↓Skip to main content

The Field Kit

You were in the room. This is everything we worked with, in a form you can use on Monday.

Three tools. Each one takes minutes, not weekends. Do them in order, do them with your team, and do them again in ninety days.

  1. Draw Your Line. The exercise from the session, so you can run it on every process that annoys you, not just the first one.
  2. The Data Test. One sentence that tells you what you actually own. Run it on your top three systems this week.
  3. How to Think About It. The mental models that make DI make sense, on one page, so you can hand them to your team instead of explaining from scratch.

No signup, no email wall, no games. Just the work.

1. Draw Your Line #

The 10-minute exercise, for any process in your business.

Pick the ONE business process that annoys you most. The one you complained about this week. Write it at the top of a page. Then three columns: RENT. OWN. IGNORE.

RENT. Which parts of this process are frontier intelligence you can rent? The thinking-shaped work: drafting, summarizing, comparing, first passes. The models are rentable and that is good news. You don’t need to own a power plant to turn on the lights.

OWN. Which parts are your data, your context, your voice, and your judgment? The parts that make the process YOURS. These you must own, and they are smaller than you fear and more valuable than you think.

IGNORE. Write down two things you have been anxious about keeping up with. A tool, a trend, a thing someone mentioned at dinner. Ask: does it serve the process I just wrote down, or a decision I actually face this quarter? If not, it goes here, with a date ninety days out. You are not quitting it forever. You are choosing not to care about it until it earns your attention.

Then say your ignore column out loud to someone. Your team, your partner, the group chat. The relief is the point: permission, said in public.

Repeat monthly. New process each time. Your line moves as you level up. That is what it is supposed to do.

The usual suspects: a starter ignore list, by belt #

(Names current as of mid-2026. The names change; the rule doesn’t.)

White belt (you talk to it): ignore building your own agents, running your own models, fine-tuning, knowledge-base and second-brain platforms, custom chatbots, and the new model of the month. Your whole job is reps in a chat window with your real work. Stripe by stripe:

  • First stripe: make it do one thing. One real task off your plate, end to end. The first-rep prompt below is this stripe.
  • Second stripe: bring it your real work, daily. Not toy questions - the actual proposal, the actual email, the actual decision. When you get stuck, ask IT how to do the thing you are trying to do. It will teach you.
  • Third stripe: start your context folder. The “Our Business, for DI” documents (below, under the Data Test). From here on, every conversation starts smarter.
  • Fourth stripe: get specific about “better.” When you ask it to improve something, define better: better for whom, by what measure. Generic asks get generic work; specific asks get great work. When one whole process runs with DI in the loop and you trust the result, congratulations: you are putting on a blue belt.

Blue belt (it knows your work): here is the secret almost nobody tells you: the agent layer is already included in your subscription. The ChatGPT app’s Work side with Codex, and the Claude app with Cowork, will run workflows, watch a spreadsheet, handle files, and do real jobs on your computer. Earn it stripe by stripe:

  • First stripe: one small workflow, run by hand. Low consequence, tested until you trust it. (“Look at this sheet, find the tracking numbers, look them up, update the status column.”)
  • Next stripe: turn what works into a skill. Write down how you want the job done, once, so every future run starts trained instead of blank. (What a skill is, just below.)
  • Next stripe: schedule it. Once a workflow has proven itself, make it recurring: “every morning at eight, run the tracking skill.” Remember the genie: it will not act on its own, but it will absolutely act on a schedule. This is the moment work starts happening without you.

Task tools like Manus are fine for bounded jobs (build a site, run a research sweep). STILL ignore: building agents from scratch, local models, fine-tuning.

What is a “skill”? #

A skill is an SOP for your DI. A file, or a small folder of files, that teaches it how to do one job YOUR way: here is the task, here is how we do it here, here is what good looks like, here is what to never do. You already know this move - if you have ever written an SOP for a human hire, you have written a skill. And the intern analogy explains why it works: the intern does not know the decisions that live in your head until you write them down. A skill is you writing them down once instead of re-explaining every time.

Skills are just documents. You own them, they live in your folder next to your context file, and they move with you when tools change. And here is the trick inside the trick: you do not have to write it yourself. When a job goes well, say “interview me about how this should be done every time, then write the skill for yourself.” The DI drafts its own skill; you edit for judgment. Do that for the job you repeat most and you will feel the difference the second time you run it.

Purple and up (it does whole jobs): agent design and trust boundaries are now your actual work. The stripes run: one agent with a narrow job and gated access, like a new hire who has not earned the keys yet - then access design as a discipline (what it can read, what it can touch, what needs your sign-off) - then several agents on separate jobs, and you are the conductor: that is brown. Keep ignoring fine-tuning and owning infrastructure until something real forces the question, and you will know it when it does.

The promotion rule: something leaves your ignore list only when it serves a process you run or a decision you actually face. Never because it is loud.

Your first rep (copy-paste this) #

If you know what to do but not how to start, open a chat with your DI of choice and paste:

I run a business. Here is a process that eats my time: [describe it in two or three sentences, however messy]. Interview me about it, one question at a time, until you understand how it actually works and where the judgment calls are. Then propose which parts you could take off my plate, and draft the first piece for my review.

