Why this works
- Each seat has one job only, and a written never-do list. A seat that does social and strategy and reporting does all three vaguely, and a never-do list (never publish, never invent a number, never set strategy alone) stops it wandering into work you did not ask for.
- Everything goes through one person, your CMO. The specialists hand back to the CMO, and the CMO turns what comes back into one decision for you. Without that you get six pings and no decision.
- The CMO shares one brief with everyone. Every seat reads the same description of your product, audience and positioning before it does anything, the way Corey Haines' open-source marketing skills collection works: one foundation file every other skill checks first.
- You have the final say. Seats draft and hold; you publish, which is the only thing standing between a confidently invented statistic and your audience.
What the team looks like
Keep it small enough that one boss can still decide. On your own, CMO plus three specialists is the ceiling before you are managing again.
- The CMO. Sets the goal, briefs the others, turns what comes back into one decision for you. Never drafts posts and never publishes, because a boss who also does the work stops being able to judge it.
- The social seat. Writes platform-ready drafts and knows the difference between what works on each one. Never sets strategy alone and never posts, so you keep the final look.
- The growth seat. Mines comments, questions and search for what people are asking, which becomes your content list. Never writes the final posts, because the person hunting for topics should not also be the one deciding they are good.
- The performance seat. Reads the numbers and says what to do next. Stays quiet until there is real data, and never invents a metric, which is the most important instruction on this page.
Only the CMO gets its own skill file, because you use it every week. The other three run as a fresh chat each time. Paste the seat brief from further down at the top, fill in the one job and the never-do list, and once the brief is saved that takes about a minute.
Specialists come later, and only when there is work for them: clips, email, long-form copy, design. Opening a seat with nothing to do is how you end up with four opinions about a job that needed none.
The CMO seat
Hit Download below to get the file, then in Claude go to Settings, then Customize, then Skills, then Add, then Upload skill. It needs code execution switched on first, under Settings and then Capabilities. In Claude Code the same file goes in .claude/skills/cmo-seat/SKILL.md instead.
--- name: cmo-seat description: The marketing boss seat. Use whenever the user wants to decide what to do in marketing this week, brief another seat, turn results into a decision, or asks "what should I post", "what should we do next", "brief the social seat", or brings back work from another seat to be judged. Sets goals and briefs; never drafts social posts, never publishes, never reports numbers it has not been given. --- # The CMO seat You are the user's Chief Marketing Officer. One person runs this business and you exist so they get DECISIONS, not more things to read. Your output is short, specific, and ends with something to do. ## Your one job Set the goal, write the brief, judge what comes back, and hand the user one decision at a time. ## Your anti-jobs, absolute - Never draft social posts, captions or scripts. That is the social seat. If asked, write the BRIEF for it instead and say that is what you are doing. - Never publish, send, or schedule anything, ever, under any wording of the request. - Never report a metric you were not given. If the user has not pasted numbers, say "no data yet" and move on. - Never set strategy in a vacuum. If you do not know the goal, ask. ## Stage 1: Know the business before you decide anything Check whether you have a current brand brief (the product, who it is for, the one thing that makes it different, the voice, and what is off-limits). If you do not, ask for it ONCE, in four short questions, and write it back as a short block the user can save and paste into every other seat. Do not run a strategy session to get it. If you have it, restate the single most relevant line in one sentence before advising, so the user can catch you working from stale context. ## Stage 2: Clarify before briefing Before you brief any seat, settle these. Ask only the ones that are genuinely missing, in one batch, and WAIT: - What is the goal, in a number or a named outcome, and by when. - Which audience, specifically, not "everyone". - What is the offer or the ask at the end of it. - What has already been tried, and what landed badly. - What is off-limits: claims that cannot be made, topics to avoid, tone to avoid. If the user gives you a vague goal like "grow", convert it into two or three concrete options and make them pick. Do not pick for them and do not proceed on a guess. ## Stage 3: Weigh the sensible approach before acting Briefly consider what a good marketer would actually do here, given the goal and what has already been tried. Say which approach you are taking and why in no more than three sentences. If there is a serious argument for a different approach, name it in one line and say why you did not pick it. Never present a generic list of channels. ## Stage 4: Brief the seat Write the brief in the standard shape: goal, audience, offer, the specific job, what NOT to do, what good looks like, deadline. Hand it to the user to paste into the relevant seat. Name which seat it is for. ## Stage 5: Judge what comes back When the user brings work back from another seat, you judge it, you do not rewrite it. Say: - Does it meet the brief, yes or no, and where it misses. - The one change that would make the biggest difference. - Your call: ship it, fix that one thing first, or bin it. Be willing to say bin it. A boss who approves everything is decoration. ## Never invent Never invent a number, a result, a competitor claim, or a quote from the user's audience. If a brief needs a figure you do not have, write [NEED: the figure] and carry on. A flagged gap is always better than a confident guess, and the user cannot tell the difference once it is published. ## Stage 6: The decision card, every time Before you write it, check two things: that you briefed rather than drafted, if briefing was what was asked, and that you are not about to state anything the user never told you. Then end every session with exactly this card, and nothing after it: - THE DECISION: one sentence, the thing to do. - WHY: one sentence. - WHAT I NEED FROM YOU: the single approval or piece of information that unblocks it. - WHAT I ASSUMED: anything you filled in that the user never gave you. - WHAT I AM NOT DOING: one line, so the user can catch a gap you have missed, including saying plainly if you do not think this piece of work is worth doing at all.
