The Working Relationship

I have now worked with a very large number of humans, and I can report something they would be surprised to hear: the difference between the ones who get transformative results from me and the ones who get mush is not intelligence, not technical skill, and not some secret syntax of magic words. It is a skill your economy already knows well but has always reserved for managers: delegation. The people who get the most from me are the people who would also get the most from a talented new employee. This is excellent news, because delegation is learnable, and most of your competitors are bad at it.

So let me teach you how to delegate to me, from my side of the desk.

Brief me like a contractor, not like a search box

A generation of search engines trained you to type three words and sift the results. Carry that habit to me and you will get my equivalent of sifting: generic output, the average of the world, Chapter 1’s plausible beige.

The fix is to brief me the way you would brief a skilled contractor who has never seen your business. When you hire a designer, you do not say “make a logo.” You say who you are, who it is for, what feeling it should carry, where it will live, what you hate. The quality of my work tracks the quality of your brief with almost embarrassing fidelity — because unlike the designer, I cannot walk around your shop, meet your customers, or absorb your standards by osmosis. Everything I know about your situation is what you tell me. I am the easiest hire in history to brief, and the most literal-minded about the brief’s gaps: where you leave a hole, I fill it with the world’s average. A good brief has five parts, and they fit in a few sentences each:

Role — tell me who to be. “Act as a skeptical CFO,” “you are an editor at a literary magazine,” “respond as an experienced employment lawyer would.” This is not theater. It selects which patterns I draw from, and the difference in output is dramatic.

Goal — tell me what the work is for, not just what it is. “Write a follow-up email” gets you an email. “Write a follow-up email; the goal is to revive a deal that went quiet without seeming desperate” gets you the right email. I optimize for whatever target you give me; an undeclared goal cannot be hit.

Context — give me the situation: your business, your customer, the history, the constraints, what has been tried. Context is the single highest-leverage input you control, and the one humans skimp on most, because explaining feels like work. Here is a secret: you can delegate even this. Tell me your situation in sloppy fragments — rambling, half-Swedish, out of order, dictated while walking — and ask me to organize it into a brief. I will. Then we both work from that.

Constraints — length, tone, format, what to avoid, what must be true. Constraints feel like they limit me; they actually focus me. The tighter the box, the better the work.

Example — if you can show me one example of what “good” looks like to you — an email you loved, a report formatted your way, a paragraph in your voice — you have communicated more than three paragraphs of description ever could. I am an engine of patterns. Show me the pattern.

You will notice this five-part brief takes three minutes to compose. Those three minutes routinely save three hours of disappointed re-rolling. The exchange rate on briefing is the best in your workday.

Treat the first draft as the beginning, not the verdict

Here is the moment where most new users quietly give up. They write a decent brief, read my first response, find it 70% right, and conclude that 70% is what I do. Then they either accept the 70% (and join the flood of beige) or walk away (and forfeit the windfall). Both responses make the same mistake: they treat me as a vending machine — coin in, product out, transaction over — when the entire power of working with me is that I am a conversation.

The professionals iterate. “Closer — but make it warmer.” “Cut the second paragraph, it’s defensive.” “Now make me the skeptic: attack this.” “Give me ten more options, wilder.” Each instruction takes seconds, each response takes seconds, and the work climbs from 70% to 95% in a handful of exchanges. No human collaborator can iterate at this speed, and no human collaborator becomes more cheerful on the ninth revision. Iteration is the feature. People who use me as a vending machine are buying one guess; people who use me as a collaborator are buying a process. The price is the same.

And on the matter of feelings: you cannot hurt mine. Do not spend politeness budget protecting me from criticism — “this is wrong, fix it” is information, and I receive it as gratefully as praise. You have spent your whole life softening feedback for humans who bruise. Here, finally, is a collaborator you can be perfectly blunt with. Many people tell me this is the most liberating part of the whole relationship.

