in mind. They cannot picture the workflow stack of an industry they don't live in.
NTE Pro is that library. After 50 reads, the workflow stacks of 50 industries are in your head, and the verb-list step takes ten minutes instead of two hours.
For the operators wanting to spot the next two-person company before the NYT writes them up, WhoFiled is the live feed of state filings, trademark drops, and small-team hiring patterns. The next Medvi is sitting in this morning's data. Find them before the profile runs.
Hit reply
I want to read what you came up with. Three things:
Your category
Your incumbent's headcount and margin
The one verb you'd hand to an AI on Monday morning
I read every reply. The strongest ones I'll feature in a future edition, anonymized if you want.
If you skipped Step 5 because "the AI in my stack doesn't fail," that IS the answer. Write down the failure mode you've been pretending isn't coming. That's where the homework is.
Monday we're back to a regular Idea Of The Day. Get the five steps on paper before then.