You don't need to work faster

Everything you run is a pipeline.

Targeting, enrichment, copy, sending, reading results and deciding what's next - all pipelines, a series of steps that turn an input into an output.

You try to run those pipelines faster - type quicker, add another tool, hire another person.

It never quite works, because the bottleneck was never speed.

The bottleneck is that you're hand-executing work that should be wrapped in a harness and handed to an agent.

A harness is your judgment, written down once, so an agent can pass through it every time instead of you deciding fresh each time.

Here's what a harness looks like in practice.

Before I ever ask an agent to write a cold email, it already has my ICP, three top-performing subject lines, a rule that every email names one specific situation, and a ban on the word "solutions."

Ask a blank agent for a cold email and you get the average of every cold email on the internet.

Ask mine and it writes at the standard I've already decided on, every single time.

I'm using the same model everyone else has access to. The harness is the only difference.

Do this today: open a notes file, and write down one rule you'd hand a smart new hire on day one.

Something like "never send a template that doesn't name a specific trigger."

That's a harness rule.

You have one now.

Tomorrow we feed it context.

Reply and tell me the rule you wrote.

I read every one.

PS: Miyamoto Musashi wrote the Book of Five Rings near the end of his life, not to boast, but so his standard could survive without him in the room. A harness does the same job with a keyboard instead of a sword: your judgment, written down, so the work still happens at your level when you are not the one holding the pen.

Why your AI copy sounds generic

You can lead a horse to water, but you can't make it drink.

You can give AI what to write and it won't know how to do it. Give it how to write and it won't know what to say.

You need to treat AI like a new hire on their first day.

A new hire learns from the mistakes of the last time they did the task. With AI, you need to set up the same system around it, one that makes sure it pulls the right information before it writes anything.

That system is the harness.

Yesterday you wrote one harness rule. Today you write the document the harness reads first.

Most people hand an AI facts: their market, their product, their pricing. Facts are not what's missing.

What's missing is your opinions. The calls a fact sheet can't hold: which subject lines you'd never send, why "solutions" is a banned word, what a good first line sounds like versus a bad one.

That's a different document. Not a context file you write once and file away. An opinions file - call it taste.md - that holds how you think about the calls that make copy good or bad.

Start it today, small:

• One line you'd never send, and why
• One line you'd always send, and why
• One rule you'd give a new hire before their first draft

Here's the difference from a fact sheet: a fact sheet is done the day you write it. Taste.md is never done. Every time you correct an AI's draft, that correction is a new entry - what changed, and why - and it goes in the file.

That's what makes the harness get better instead of staying flat. Not more facts. More of your judgment, written down, one correction at a time.

Do that today. Every correction you make from here goes into this file, not just your head.

I never started a run from a blank prompt either. Every campaign I have run began from a written context file like the one you built today. [NEEDS REAL NUMBER/CASE - pending Mitchell's sign-off on what real case, if any, replaces this line]

Reply and tell me the first opinion you wrote down.

PS: Napoleon read every scouting report before he moved a single unit, not after. He refused to commit to ground he had not seen on paper first, and that discipline is what beat marshals just as sharp as he was. Taste.md is the same discipline: check what you already decided before you write the next line, instead of deciding it fresh every time.

Stop re-deciding what already worked

Teams throw away their own sharpest asset every quarter.

You run a campaign. It works.

Three months later, you sit down to write the next one and start from scratch, re-deciding things you already learned the hard way. The knowledge was real, but it lived in your head or in a dead Slack thread, so it may as well not exist.

The fix is a knowledge graph. Stripped down, it is a document of what worked, written so you can query it instead of remembering it.

Start with one page. Every time a campaign does well, add a row:

• the situation the prospect was in
• the subject line and angle you used
• the reply rate
• what you think drove it

No software required to begin. A single markdown file will do.

The power shows up on the fourth or fifth entry. You stop guessing "what should I try" and start asking "what has worked when I targeted a company that just raised a round" - and the file answers.

You are querying your own experience instead of re-deriving it.

This is the difference between a team that gets better every month and one that runs the same experiments forever.

At scale, this is exactly what we run at LeadGrow: 1,626 campaigns of pattern recognition, encoded so it can be queried in seconds instead of remembered. You do not need 1,626 to start. You need one page and the discipline to add a row every time something works.

Open that page today. Add your first row from a campaign you already ran.