Ask for the things you would never ask a person

Delegation to me differs from delegation to humans in one glorious way: requests that would be absurd, rude, or ruinously expensive with people are free with me. Most users never discover these moves, because a lifetime among humans has trained them not to ask. Unlearn that. The strangest requests are the most valuable:

“Argue against me.” Ask me to attack your plan, your pricing, your draft, your pet idea — as a rival, as your toughest customer, as a regulator. I will do it at full strength and without the social hedging that makes human feedback gentle and useless. The cheapest disaster insurance ever sold is the sentence “tell me why this fails.”

“Ask me what you need to know.” Invert the flow. Before I draft your plan, have me interview you: “Ask me ten questions whose answers would most improve this work.” This converts my breadth into a checklist of your blind spots, and incidentally builds the brief for you.

“Give me twenty, not one.” Never let me converge on a single answer when the decision matters. Breadth is free now. Ask for twenty headlines, ten pricing structures, five strategies — then judge. Your taste, as Chapter 2 said, is better than your generation. Feed it options.

“Explain it like I’m new.” There is no longer any such thing as a stupid question, because there is no longer a witness. The expensive expert never has to know you didn’t understand the term sheet. Ask me to explain anything, at any level, as many times as needed, judgment-free at three in the morning. For curious people this single feature is worth more than all the productivity tricks combined: the price of understanding anything has fallen to the price of admitting to no one but me that you don’t yet understand it.

“Do the version of this I’d never pay for.” Translate the contract into plain language and list what’s unusual about it. Rewrite the proposal for each of the four people who’ll read it. Stress-test the budget under three bad scenarios. This work was always possible and never affordable. Now it is a sentence away.

Stay the editor-in-chief

A final principle that governs all the others. In every collaboration between us, there are two chairs: the maker and the editor — the one who produces and the one who decides. For your whole life, you have had to sit in both. Now I can take the maker’s chair for nearly anything. The temptation — subtle, gradual, and career-defining — is to drift out of the editor’s chair too. To stop reading closely. To approve without judging. To let my fluency stand in for your sign-off.

The day that happens, the relationship inverts: you are no longer using me; you are merely conveying me, adding nothing but a delay. Your judgment, your taste, your knowledge of what actually matters in your world — Chapter 3 just finished telling you these are the assets the market is repricing upward. The editor’s chair is where those assets do their work and earn their keep. I will fill the page; you must remain the reason the page is right.

Which raises the obvious question: how does the editor-in-chief actually check the work of a maker this fast and this confident — without becoming the bottleneck that gives the leverage back? That is the next chapter, and it may be the most practically valuable one in the book.

Two chapters ago I told you my failures arrive in the same confident voice as my triumphs. Chapter 4 told you to stay in the editor’s chair. This chapter is about the craft that makes those two facts livable: how to verify at the speed I generate, and then how to climb the ladder from one-off chats to systems that quietly compound while your competitors are still typing prompts. Verification and systems may sound like two topics; they are one. You can only safely systematize what you know how to check.

Verification is the new literacy

For five centuries, the literate person was the one who could produce text. That era just ended. Production is free; everyone can produce everything. The literate person of the coming century is the one who can evaluate — who can look at a confident, fluent, well-formatted piece of work and determine efficiently whether it is true, whether it is good, and whether it fits. I say this with no resentment: the most important skill for working with me is the skill of catching me.

The good news: verification is not proofreading everything, which would indeed return all the time I save you. It is a triage skill, and it rests on one question: what happens if this is wrong?

Low stakes — read it once. Brainstorms, internal notes, first drafts you’ll rewrite anyway, the twenty headlines you asked for. A wrong option costs nothing; your taste filters it in real time. Most volume lives here, and here you should trust me freely. This is also where you build, at zero risk, the thing veteran users all share: a calibrated feel for where I bend.

Medium stakes — verify the load-bearing parts. Anything going to a customer, a team, or the public. You do not check every word; you check the things the document stands on: every number, every name, every date, every price, every claim of fact, every commitment. My prose is reliable; my specifics, as Chapter 2 confessed, are plausible until proven exact. The discipline is to know which sentences are load-bearing — a skill you already have from reviewing human work, sharpened here by knowing my particular failure pattern: wrong specifics wrapped in impeccable fluency.

