{"id":"2093330541177352217","url":"https://x.com/VibeMarketer_/status/2093330541177352217","text":"https://x.com/i/article/2093330534495916032","author":{"name":"J.B.","username":"VibeMarketer_","avatarUrl":"https://pbs.twimg.com/profile_images/1950293861793677312/ldQniJt5_200x200.jpg"},"createdAt":"Fri Aug 28 13:31:07 +0000 2026","engagement":{"replies":158,"retweets":422,"likes":2884,"views":283878},"article":{"title":"How to Build a One-Person Media Company With Hermes Bots","previewText":"i built a team of six hermes bots that gives one person the research, writing, editing, and distribution capacity of a complete media company.\nmost people use ai to write faster. that is useful, but","coverImageUrl":"https://pbs.twimg.com/media/HQ0BAvOawAAgoRo.jpg","content":"i built a team of six hermes bots that gives one person the research, writing, editing, and distribution capacity of a complete media company.\n\nmost people use ai to write faster. that is useful, but writing is no longer the bottleneck.\n\nthe difficult part is consistently finding ideas worth covering, developing an original angle, and distributing each one without publishing the same recycled post across five platforms.\n\ni trained the system on the strongest research, packaging, evidence, and distribution patterns from my own content process, then stored them in a shared obsidian graph. \n\nthis is the exaxt system that has helped me generate over 5.8 million impressions on X account in just the past 4 weeks. \n\n![](https://pbs.twimg.com/media/HQ0BA83aoAATcge.jpg)\n\none bot searches for trends, customer questions, authority clips, and promising source material. another verifies the research. a strategist finds the strongest angle. a writer develops the long-form piece. a distribution bot rethinks it for x, linkedin, newsletters, carousels, and video. an editor checks the entire package before it reaches me.\n\nevery bot is trained on the same obsidian content brain: my voice, audience, offers, proof, hooks, platform playbooks, and lessons from previous posts.\n\ngive the team one strong idea and it returns the research, flagship piece, and distribution campaign required to own that idea across every platform you use.\n\nthis guide shows you how to build the complete system.\n\n## a media company is a loop\n\nthe valuable part of a media company is not the number of drafts it produces. it is the loop connecting attention, research, editorial judgment, distribution, and feedback.\n\nthe complete loop looks like this:\n\nidea → research → angle → long-form → distribution → review → performance → updated playbooks\n\nbreak any connection and quality drops.\n\nif research never reaches the strategist, the angle becomes generic. if the writer never sees the source packet, claims drift. if distribution starts from a finished article without understanding its argument, every platform receives a shortened version of the same thing. if performance never updates the playbooks, the team repeats the same mistakes forever.\n\nthis is why six independent chatbots are not a media company. they need clear ownership, shared context, structured handoffs, and one feedback loop.\n\nhermes agent's august 16 release gives us the pieces required to build it. bot mode adds a visible roster of named bots and communication between them. profiles give each specialist separate memory, sessions, skills, and instructions. kanban gives the work a durable path across those profiles, including dependencies, comments, review, retries, and human input.\n\nuse each layer for the job it is good at:\n\n- bot mode makes the team visible and easy to talk to.\n\n- profiles keep each role focused and prevent one bot's memory from becoming everyone else's memory.\n\n- obsidian holds the shared editorial knowledge all six bots are allowed to use.\n\n- kanban moves real assignments across the team without relying on one long conversation.\n\neach layer owns a different part of the operation. the bots make editorial decisions, the obsidian vault preserves shared knowledge, and kanban keeps the work moving between them.\n\n![](https://pbs.twimg.com/media/HQ0BBI7bUAAxGML.jpg)\n\n## build the shared content brain first\n\nstart by building the knowledge all six roles need to make compatible decisions. once that foundation exists, each bot can work from the same editorial standards.\n\nobsidian gives that knowledge a useful shape. each markdown file holds one part of the system, while links show how voice, audience, hooks, platforms, and performance affect one another.\n\nthat graph becomes the team's shared content brain.\n\ncreate this structure inside your obsidian vault:\n\nmedia-company/\n\n- index\n\n- brand: voice, audience, offers, proof\n\n- discovery: signals, customer questions, authority clips\n\n- engine: angles, hooks, repurposing, review, performance\n\n- platforms: x, linkedin, newsletter, video, carousel\n\n- campaigns: one folder for every active campaign\n\nobsidian makes the relationships visible, but the files remain plain markdown. hermes can read and update them without requiring a special database or proprietary content tool.\n\nthe index file is the entry point every bot reads first. it defines:\n\nwhat we publish about\n\n- applied ai systems for operators\n\n- practical agent builds\n\n- new tools with a specific business use\n\nwho we publish for\n\n- founders\n\n- marketers\n\n- ai operators\n\nwhat every campaign must contain\n\n- one clear reader outcome\n\n- one central claim\n\n- direct evidence for consequential claims\n\n- one reusable framework, workflow, or decision rule\n\n- platform-native distribution\n\nrouting\n\n- new signals go to signal scout\n\n- approved signals go to researcher\n\n- complete source packets go to content strategist\n\n- approved angle briefs go to long-form writer\n\n- approved flagship pieces go to distribution bot\n\n- every external asset goes through editor\n\nhuman approval\n\n- required before publishing\n\n- required before changing voice, audience, offer, or evidence rules\n\n- required before a performance lesson becomes a permanent playbook rule\n\ndo not turn the vault into an archive of every thought the bots produce. store accepted knowledge, current rules, reusable examples, and links to source material. campaign drafts belong in the campaigns folder, where they can be reviewed or discarded without polluting the team's long-term memory.