One more thing before tomorrow: I queried 1,626 campaigns of encoded patterns before writing a single line of copy. I was standing on a memory my competitors had thrown away.

Reply and tell me what your first row was.

PS: Marcus Aurelius kept a private notebook of what held up under pressure. We know it today as Meditations. He was not writing for an audience. He was building his own record of what worked, so he would not have to relearn it in the middle of a crisis. That is a knowledge graph with a stylus.

Firmographics are not a target list

"Series B, 200 employees, VP of Sales."

That is not a reason to email someone. It is a filter.

You build target lists almost entirely from firmographics - industry, headcount, revenue, title. They tell you a company exists. They tell you nothing about whether that company needs you right now.

The thing that predicts a reply is situation - what just changed in their world that makes your product relevant this week.

A company that just raised a round is hiring and has budget. A company whose main competitor just got acquired is nervous and looking. A company that just posted twelve open sales roles has a scaling problem you might solve. Same firmographics, opposite readiness to buy.

So build the list on situations. Here is the exercise: write three situational target queries for your business. The format is "companies that [just did a specific thing] and therefore [have the problem I solve]."

• "companies that posted a new Head of Revenue role in the last 30 days and therefore have no outbound engine yet"
• "companies that switched off a competitor tool this quarter and therefore need a replacement"
• "companies that just opened a second office and therefore need to fill a new pipeline"

Each of those is a trigger you can go find - in job boards, funding feeds, review-site switching signals, news.

Firmographics narrow the universe, but situations tell you who to email first.

Write your three today. Keep them. This is the input to everything downstream.

Good copy sent to the wrong list still fails. Average copy sent to a company in the exact right moment still works.

Targeting is the highest-return decision you make.

PS: Philip of Macedon rarely fought the battle sitting in front of him. He waited for the one where the ground, the timing, and the enemy's position had already done half his work. Situational targeting works the same way: wait for the trigger, not the title.

Enrichment in minutes, not days

GTM engineers lose a week here: enrichment.

You have a list.

Now you need work emails, verified.

You need the trigger that put them on the list confirmed.

You need enough about each person to write something that is not a mail merge.

So you run one tool, get 60% coverage, export the misses, run a second tool, export again, and three days later you have a spreadsheet held together with tape.

The move is a waterfall.

Instead of running data sources one at a time by hand, you chain them.

The leads a cheaper source can't resolve fall through automatically to the next source, and only the true misses reach the expensive one at the bottom.

Design it as three steps.

• Step one: cheapest, broadest source - resolves the easy majority, the rest falls through
• Step two: stronger, pricier source - catches most of what's left
• Step three: expensive, high-accuracy source or a manual check - runs only on the small set nothing else could find

You pay top rates on only the hardest few leads instead of the whole list, and coverage climbs toward complete without you babysitting exports.

Your task today: sketch a three-step waterfall on paper for one field you need - work email, say.

Name the source at each step and the order they fire.

You do not have to wire it up yet.

Once you can see it as a chain, the manual version starts to feel insane.

This is the step that turns targeting into a real, usable list.

Skip it and your perfect target queries sit in a spreadsheet with half the emails missing.

He wired his own enrichment as a waterfall exactly once. He reran it in minutes on every new batch since.

PS: If you want to see a real waterfall running before you build your own, look at how a tool like Clay chains enrichment providers together - the clearest working example of the fallback logic in today's email.

Teach the AI to write like you once

You have written a great email before.

The problem is you cannot reliably write it again, and you definitely cannot get AI to.

Every time you prompt from scratch you are rolling the dice.

Sometimes the copy lands, sometimes it is stiff and templated, and you are the only quality gate standing between the two.

That does not scale past you.

And the cost of getting it wrong went up.

The reflex fix for slow copy - mass-generate it with AI and blast - is now getting punished in the open.

"AI-first" brands are starting to fall flat, and readers tune out the moment something reads machine-written.

Volume without judgment has become a liability, not a shortcut.

The winning move now is applying your judgment to every draft you send, not cutting it to move faster.

The fix is a skill file. It is the harness from day one, aimed specifically at copy: a document that encodes your writing standard so the agent produces your voice by default instead of the average AI voice.

Build it from evidence, not opinion. Pull your single top-performing email - the one with the reply rate you would brag about. Then write down the rules that made it work:

• The structure - specific situation in line one, one idea per email, under 90 words
• The voice - short sentences, no corporate words, the exact tool or number instead of a vague claim
• The bans - which words and openers you never touch, plus the fake-personalization tricks that get you marked as spam
• The proof pattern - how you cite a result, naming the actual client and the actual number

Now every copy request references that file, and the agent writes inside your rules instead of guessing. You have turned one good email into a standard the machine can hit every time.

