Tim Ferriss has a framework I keep thinking about: if you want to see the products of the future, look at what ultra-wealthy people buy, then figure out how to make it available to everyone.

The rich have personal drivers. We got Uber. The rich have personal shoppers. We got Stitch Fix. The rich have personal assistants. We got… well, we’re still figuring that one out.

But here’s the one that caught my attention: the rich have personal chefs.

Not because rich people can’t cook. Because they figured out something the rest of us are just now realizing: the cognitive overhead of feeding a family (deciding what to make, every single day, while accounting for everyone’s preferences, allergies, and schedules) is a job. A real one. And if you can afford to outsource it, you do.

I’m building the version of that for everyone else. And in the process, I’ve become convinced that vertical AI agents (not chatbots, not copilots, not general assistants) are the actual product revolution we should be paying attention to.

The Problem With General-Purpose AI

We’re living through an interesting moment. The foundation models are extraordinary. GPT-4, Claude, Gemini: they can write, reason, code, and analyze at a level that would’ve seemed like science fiction three years ago.

And yet.

Most people’s daily interaction with AI is still some version of opening ChatGPT and staring at a blank text box. The experience is powerful but directionless. It’s like being handed a Formula 1 engine and told to figure out the car yourself.

The people who get the most value from AI right now are the ones who already know what to ask. Developers who know how to prompt. Power users who’ve built custom GPTs and automation chains. People, frankly, like the ones reading this post.

But for the other 95% of the population (the parents, the small business owners, the people who don’t think about AI as a tool but as a vaguely impressive thing they saw on the news) the blank text box is a wall, not a door.

This is the gap vertical agents fill.

What I Mean by “Vertical Agent”

A vertical agent isn’t a chatbot with a system prompt. It’s an AI that’s been purpose-built around a specific domain, with opinions, context, workflows, and integrations that make it useful without requiring the user to be an AI expert.

The difference is the same as the difference between Google and a travel agent. Google can answer any question about flights. A travel agent asks you where you want to go, what you care about, remembers you hate layovers, and books the trip. One has capability. The other has capability plus context plus initiative plus taste.

When I started building Dinner Solved AI, I didn’t want to build “ChatGPT but for recipes.” That already exists in a dozen forms and it’s moderately useful. I wanted to build Martine: a character with a name, a point of view, and a job.

Martine is your AI personal chef. Not a recipe search engine. Not a meal planning dashboard. A chef. You talk to him. You tell him your daughter won’t eat onions, your partner is doing low-carb, and you’ve got chicken thighs and broccoli in the fridge. He doesn’t give you ten options and ask you to pick. He tells you what to make. He gives you the recipe. He builds the shopping list. He sends it to Instacart or Postmates.

That specificity is the product. And I think it’s the template for what’s coming across every domain.

The Tim Ferriss Framework Applied

Let’s play this out across a few categories.

The rich have personal chefs → Vertical agent: AI personal chef. This is Martine. He handles the planning, recipes, shopping lists, and grocery delivery. The cognitive load of feeding a family, which runs roughly 400 hours per year for the average parent, gets compressed into a 30-second conversation.

The rich have wealth managers → Vertical agent: AI financial advisor. Not a robo-advisor that rebalances your index funds. An agent that knows your income, your debts, your goals, your risk tolerance, and proactively says “you’re spending 40% more on dining out this month. Want me to adjust your savings plan?”

The rich have executive assistants → Vertical agent: AI chief of staff. Not a calendar bot. An agent that knows your priorities, triages your email, preps you for meetings, follows up on action items, and flags when your week is going off-track. The things a great EA does that no scheduling app replicates.

The rich have personal trainers → Vertical agent: AI fitness coach. Not a workout generator. An agent that adjusts your program when you say “my knee hurts,” remembers you hate burpees, knows you’re training for a half marathon in April, and holds you accountable without being annoying.

The rich have private tutors → Vertical agent: AI tutor. Not a homework helper. An agent that knows your kid’s grade level, learning style, strengths, and weaknesses, and builds a personalized curriculum that adapts in real-time.

In every case, the pattern is the same: take a service that costs $50-500/hour when delivered by a human, wrap it in an AI agent with deep domain knowledge and persistent memory, and deliver it for $20-30/month. That’s a 100x cost reduction with (if the agent is well-built) 80% of the value.

The first wave of AI products asked “what can AI do?” The next wave asks “what do people actually pay for?” Those are very different questions, and the second one leads to much better businesses.

Why Character Matters

One of the least-discussed and most important decisions in building Martine was giving him a name and a personality.

This might sound like a branding exercise. It’s not. It’s an architecture decision.

When you give an AI agent a character, three things happen:

1. Users form a relationship. People don’t form relationships with dashboards. They form relationships with entities that have names, preferences, and consistency. Martine has opinions. He’ll push back if you’re doing pasta for the fourth night in a row. He’ll suggest something you haven’t tried. He remembers what your family liked last time. This isn’t a feature. It’s the foundation of retention.

2. The interaction model becomes natural. Nobody needs instructions for how to talk to Martine. You talk to him the way you’d talk to a friend who’s a great cook. “Hey, what should I make tonight? I’m exhausted and my kid won’t eat anything green.” That’s not a prompt. That’s a conversation. The learning curve is zero.

3. Trust compounds. Every time Martine gets it right (suggests a meal the whole family eats, remembers your partner’s allergy, nails a recipe for a Wednesday when you’re too tired to think) trust increases. And trust is what keeps people paying $29.99/month. Not features. Trust.