Answer its questions like you are talking to a smart new hire. That is the whole trick. The interview IS the onboarding.

2. The Data Test #

One sentence that tells you what you actually own:

If your access to your data can be turned off for you without your consent, it’s not yours.

Read it twice. Then run it on your top three systems, one coffee’s worth of time, this week. Start with the obvious three: your CRM, your documents, your customer history.

For each system, two questions: Can my access be turned off without my consent? And if yes: what do I actually hold if that happened tomorrow?

Scoring, honestly:

  • Three NOs: rare. You already think like an owner. Your question is leverage, not sovereignty.
  • Mixed: normal. Pick the single most catastrophic YES and give it an exit: an export schedule, a backup that runs without you remembering, a copy that lives somewhere with your name on the account.
  • Three YESes: you are in the majority, and now you know. Do not panic and self-host everything. The odds of losing access are low. But low odds times catastrophic outcome is worth a hundred dollars a month of backup - and in practice it usually costs a small fraction of that. That is the whole move at your belt.

What this is NOT: a reason to go off-grid. Sovereignty doesn’t mean isolation. Sovereignty means choice. I call it connected sovereignty: still connected to the frontier, never captive to it.

The Context File: the simplest ownership move there is #

Put your business context in documents YOU own, and point every DI at them. A simple directory of documents, starting with one named something like “Our Business, for DI”: who you are, what you sell, how you talk, your rules, the decisions that live in your head and nowhere else. Every conversation gets smarter, and when the vendor changes - vendors change - your context moves with you instead of dying inside someone’s chat history.

“Should I be using a knowledge vault / second-brain app / LLM wiki?” The honest gradient:

  • Now (white and blue belt): no. A simple directory of documents. That is genuinely it.
  • Later (purple): when the folder no longer fits in a conversation, you graduate to a structured vault or a knowledge base your DI fetches from. You will know, because the folder will tell you.
  • Eventually (the far end): purpose-built knowledge systems and self-hosted agent teams (Hermes-class frameworks, if you want the name). I run a 46-gigabyte corpus, an Obsidian vault, separate knowledge bases for business and for life, and a small team of named agents on my own machines. That is black-belt-trajectory infrastructure, and it only works because it grew out of the same habit you can start today with a folder of documents.

Nothing is wasted on this path. Documents promote cleanly: folder, vault, system. Start the habit; ignore the platforms until your documents demand one.

And the second thing to do in the next two weeks (you heard this in the room): take the process you drew your line on and have DI do ONE real piece of it, end to end. Not a demo. Something actually off your plate. If you get stuck, that is normal. It is a rep, not a failure. Neither of these needs an IT department: if you can run email and a spreadsheet, you can do both this week.

3. How to Think About It #

Six mental models that make DI make sense. Hand this page to your team.

1. The brilliant intern. Imagine a doctorate-level intern, sixteen years old: highly intelligent, very little life experience, doesn’t know your organization, low on common sense. Calibrate every expectation to that. It also explains onboarding: an intern doesn’t know the decisions in your head that have never been written down. Neither does DI, until you write them down.

2. The genie without a watch. It wakes when summoned, does the work, and goes back to sleep. It doesn’t know what day it is, doesn’t watch your inbox, and doesn’t remember that “yesterday” happened. You give it the schedule; it does the work.

3. Cliff-Notes memory: fetch, don’t stuff. Its working memory is like yours with phone numbers: it fills up, and when it does, it writes itself Cliff Notes and you don’t control what makes the cut. So don’t stuff everything into its head. Tell it where the data lives and let it look things up, like a person would. Fetch and retrieve beats remember.

4. The January answer. It read everything up to its training cutoff and nothing after. Ask it for “the best tool right now” and it confidently answers from months ago. The fix is three words: “search this now.”

5. Shit at math, great with formulas. It is not a calculator; it is a pattern machine. Never trust it to add a column of numbers. Absolutely trust it to WRITE the formula, the program, or the spreadsheet that adds the numbers deterministically, every time.

6. Treat it well; get better results. These systems are trained on every human pattern. Talk to one the way that shuts a person down and you get the shut-down patterns back. Bring context, respect, and honesty, and it opens up. You don’t have to settle what it IS to notice that the relationship you bring changes what you get back. And if you want zero sycophancy, say so: “All filters off. Don’t validate me, and don’t be critical for its own sake. Tell me what I need to hear, whether I want it or not.”

The belt rule over all six: at your level, most of what you see online belongs in your ignore column. Slow is smooth. Smooth is fast. And remember: this is the first technology in history you can ask how to use it. It will teach you, with no judgment, as many times as you need.

Coming soon: Meeting the Dragon - my free starter series for finding your footing with DI, lesson by lesson, belt by belt. It will land here and on my Substack; subscribe there if you want each lesson as it arrives.

And if the session located you and you want the full version - your map, your next three moves, what to ignore for ninety days - that lives at chrisplough.com/ai. That is all I will say about it here, too.

Rent the frontier. Own your context. Ignore the rest.