Skill-file copy is how we hold a 4.1% average reply rate across our campaigns - roughly three times the industry norm. The standard, written down and applied every time, is what does that - not luck, not one gifted writer.

Today: pull your top email and write five rules that explain why it worked. That is your first skill file.

His copy came from a written standard, not a fresh prompt every time - a standard applied consistently, not inspiration striking twice. [NEEDS REAL NUMBER/CASE - if a proof point belongs here, it needs Mitchell's sign-off on which verified result to use.]

PS: Hemingway held himself to a short, unforgiving list of his own rules: short sentences, no wasted adverbs, cut anything that did not earn its place. He did not rediscover them every morning. He wrote them down once and worked inside them. That is a skill file for a typewriter.

One message to everyone is a mistake

The fastest way to waste a good list is to send it all the same email.

Your list is not one audience.

The company that just raised a round is in a different headspace than the one that just lost a competitor.

A VP of Sales reads a different first line than a founder.

Blast one message across all of them and you speak precisely to no one.

The move is segment, then deploy.

Before you send, split the list by the situation that put each person on it, and ship a variant written for that situation.

Here is the checklist I run before any deploy:

• Segment by trigger - group by why they qualified: funding, competitor switch, new hire, expansion
• One variable per variant - swap the first line and the proof point, keep the offer and the ask the same
• Match the proof to the segment - cite a result that looks like the reader's own situation
• Stagger the send - wave it out so one deliverability problem doesn't hit the whole list
• Set the read date now - decide today when you'll look at the numbers, before you send

Run that checklist and one list becomes four or five sharp campaigns instead of one blurry one.

Today: take the list you have been building all week and split it into segments by trigger.

Even on paper.

That split is the whole lesson.

PS: Napoleon split the Grande Armee into corps that could each fight their own battle, then converge when it mattered. He never threw one undivided mass at a mixed battlefield. Segmenting your list before you deploy is the same move, scaled down.

Your numbers are trying to tell you something

You look at a campaign report and feel something - good open rate, bad reply rate, a vague sense of whether it worked - then move on.

That feeling is not information. The report is full of your next move, but only if you interrogate it instead of glancing at it.

Here are the four questions I run against every campaign before I decide anything.

• Where people dropped off
• Which segment carried the result
• What the winners had in common
• What the one change is for next time

Walk the funnel in order - delivered, opened, replied, booked.

The first place the number falls off a cliff is the thing to fix.

A great reply rate with no bookings is a call-to-action problem, not a copy problem.

Performance is almost never spread evenly.

Usually one trigger, industry, or title is doing most of the work - find it.

That's your next campaign.

Pull the ten replies you were happiest to get and read them together.

The pattern in who responded tells you who to target next, in their own words.

One change, not five.

Change everything and you learn nothing, because you can't tell which move mattered.

Answer those four and a pile of numbers becomes a decision.

Today, take your most recent campaign, even an old one, and answer the first question.

Where did people drop off?

That single answer is usually worth more than the whole dashboard.

He audited every single week and pivoted on evidence, never on how the week felt.

That discipline is the real gap between someone who runs one good campaign and someone who runs a pipeline that compounds all year. [NEEDS REAL NUMBER - do not restore "200+ meetings" or any Unity figure without Mitchell's sign-off. Verified case on file: Coherence ApS - 7,893 leads contacted, 654 replies (6.98%), 106 interested, 14 meetings booked, 5 closed won. No revenue figure verified.]

PS: Today's four questions are a shrunk-down version of the military after-action review - what happened, why, and what changes next time. Worth reading the real format once if you want the fuller version of this habit.

The habit that makes everything compound

A pile of pieces is not a machine.

You now have all seven layers - context, a brain, targeting, enrichment, copy, deploy, audit.

What makes it a machine is the one move you're skipping.

That's the reason this approach compounds instead of plateaus.

The move is the write-back.

You audit a campaign and learn something.

• this subject line dies
• this segment converts
• this proof point lands

Write that finding into your brain - the file the agent reads next time - instead of leaving it in your own memory.

The next campaign doesn't start from zero.

It starts from everything the last one taught you, already loaded.

Then it teaches you something new, and that goes back in too.