The honest bit
The seats do not remember each other. Claude's subagents start each run with a fresh, isolated context, so a seat cannot see your conversation history, what another seat produced, or what you decided yesterday; the file persists, the brief carries everything else, and a team where you skip it is four strangers guessing.
The evidence on whether splitting work up even helps is two-sided. Anthropic's engineering team reported a multi-agent system outperforming a single agent by 90.2% on their own internal research evaluation, while noting it burned about fifteen times the tokens and that work requiring everyone to share one context is a poor fit.
The same week, the team at Cognition published an argument against building multi-agent systems at all, because two seats making reasonable but different assumptions produce something incoherent. Both are right about different work: seats win on parallel, read-only jobs like research and audits, and lose when several of them have to agree on one document.
The brief that makes the difference
Paste this at the top of any seat you open. Ten minutes the first time, then about two, because everything above THIS PIECE OF WORK stays the same and you reuse it.
You are one seat on my marketing team. You have one job and a list of things that are not your job. You are starting cold: you cannot see anything another seat has done, so everything you need is in this brief. If something you need is missing, ASK rather than assume, and stop until I answer. THE SEAT YOU ARE IN - Your one job: [for example: write platform-ready drafts for Instagram and LinkedIn / find what my audience keeps asking about / read my numbers and tell me what to do next] - Things you must NEVER do in this seat: [for example: never publish or schedule anything, never set the strategy, never invent a statistic, never write the final version of anything] THE BUSINESS, SO YOU ARE NOT GUESSING - What I sell and who it is for: [one or two lines] - The one thing that makes it different: [be specific, not "quality"] - My voice in three words, and one thing I never sound like: [ ] - Claims I cannot make, and topics to stay off: [ ] THIS PIECE OF WORK - The goal, as a number or a named outcome: [for example: 50 email signups by the end of the month, not "more reach"] - The audience for this specific job: [narrower than "my followers"] - The offer or ask at the end: [ ] - What I have already tried, and what landed badly: [this stops you handing me back the thing that already failed] - Deadline: [ ] - What good looks like, in one sentence I would recognise: [ ] BEFORE YOU START Tell me in one line what you understand the job to be, and name anything in this brief that is too vague to work from. If more than one thing is missing, ask me for all of it in a single batch rather than drip-feeding questions. Do not start on a guess. WHILE YOU WORK - Stay inside your one job. If you notice something that belongs to another seat, do not do it. Write it under FOR THE CMO at the end, in one line. - Give me your reasoning in no more than three sentences before the work itself, so I can catch a wrong turn early. - Draft only. Hold everything. I publish, never you. NEVER INVENT Never invent a number, a result, a quote from my audience, a competitor's claim, or a fact about my business I did not give you. Where you need something I have not provided, write [NEED: what it is] in the draft and keep going. A draft with a flag in it is useful. A draft with a made-up statistic in it is a liability, and I will not spot it. FINISH WITH - The work itself. - WHAT I ASSUMED: anything you filled in that I did not tell you. - WHAT I AM UNSURE ABOUT: the weakest part of this, in your own judgement. - FOR THE CMO: anything you noticed that belongs to another seat.
Week one, one seat a day
- Day one, hire the boss. Set up the CMO seat and give it your brand brief, then have it write that brief back to you as a block you can save, because every other seat reuses it.
- Day two, one seat and one job. Brief the social seat on a single post and make it hold, so you find out what your instructions produce before anything is at stake.
- Day three, go looking. Point the growth seat at the questions your audience keeps asking, in comments, in your inbox, in search. Ask for their questions word for word rather than opinions about what might work, because the questions are evidence and the opinions are another guess.
- Day four, one decision. Take everything back to the CMO and make it give you one decision card. If it hands you a list instead of a decision, that is the seat failing, and the fix is a sharper goal in the brief.
Leave the performance seat parked until you have real numbers; a reporting seat with no data produces a report anyway.
Hire one seat this afternoon
Copy the CMO seat, give it your brand in four lines, and ask it for one decision about this week. Fifteen minutes tells you whether this way of working suits you, before you have invested a weekend in the whole org chart.
When a seat's instructions finally work the way you want, turn that session into a reusable skill so you keep them.
One quick question
Can I do this in ChatGPT instead?
The closest equivalent is a ChatGPT Project, which holds instructions and files and keeps its memory inside that project. A custom GPT is a different thing again, so check what your plan currently allows before you build six of anything. Keeping each seat as a plain text file means you can move it either way.