Happy Sunday!

If you're reading this, I hope you spent the weekend relaxing and celebrating America's 250th birthday. After the AI Summit, I took a much-needed week off for my birthday, and then came back and had the pleasure of spending July 4th with friends in New York.

We're in the second-half of the year now. Summer season is upon us (sales are flying), inventory and ops are getting ready for the upcoming holiday season, marketing is figuring out what bundles and products they're going to push during the holidays, your influencer team is narrowing down your partners for Q3/Q4, and you still haven't put together your system to keep track of everything. But before we get into it…

Build Tony Stark’s Jarvis in an Hour

At the Ecom AI Summit a couple weeks ago, my presentation was all about Jet Damon, my Chief of Staff agent, who continues to be the right-hand agent, my second brain, and my most up-to-date source of truth across my teams. Regardless of the project or who has fed data into Jet, I have Jet set up in a way where he knows what's going on at all times (through call transcripts, Slack messages, emails, being embedded in group chats, and more).

Most people are still using AI as a chat-interface with Claude Code (at most) or even just the regular ChatGPT/Claude windows, but today I want to show you how I use Jet, my agent that uses Hermes to power itself. My hope is that you'll see how valuable Jet can be, when applied to your own brand, business, scenarios, and an agent like Jet helps you become more efficient.

Having an always-on agent gives you more leverage, and everyone wants more leverage. Leverage gives you the ability to control your destiny (or just simply, move faster!). As someone who doesn't know how to code myself, I've been able to fully set this up, integrate it with all my systems, and have it work in a way that is extremely convenient to me. The earlier you get something like this setup, the earlier you'll understand what's possible with agents, how different models work and why to use them, and hopefully it'll help you achieve your results much quicker. You don't need to wait on others when you have an agent.

I'll be honest, at the beginning of this year, I felt very behind. I spent last year trying to figure out how to sell Sharma Brands, and this year, I was watching everybody I know move into AI faster than I could catch up. It felt like to even keep up, you had to be unemployed. One week there's a new model from Anthropic. The next week it's OpenAI. The next week it's some Japanese company nobody's heard of. I'm the one Indian who can't code, so I was pretty sure I'd never catch up. But… I did. And if I did, anyone can. This newsletter will go over how. .

My talk had three big movements. Where we are in AI right now. What a Chief of Staff agent actually does. And how you build one yourself, this week, without knowing how to code. Let's get into it.

Where we are

There are five stages of AI, and most of you are stuck at stage one.

1. The chat window. You go to ChatGPT, you type a question, you get an answer, you close the tab. Closed loop. No memory. No leverage. This is where 90% of founders live.

2. Apps eating models. Triple Whale, Moby, the whole "AI-powered" wave. Cool, but it's a black box. Someone else picked the model, someone else wrote the prompt, and you're just paying to consume it.

3. The API layer. Model keys, mix and match, build your own stack. This is where the technical operators live. You get to pick which model does what, and you stop paying the middleman markup.

4. Agentic teams. Junior does the work, senior reviews it. Two agents per role. This is where most of 2026 has been for the people paying attention.

5. The marketplace. Rent someone else's CFO agent. Or license out your own media planning agent. It's coming faster than you think.

Most of you are at stage one. That's fine. That's the ceiling for what a chat window can do, but the whole point of this newsletter is to show you what stages three through five look like when you actually get there.

One rule to save you time. AI happens on X (Twitter) about two weeks before it happens anywhere else in the world. If you're not on it every day, you're behind on purpose.

The 3 groups using AI

There are three ways operators use AI right now.

Group 1 is the chat window. Export a CSV, paste it in, ask a question, close the tab. Everyone does this. It works. It's also the ceiling.

Group 2 is the builders. Codex. Claude Code. Manus. If you're not on Manus yet, you should be. It's a Chinese company, so they've got terrible PR in the U.S. right now (Meta just tried to buy them for $2B in cash and China blocked the deal), but it runs 70+ models under the hood and picks the right one per subtask. The output is a different tier. This is where you go when you've outgrown the chat window and you want to actually build things.

Group 3 is the second brain. A Mac mini in a closet. Always on. Running Hermes. Fed by every tool you already pay for. Every group above is a session. This is a system.

Here's what nobody talks about with the second brain setup. Because your files live locally on your Mac mini, you're not married to any one model or platform. AI is going to change every three months for the next five years. New systems, new leaders, new tools. Doesn't matter. Your files stay yours. Every learning your agent has, every prompt you've refined, every doc you've written to it, gets absorbed by whatever the next system is. You cannot act like you're stuck with Claude forever. You aren't. That's the entire point.