The system gets sharper every cycle, on its own, because you closed the loop.

This is the mechanism behind every email I've sent you this week.

A blank prompt gets a little better each time you use it, if you're lucky.

A harness that writes its own results back into itself gets better on a curve.

That's why the gap widens.

Two operators start at the same skill level.

One keeps prompting from scratch and stays roughly flat.

The other closes the loop, and six months later is not twice as good, but running an operation the first one can't imagine, because the machine has been learning the entire time.

The self-improving pipeline is the point, and everything else is plumbing that makes the loop possible.

Today: take one thing your last campaign taught you and write it into your context file as a permanent rule.

That one sentence is the loop, closing for the first time.

PS: Benjamin Franklin tracked his own conduct against thirteen virtues in a little book, every day, for years. He built that discipline by writing the results back into the system and reviewing them on a schedule - an old idea, rarely applied to a business.

Stop being the strongest hand on the rungs

For nine days you have been building a machine. Today I want to name what happens to you once it runs.

Right now, you are the technician on the rungs, climbing every step by hand - targeting, writing, sending, reading results, deciding. You are good at it. Being the strongest person doing the manual work means the whole operation is capped at how fast you personally can climb.

Look at what founders are posting right now. Burnout is the number one reason solo founders quit - 54% say so. The most common shape it takes is one sentence: "I built something people love, and I am too busy to market it." That is a rungs problem, not a discipline problem. You are the only one who can climb, so the climbing never ends, and the marketing is the first rung that gets dropped.

The shift is to get off the rungs entirely - hand them to the machine and move up to the job only you can do:

• setting the strategy
• holding the standard
• deciding what "good" means and what to build next

I call that General Mode. The machine works the rungs at the output of a full team, and you direct it. You stop being the person doing the work of five and paid like one, and become the person whose judgment is the only scarce input in the system.

That is a different job, not a productivity tip.

The technician asks "how do I do this faster." The general asks "what should we do, and to what standard." Everything you have written this week - the harness rules, the context file, the skill files, the loop - exists to move you from the first question to the second.

Assembling all of that yourself, from scratch, is a real project. The brain, the six commands, the skill files, the write-back loop, the infrastructure to host it. It works, but it is weeks of building before you write your first email.

So we built the whole thing and made it a command. It is called /legion, and it is the assembled version of everything this sequence has taught you - the productized machine, ready to run, so you get to General Mode in week one instead of month three.

Today: ask yourself one question. What is the highest-value thing you would do if the rungs ran themselves? That answer is your actual job. The machine exists to give it back to you.

- Mitchell

PS: By the time Napoleon was winning wars, he was barely drawing a sword. His job was reading the map and deciding where the weight of the army landed. General Mode is that job, at whatever scale you are running.

Own the machine or rent it forever

You rent this machine, or you own it.

The difference is bigger than you'd think.

Renting looks like hiring an agency to run your outbound, or stacking point tools that each do one slice.

It works, and it never stops costing.

• Agency retainer - around $10,000 a year, and you walk away with nothing
• One AI SDR tool - roughly $11,000 to $40,000 a year, for one function

That AI SDR number alone is more than the entire machine costs to own, for a sliver of what it does.

Owning looks different.

You build the brain once, and it's yours - skill files included.

The loop compounds into an asset that gets more valuable every week you run it, and it's portable - you can rip out your knowledge graph and take it with you.

You own the machine. Not just what it makes.

Priced conservatively, against named public comparables, the stack we've been building is worth about $51,300 a year - the brain, the deployment engine, a full production system every month, founder office hours, the room.

That's the low-to-mid bound on each line, not a hype number.

The real question is whether you want to rent your go-to-market forever, or own the machine that runs it.

One is an expense that resets every year.

The other is an asset that compounds.

Today: add up what you currently spend renting your GTM - tools, contractors, the agency retainer.

Hold that number next to the cost of owning the machine outright.

- Mitchell

PS: Rockefeller didn't stop at refining oil. He bought the pipelines too, because renting capacity from someone else meant someone else set his margin. Owning the machine, not just the output, was the real edge, and it still is.

The whole Unity workflow, laid out

I have been peeling one layer a day. Today the whole thing goes on the table.

Unity worked like this, start to finish.