I think every vertical agent will eventually converge on this: a named character with domain expertise, persistent memory, and the ability to take action (not just suggest). The agents that feel like tools will lose to the agents that feel like people.

The Technical Architecture That Makes This Work

For the builders in this group, here’s what’s under the hood.

The naive approach to an AI meal planner is: take a foundation model, give it a recipe database, add a system prompt, and let users chat. This works okay for single-turn queries (“give me a chicken recipe”) but falls apart for the actual use case.

What Martine actually needs to do:

Persistent memory across sessions. He needs to remember that your daughter is allergic to nuts, your son only eats five things, and your partner started keto last week. This isn’t conversation history. It’s structured knowledge that persists and updates.

Multi-constraint optimization. “Give me a dinner that’s nut-free, keto-friendly, uses chicken thighs, takes under 30 minutes, and includes at least one thing my picky 7-year-old will eat” is a constraint satisfaction problem, not a text generation problem. The model needs to reason across multiple overlapping requirements simultaneously.

Action execution. Martine doesn’t just suggest. He builds a shopping list and sends it to a grocery delivery API. This is the difference between an agent and a chatbot. An agent takes action in the real world. The shopping list that appears in your Instacart cart is where the magic happens.

Taste modeling over time. The more a family uses Martine, the better he gets. Not just at respecting restrictions, but at understanding preferences. You tend to like Mediterranean flavors on weeknights, you do more adventurous cooking on weekends, your kids are more willing to try new things when it’s presented as a build-your-own situation. This is collaborative filtering applied to a family’s palate.

None of this is theoretically novel. But wiring it together into a product that feels like talking to a knowledgeable friend? That’s the hard part. And it’s the part that generic AI wrappers don’t do.

The Market Signal Nobody’s Talking About

Here’s a number that should get every builder in this group thinking: families spend $400-600/month on HelloFresh and similar meal kit services.

They’re not paying for ingredients (grocery stores sell those cheaper). They’re not paying for recipes (the internet has millions for free). They’re paying to not have to decide. They’re paying for the cognitive relief of having someone else answer “what’s for dinner?”

HelloFresh is a $7 billion company built on removing one daily decision from people’s lives.

Now imagine removing that same decision for $29.99/month instead of $400. No physical supply chain. No ingredient waste. No boxes on your porch. Just an AI agent who does the thinking, and works with the food you already have or sends what you need.

That’s not a 10% improvement. That’s a structural disruption of a multi-billion dollar market.

And this pattern repeats everywhere. Look at any service industry where people pay a premium for personalized human attention, and ask: can a vertical agent deliver 80% of the value at 1% of the cost? If yes, that’s a company.

What I’ve Learned Building This

A few tactical observations for anyone in this group building (or thinking about building) vertical agents:

Start with the Sales Safari, not the model. Before I wrote a line of code, I spent weeks in Reddit threads, Mumsnet forums, and parenting Facebook groups reading how people actually talk about dinner. The language they use (“relentless,” “dread,” “I’d rather clean bathrooms”) became the language of the product. The pain points they describe became the features. Build for real pain expressed in real words, not for what you imagine the market wants.

The moat is in the workflow, not the model. Anyone can plug GPT-4 into a chat interface and call it a meal planner. The moat is in the integrations (grocery delivery APIs), the persistent memory (family preference modeling), and the character (Martine’s personality and trust). Models are commoditizing fast. Workflows aren’t.

Name your agent. I said this above but it bears repeating. The moment we named Martine and gave him a personality, everything about the product snapped into focus: the UX, the marketing, the retention strategy, the voice. “Ask Martine” is fundamentally different from “use our AI meal planner.” One is a relationship. The other is a feature.

Price like a service, not an app. If you price your vertical agent at $5/month, you’ve told the market it’s an app. If you price it at $20-30/month, you’ve told the market it’s a service. The positioning matters more than the number. We landed on $29.99/month (with a founding member deal at $12.99 locked forever for early adopters) because Martine isn’t competing with Mealime at $3/month. He’s competing with HelloFresh at $400/month. That’s the comparison we want customers making.

The Future Is Vertical

The next wave of meaningful AI products won’t come from building better foundation models. They’ll come from builders who deeply understand a specific human problem and wrap an AI agent around it with the right domain knowledge, the right integrations, the right personality, and the right price.

General-purpose AI is infrastructure. Vertical agents are products. And products are what people pay for.

The rich have always had access to personalized human expertise: chefs, advisors, coaches, tutors, assistants. The cost of that expertise has kept it out of reach for everyone else.

AI agents change that equation permanently. Not because the AI is as good as the best human expert (it’s not, and it doesn’t need to be). But because it’s available at 6 PM on a Tuesday when you’re exhausted and your kids are hungry and you just need someone to tell you what to make for dinner.

That’s not a toy. That’s a revolution. And we’re just at the beginning.

I’m Tim, and I’m building Dinner Solved AI, an AI personal chef named Martine for parents who are tired of “what’s for dinner?” The first 100 founding members get $12.99/month locked in forever (regular price is $29.99, still a fraction of what a personal chef or meal kits cost). I’d love feedback from this community.

And if you’re building a vertical agent of your own, I’d love to swap notes. DM me or drop a comment.

The Personal Chef Test: Why Vertical AI Agents Are the Real Product Revolution