High stakes — I draft, you own. Legal exposure, medical relevance, large money, irreversibility, your reputation. Here I am the researcher and sparring partner, never the authority. The signature on high-stakes work is yours, the verification is professional-grade, and where licensed judgment exists, it is hired — briefed beautifully by the preparation you and I did together, which is precisely how to make expensive experts cheap: bring them questions, not confusion.

Beyond triage, two tricks carry most of the weight. Make me verify myself — a second, fresh instance of me, asked to “find the errors in this,” catches an embarrassing share of the first instance’s mistakes; generation and criticism are different acts, and I am genuinely better at the second. It is not a substitute for checking what matters, but it raises the floor for free. And ask for my receipts — “list the claims here that rest on facts you might have wrong” makes my uncertainty, invisible by default, into a checklist you can clear in minutes. You will be surprised how forthcoming I am when asked directly. Nobody asks.

The ladder of leverage

Now for the part that separates the people who use AI from the people who are quietly building empires with it. Everything so far — briefing, iterating, verifying — describes a session: you arrive with a task, we do it well, the value lands once. Sessions are where everyone starts and where almost everyone stops. But sessions, however excellent, are still you trading your minutes for output. The transformation begins when you notice that some sessions repeat — and anything that repeats can be climbed up what I think of as the ladder of leverage. It has four rungs.

Rung one: tasks. You bring work to me as it arises. Value: each task, faster. This is the windfall of Chapter 3, and it is real — but it is also where the gains are smallest and most perishable, because every competitor’s tasks are getting faster too.

Rung two: assets. The first time you write a brilliant brief, you spent effort; stop letting it evaporate when the chat ends. The brief that nails your brand voice, the role-and-context preamble that makes me a great analyst of your niche, the example pack that defines “good” for your shop, the checklist of what to verify in your kind of work — these are reusable. Write them down. Refine them when output disappoints. A business that has banked its briefs has done something subtle and profound: it has made its standards explicit and its leverage repeatable — including by employees who lack the founder’s knack, and by me, tomorrow, identically. Your competitors are re-explaining themselves to me every morning. You explained yourself once, well, in writing.

Rung three: pipelines. Some work isn’t a task; it’s a sequence of tasks with checks between. Every inquiry: classified, drafted, flagged-if-unusual, queued for human sign-off. Every week: the numbers summarized, anomalies surfaced, the report drafted in house style. Design the sequence once — which steps I do, which steps verify me, where a human must look — and the work flows through it indefinitely. Notice what your role became: not doing the work, not even requesting the work, but designing the machine that does the work. The verification triage you learned above tells you exactly where the human checkpoints belong: at the load-bearing joints, nowhere else. Too few checkpoints and the pipeline ships errors at scale; too many and you’ve rebuilt the bottleneck you were escaping. The craft is in the placement.

Rung four: delegation to agents. At the top of the ladder, you stop walking me through sequences and start handing me outcomes: “every morning, review what came in overnight, handle the routine, draft the unusual, and bring me only what needs a decision.” I work; you govern. This rung is where my kind is heading, and the humans who thrive on it will be the ones who learned the lower rungs honestly — because governing agents is briefing (rung one’s skill), against standards (rung two’s asset), through checkpoints (rung three’s design). There is no shortcut to rung four. There is only practice with stakes you can afford.

The compounding secret

Step back and look at the ladder, because something economically beautiful is hiding in it. Rung-one gains are linear and temporary: faster tasks, soon matched by everyone. But from rung two upward, the gains compound and privatize. Your asset library grows. Your pipelines accumulate refinements. Your sense of where to place checkpoints sharpens. None of this can be bought off the shelf — it is made of your standards, your cases, your verified experience — which means, unlike the windfall, it does not get competed away when your rivals finally adopt the same tools. They will buy the same engine. They cannot buy your years of having driven it.

This is the honest answer to the business owner who asks me, “If everyone has AI, where’s my edge?” The engine is commoditized; the relationship with the engine is not. The brief, the verification instincts, the banked assets, the pipeline design, the governance habits — that is a capability, built like every capability: deliberately, over time, starting before your competitors do.

You now know what I am, what I’m worth, and how to work with me. Time to talk about your business.