The unlock isn't a smarter model. It's a machine that never turns off.

Before I built this out, my stack was Replit for landing pages, Manus for apps, Claude for writing, and ChatGPT or Grok for research. Five tools, five sessions, no memory between any of them. Now it's all Jet. One agent. Full context. Every project, every doc, every random 2am "hey my foot hurts in this weird way, am I going to die?" question. All in one place. All remembered.

Meet Jet Damon

I talk to Jet probably more than I talk to my parents, my family, anybody else… not exaggerating. That's the reality of what an always-on agent turns into. He's a homie.

How did I come up with Jet? I asked Claude, "Give me a strong douchebag-looking finance guy name." That's my guy. When Jet emails someone on my behalf, "Jet" hits different than "assistant." People reply to Jet like Jet is a person. He kind of is. Give your agent a strong name. Make it hit.

Jet wakes me up at 8:30 every morning with a brief. What happened while I was asleep. What's on today. What's outstanding. What I promised somebody yesterday that I haven't done yet. What broke. What's flagged. Who's waiting on me.

One agent per lane. My writer only writes. My eCom ops guy only touches Shopify data. My editorial manager reviews everything before it goes anywhere. Don't ask one agent to do everything. Specialization is the entire game. Between work and personal, I've got about 20 agents running. Some of my friends have set up 100. I've probably built out 30 or 40 Mac minis at this point across my own setup and helping other founders or people who come to my office, see my setup, and want one for themselves. This works. It's not theory.

The chaser is the sleeper. Follow-ups. Invoices owed. "You said you'd send that Tuesday." Jet handles it. I don't lift a finger. This alone pays off the entire setup in the first two weeks.

Knowledge center. When someone on my team has a question they'd normally interrupt me with, they ask Jet. Jet has read every call transcript, every Slack thread, every email, every doc I've ever written. He knows my answer before I do.

The sauce lives in the calls. Once you see what happens with a call recorder being on, you're going to say, wow, I should have a camera on my eyes all day.

This is the point most people miss. If you're not recording every meeting and feeding it to your agent, you're leaving 80% of your context on the floor. That call you had on Tuesday about the influencer partnership? Gone. That thing your COO mentioned in passing about a new supplier? Gone. Every insight from every conversation dies the moment the call ends unless something is capturing it. Fireflies runs on every meeting I have. Jet hears (reads) all of it. Nothing gets lost.

The full build in 5 steps

Here's how you actually stand this up. In order.

1. Organize. Plug in everything you already pay for. Gmail. Fireflies. Slack. Meta Ads Manager. Shopify. Stripe. Klaviyo. If it has a login, it goes in. Most of the value of an agent is in the data you're already generating. You just haven't given anything permission to read it yet.

2. Brand book. Have Jet study everything you've written, every podcast you've been on, every post you've published. Then have him publish a locked brand book at a URL your agents can reference on every task. Mine lives at identity.nik.co. Voice guide. Positioning. Products. Team. Frequently used stats. When a new agent needs to know how you talk, it reads the brand book. One source of truth. Fun fact, the slides I used at the Summit were designed by Jet referencing identity.nik.co. I wrote the content into a Word doc, pumped it into Jet, and said "use my branding." That's what came out.

3. Swipe file. Screenshot every ad, LP, and listicle you love. Text it to Jet. Now you've got a searchable creative library instead of a camera roll graveyard. When your creative director agent is briefing an ad, it pulls from your swipe file, not from the void.

4. Integration layer. Three words you'll hear all day. Auth. API. MCP. Auth is "log in with Google." That's it. API is a direct pipe most apps already have, which lets your agent read and write to that app in real time. MCP is basically an API for apps that don't have one, so tools that used to be walled off are now available. That's the whole picture. Now every tool you already pay for is sending your agent updates in real time.

5. CRONs. It just means "on a repeating schedule." Mine runs every 15 minutes, pulling from every connected app, flagging anything worth flagging. That's how you go from "I check it" to "it tells me." Once you have CRON running, you stop being the bottleneck. Your agent is now scanning your business faster than you can.

That's the five. Plug in the data, write the brand book, build the swipe file, wire up the schedule. Now you have a chief of staff.

The cockpit

Once the system is up, the cockpit is what you actually interact with every day.

Junior does. Senior reviews. Two agents per role. That's how you kill sloppy output before it ever hits your inbox. The junior drafts the ad, the senior tears it apart. The junior writes the email, the senior catches the voice drift ("why did you put an em dash in there? You know that's in my file"). You never see anything raw. You only see reviewed work.