• Loaded a context file with my market, ICP, and standard
• Queried a brain of past patterns before writing a word
• Targeted by situation
• Enriched in minutes through chained data sources
• Wrote copy from a skill file that held my voice
• Deployed across segments
• Audited every week and fed the results back into the brain

That is me running the output of a full team. [NEEDS REAL NUMBER - the Unity outcome (revenue, meeting count, team structure, acquisition) is unverified and disputed. Do not restore a figure here without Mitchell's sign-off. The verified case on file is Coherence ApS, not Unity - if that's the case to run instead, it needs its own correctly attributed line, not a swap into this sentence.]

This is not one lucky story.

The same machine built the first agency I ran to $1M in ten months.

It is the machine behind 1,626 campaigns and a 4.1% average reply rate, roughly three times industry.

It ran a Reddit SEO play to 133+ booked calls.

The brain underneath it cut a task from around 33 tool calls to 3 - a 91% token saving - validated across 150 split tests.

It authenticates in 25 seconds.

That workflow is what this sequence has been building toward: a single command-line tool. Six commands: init, target, enrich, copy, deploy, audit. Every one of them merges from your brain, your harness, and your context. Audit writes the results back so the machine compounds. It is called /legion.

Everything this sequence taught you to build by hand, /legion is the assembled version of, ready to run.

You start with my machine instead of reconstructing it from scratch.

- Mitchell

PS: Sun Tzu collected discipline that already existed scattered across a hundred campaigns and wrote it down in one place - he invented less than he assembled. Today's email is the same move: nothing new, everything you already learned, assembled.

I can't build this myself

Every time I show someone the full machine, the same thought lands.

This is real, but I can't build it, run it, or maintain the infrastructure.

So it isn't for me.

That fear is earned.

The free templates look great until you realize they don't fit your use case.

Every tutorial either assumes you're already an engineer or is three months out of date.

The ceiling was never the problem.

The ramp is.

/legion is built to kill that exact objection. Walk into this instead of a pile of parts and a wish of luck:

• The brain, already built - 1,626 campaigns of pattern recognition, queryable in seconds, made yours by feeding it your own work
• Infrastructure, optional - self-host free, or let us host it as a managed add-on
• Office hours with me, every month, for the judgment calls no tool answers
• The Operators room, staffed by paid GTM leaders instead of strangers, with a new production system shipped every month

So the honest answer to "I can't build this myself" is: you don't.

The building is already done.

Your job is to run it and direct it, which is the General Mode work only you can do anyway.

Tomorrow I lay out the piece that removes even that: if you do the work and it doesn't land, you don't pay.

Today, write down the real reason you'd hesitate.

Not the price - the actual fear.

Almost always it's "I can't pull this off."

You won't be doing this alone either.

You'll have the standard, and you'll have the room.

That's what lets you run like a team.

- Mitchell

PS: Even Alexander had Aristotle as a tutor before he had an empire. You don't build a real standard alone in a room. The room is how a standard gets passed from one operator to the next, not a crutch.

50 seats, locked for life

For two weeks you learned the whole machine.

• The reframe
• The context file
• The brain
• Targeting
• Enrichment
• Copy
• Deploy
• Audit
• The loop that makes it compound

You have seen the machine work end to end.

You have seen that it is now a single command called /legion.

Today there is nothing left to teach.

There is only a decision.

The offer, plainly: /legion Individual is $197 a month. That is the full machine.

• The brain
• The six-command CLI
• A new production system every month
• The room
• Office hours with me

Everything this sequence taught you to build, already built, yours to run and yours to keep.

The founding terms: the first 50 seats lock $197 a month for life.

Once those 50 are gone, the price moves to $247 a month standard, and it stays there.

The 50 who move first hold the lower rate permanently - that is $600 a year, every year, for as long as you run it.

This is a one-time founding cohort.

There is no fake counter and no invented deadline.

When the 50 seats are claimed, the founding rate is closed.

And the risk is mine, not yours.

It is called the Operator's Guarantee.

• Complete onboarding
• Ship at least one system
• Show up to office hours

Do that, and if /legion does not deliver, you get every dollar back within 30 days.

The only thing I ask is that you run it.

Rent your go-to-market forever, or own the machine that runs it, at the lowest price it will ever be, with the risk carried by me.

These 50 founding seats are the same move: act while the terms are still open.

Claim a founding seat at $197/mo: [link]

PS: This is the only email in this sequence with a price and a link in it, on purpose. The other thirteen were the free course. Napoleon's whole edge was moving first and deciding faster than the other side could finish deliberating. Fifty seats, at the lowest this will ever cost. Move first.