In partnership with

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.

Get Your Report

I don’t start my day by checking email anymore.

At 7am, Jarvis checks in with me.

It sends a full briefing:

  • Pipeline movement

  • Inbox triage

  • Deal risk scores

  • Content queue status

All processed overnight. All ready before I touch a keyboard.

This is Part 1 of a series. I’m unpacking what I built, what’s next, and how you can build your own version from scratch.

💡 - This is not a one size fits all solution. The AI I built might be totally different from the AI you build. For ideas on how to use your own personal AI assistant, ask yourself this question: "what is one task that you generally perform at a computer that you would like to never have to do again?”

Unless the answer is Zoom calls, it can probably be automated.

Jay

What Jarvis is

This project started as an email triage tool. One feature. Annoying problem. Simple fix.

It is not that anymore.

Jarvis is personal AI operating system.

It manages my sales pipeline, generates content in my voice, triages my inbox, scores my deals with MEDDPICC, researches prospects, monitors my stock watchlist, and runs 22 automated background jobs without me touching anything.

All of it lives inside:

  • A single Telegram chat

  • A custom dashboard (when I need a screen) that I can truly “live” in.

Keep reading to see what has shipped so far

The interface

So how do I communicate with Jarvis, you might ask. I can currently communicate with Jarvis in the following ways:

  • Telegram

    • 40 commands

    • Voice notes too (I can talk to it, it transcribes, then acts)

  • Email

    • Jarvis has his own email account, so I can email him

    • This is especially useful if you for emails that contain information you might want to hang on to, but do not have a purpose for it now.

  • Text

    • Jarvis has his own phone number

    • I can text him, and maybe more importantly… he can text me (as well as anyone else that I want him to…. prospects, my wife to let her know I’m running late, etc.)

  • Dashboard

    • I have a “live" webpage (password protected, of course) with an obscure web address.

    • This is the primary screen that I work out of all day long. I can check/send email, check and add to the calendar, view either my or Jarvis' Google Drive, generate ideas for my content, work my deals (including CRM updates), run email campaigns, manage my Slack channels, pull signals from prospects, and literally so much more.

  • One app

  • Any device

  • No new tools to learn

Email OS

This delivered the fastest return.

  • 7am Morning Briefing: inbox summary, calendar events, open tasks, and drafted replies waiting for review

  • Email triage every 30 minutes: unread emails categorized, ranked, and reduced to the top 5 that actually require action

  • Proactive reply drafting: Jarvis writes responses in my voice before I ask

  • One command to approve and send

I now spend about 10 minutes in my inbox.

It used to be 45.

Sales OS

This is where most of the work went.

Jarvis replaced my CRM for tracking deals. I migrated out of Notion in February. Everything lives here now.

  • Seven pipeline stages

  • Full history on every deal

What it does on the sales side

  • MEDDPICC scoring on every active deal

  • Deal velocity tracking (average days per stage, where things are stalling)

  • Gap coaching with specific questions for each deal

  • 3-touch follow-up sequence generation from deal context

  • Contact enrichment via Apollo and Clay (title, company, tech stack, intent signals)

  • Competitor research via Perplexity

  • Win/loss logging with reasons and lessons

Most of this runs from one Telegram command. Some runs automatically.

Content OS

I produce a lot of content.

  • LinkedIn posts

  • Newsletters

  • Short-form video

Jarvis generates all of it in my voice. Not a generic voice. My actual voice.

I gave it 15 files of style guides, voice rules, exemplar posts, and anti-patterns. The output comes back nearly ready to use - but sometimes I have to jump in and put my own personal touches in.

What has shipped on the content side

  • 10 hooks per topic, written to my exact cadence

  • Full LinkedIn post expansion from a raw idea

  • Newsletter draft agent that runs every Sunday at 5am

  • Content pipeline tracker across platforms

  • YouTube video ingestion: paste a URL, get a KB summary and two LinkedIn drafts

  • Performance tracking and content gap detection

The newsletter you are reading right now was drafted by Jarvis. I edited it.

Proactive automation

22+ scheduled background jobs. No input required.

A few examples:

  • 7am weekday: async standup check-in

  • 8am weekday: buying signal scan across my prospect watchlist (job changes, funding rounds, trigger events). I only get notified when there is positive movement.

  • 8:30am weekday: stock watchlist report

  • 9am daily: deal contact warmth tracker

  • 5:30pm daily: habit tracker check-in (outreach, content, deep work)

  • 5pm Friday: weekly scorecard (revenue, content, habits)

  • Sunday 5am: newsletter draft synthesis

I wake up and the briefing is already there.

I end the week and the scorecard is already waiting.

This is what makes it an operating system and not just a chatbot.

Knowledge base

Nearly 300 markdown files power everything above.

  • Voice guides

  • Style guides

  • Deal playbooks

  • Framework indexes

  • Webinar summaries

  • Competitive intelligence

When Jarvis generates content or researches a deal, it routes to the relevant files first.

That knowledge base is why the output sounds like me.

Without it, you get generic AI output.

With it, you get something that holds your voice across every tool.

What is coming next

Four things actively in progress.

  • Chief of Staff mode: more autonomous. Close-date risk alerts, market intelligence briefings, daily agenda building. Less “answer when asked,” more “flag this before you think to ask.”

  • Auto-repurposer: approved LinkedIn post automatically reformats for other platforms.

  • Mobile layer: a mobile-native interface. Early, but in progress.

  • Multi-agent: today it runs single-threaded. Moving toward parallel agents handling different workflows at the same time.

Where this goes by end of year

If the trajectory holds, Jarvis operates without commands by year-end.

  • No “draft me a follow-up.” It notices the deal went quiet and messages me with a draft already written.

  • No Salesforce. No HubSpot. Full pipeline management inside a system I control, with AI-weighted forecasting that updates after every interaction.

And the architecture that runs my sales OS can run:

  • An operator’s back office

  • A marketer’s content stack

  • A recruiter’s sourcing engine

The tools change. The pattern does not.

How to build your own version

I’ll go deep on this in future issues. Here’s the five-step mental model to get started.

Step 1: Pick your interface

Telegram.

  • Bot API is simple

  • Works on any device

  • Voice notes are built in

Your assistant lives in a chat, not another app.

Step 2: Pick your brain

Claude (Anthropic).

Specifically for tool execution. Sonnet or Opus handle tool calls, reasoning chains, and context-aware responses without custom training.

Step 3: Define the tool pattern

Every capability is a “tool”:

  • Name

  • Description

  • Input schema

You give Claude the list. It decides which tool to call based on what you said.

This is Anthropic’s tool-use pattern.

Once you understand it, every new capability is just adding another entry.

Step 4: Host it somewhere that runs 24/7

Railway.

Push to GitHub, it deploys.

About $5/month.

Your automations run while you sleep.

Step 5: Start with one thing

Pick the most painful daily task.

  • One tool

  • Use it for two weeks

  • Then add the next one

That’s how this went from email triage to what it is today.

Paid subscribers get the actual starter skeleton below.
The file structure and core tool pattern that powers everything above. If you're going to build this, start there instead of from scratch.

Talk next week.

The Jarvis Skeleton

Quickstart for paid subscribers | sellingwithai.vip

1) The starter skeleton

The actual file structure and a barebones version of the core tool pattern.

It’s a working Express server with:

  • A JARVIS_TOOLS registry wired up

  • One example tool implemented end-to-end

You clone it, add your API keys, and start building.

2) The blueprint doc

Not code. Decisions.

Both are linked for paid subscribers below.

Reply

Avatar

or to participate

Keep Reading