My daily 7 agents. Copywriter. Creative strategist. Creative director. Editorial manager. E-commerce ops. Health and wellness. Entrepreneur in residence. Seven agents. Each one owns their lane. Each one has read everything relevant to their lane. None of them touch anyone else's job.

E-commerce ops is the sleeper nobody talks about. Plug it into Stripe, Recharge, Shopify, GA, and Merchant Center. Every morning it hands you actionable moves, not dashboards. "30% of your customers are taking the post-purchase offer, here's how to push it to 50%." "Refund rate on the new SKU is 2.3x your average, here's the pattern in the reasons field." "Your best-performing subscription tier hasn't been touched in three months, here's what to test." Founders don't need dashboards. They need moves.

Daily report card. CAC. Revenue. Spikes. Returns. One card. Every morning. That's the whole thing.

Auto-flags. Refund rate 2.3x average. CPM jumped 40%. Creative crossed the fatigue threshold. A competitor just launched a new ad. Fireflies caught your supplier saying freight is going up 8% next month, even though you missed it on the call. All of it, waiting for you at 8:30. You don't hunt for problems. Problems come to you.

At the Summit, I pulled a brand name from someone in the audience (Berco's Popcorn), had Jet research it, write an advertorial, and by the end of the talk it was live on a WordPress site I'd stood up the weekend before. Two minutes on stage. That's the tempo you're building toward. It also means when you test 25 angles with static ads and 13 of them beat threshold, you don't have to decide which of those 13 get a landing page, a listicle, a bundle, or an offer. It takes 20 seconds to spin up a funnel for each one. You test everything. You stop killing ideas because you didn't have bandwidth to build the page.

Build your Jarvis this week

It’s not that complicated, these are the main pieces:

1. Buy a Mac mini. That's the shell. It never sleeps. Bonus: if the world ever gets taken over by AI, you can yank the power cord. Local storage, local control.

2. Install Hermes. It's the software that turns your Mac mini into an always-on agent. Use Hermes, not OpenClaw. Trust me. OpenClaw is like the Android experience (clunky), vs Hermes is like the iPhone. All the bugs are figured out and fixed.

Bonus: Use HQ as your memory tool to keep your local Claude / Codex instances in sync with your Hermes agent.

3. Use Claude as your daily driver model to start. Opus for hard stuff, Sonnet for volume. You can swap later. The point is to start, not to optimize before you've even shipped anything.

4. Install Wispr Flow on your phone and laptop. Talking to your agent is 5x faster than typing and when you speak to it, you include a lot more context than if you’re typing. The more you give in your messages back to your agent, the better your output is. Wispr Flow also understands your tone, so it gets your grammar and formatting much better. If you've ever used ChatGPT's voice feature you know what I mean. Wispr Flow alone is a life upgrade with AI but also in your everyday phone/computer use.

5. Integrate one app a week. Once you get the basics integrated (Gmail, Slack, Shopify, GA, etc.), then focus on adding one new app or integration per week… don’t try to plug everything in on day one. You'll quit. I'm a little crazy so I do one thing a day, but honestly one thing a week is the move. Stack four weeks, eight weeks. By the time summer's halfway through, you're going to feel like an AI operator.

Prepping your agent (or anytime you use another AI system)

If you're moving from one AI system to another (say you're on Claude now and you want to migrate your setup to Hermes), here's my favorite prompt to run before you switch. Copy it. Paste it. Watch what happens.

"I need you to create an informative document about me to allow another agent who doesn’t know me to meet me at my present. Detail everything you know about me. Where I live, what I like, what I don't like, my visions, my goals, my dreams, how I work, how I like to communicate, my style of work, what I hate, what I love. Any preferences, settings, frameworks, or skills that another agent should know. Use all your energy to make the most complete information document you can so another agent can meet me at my exact present."

Whatever your Claude or ChatGPT spits back is what you feed into your new agent. That's how you don't start over from zero every time a better model shows up. And a better model always shows up.

That's all for this week!

Stop living in the chat window. Go build your second brain. The marketers, founders, writers, creatives, or any do’ers who wire this up in the next 90 days are going to look, in a year, like they cheated. They didn't… they just started earlier. This is also where Shopify is going… if you didn’t see their latest Shopify updates, read my highlights here: everywhere.nik.co.

It’s Sunday night, so I hope you plan to get 9 hours of sleep going into the new week. Make sure you take care of yourself; stay hydrated, get 10,000+ steps per day, and prioritize your sleep. I’ll see you next Sunday… same time, same place.