\n\nthe shared brain should become more selective as it grows.\n\n![](https://pbs.twimg.com/media/HQ0BBS6bsAACUGF.jpg)\n\n## give six bots six different jobs\n\nthe fastest way to ruin a multi-agent workflow is to give every bot the same broad instruction: make great content.\n\neach bot needs one decision to own, one deliverable to return, and a clear point where it must stop.\n\nuse this contract for every profile:\n\n- owns: the decision this bot is responsible for.\n\n- reads: the files and handoff fields it may use.\n\n- returns: the exact artifact the next bot receives.\n\n- must not: decisions that belong to another bot or a human.\n\n- done when: observable conditions that make the handoff complete.\n\nnow create the team.\n\n1. signal scout finds ideas with a reason to exist now\n\nsignal scout watches product launches, research, customer questions, recurring objections, strong authority clips, and conversations already attracting attention.\n\nit does not decide the final thesis or start drafting posts.\n\nfor every candidate, it returns:\n\n- what happened;\n\n- why the audience may care;\n\n- the original source;\n\n- the strongest authority clip or proof object;\n\n- the question the finished piece could answer;\n\n- how quickly the opportunity will decay;\n\n- a short reason to reject it when the signal is weak.\n\nthis bot should discard far more ideas than it approves. its job is to protect the rest of the team from spending hours polishing a topic nobody needed.\n\n2. researcher builds the evidence package\n\nresearcher receives an approved signal and turns it into a source-bound evidence package.\n\nit verifies the original claim, finds primary sources, checks the surrounding context, records useful numbers, and separates verified facts from inference.\n\nits handoff includes:\n\n- the current event or source that creates urgency;\n\n- three to seven verified claims;\n\n- direct urls for every consequential claim;\n\n- relevant quotations or timestamped clips;\n\n- contradictions and missing evidence;\n\n- what the sources do not prove;\n\n- two or three mechanisms worth explaining.\n\nresearcher does not select a sensational headline and then search for supporting evidence. it hands the strategist a bounded set of facts strong enough to support an original argument.\n\n3. content strategist finds the story inside the research\n\ncontent strategist turns the evidence package into one editorial decision.\n\nit chooses:\n\n- the reader;\n\n- the outcome;\n\n- the central tension;\n\n- the thesis;\n\n- the most useful format;\n\n- the flagship headline;\n\n- the reusable object the reader will leave with;\n\n- the distribution angles that could later stand alone.\n\nthe output is an angle brief, not a draft:\n\n- reader:\n\n- reader outcome:\n\n- current source:\n\n- central tension:\n\n- thesis:\n\n- what becomes possible:\n\n- flagship format:\n\n- reusable object:\n\n- proof required:\n\n- sections:\n\n- distribution entryways: proof, mechanism, workflow, risk, and result\n\none complete angle is more useful than ten interchangeable ideas. the strategist should return one recommendation and explain why the rejected directions are weaker.\n\n4. long-form writer creates the flagship piece\n\nthe writer receives the approved angle brief, evidence package, voice file, and the relevant article patterns from the vault.\n\nits job is to create the deepest and most reusable version of the idea. depending on the campaign, that might be an x article, newsletter, guide, or video essay.\n\nthe flagship piece should contain:\n\n- an outcome-led headline;\n\n- a first screen that makes the result tangible;\n\n- visible architecture;\n\n- source-backed claims;\n\n- a complete workflow or framework;\n\n- examples at the moments where a reader could get stuck;\n\n- a compressed ending that makes the idea easy to remember.\n\nthe writer does not create every platform asset. it produces the source material from which the distribution bot can develop several different stories.\n\n5. distribution bot rebuilds the idea for each platform\n\nrepurposing usually fails because the system treats formatting as distribution.\n\nan article does not become an x post because it lost 1,500 words. a newsletter does not become a carousel because its paragraphs were placed on slides.\n\ndistribution bot returns to the angle brief and asks what part of the idea fits each platform's consumption pattern.\n\nfor x, it might isolate the sharpest claim, a surprising proof point, a build sequence, or an authority clip.\n\nfor linkedin, it might develop the operator lesson, the internal decision, or the before-and-after workflow.\n\nfor a carousel, it should choose the framework that becomes clearer when shown visually.\n\nfor video, it should build a spoken narrative around the tension, demonstration, and result.\n\nfor the newsletter, it can add the nuance, examples, and personal context that would overload a short post.\n\nthe requirement is simple: every asset must give someone a reason to consume it even if they already saw another part of the campaign.\n\n6. editor protects the whole operation\n\neditor receives every asset together, not one at a time.\n\nthat lets it catch problems a platform-specific review would miss:\n\n- five hooks making the same claim;\n\n- the same opening story repeated everywhere;\n\n- unsupported facts introduced during repurposing;\n\n- tone drifting between platforms;\n\n- a carousel that adds no value beyond the article;\n\n- a cta that does not match the reader's stage;\n\n- one platform receiving far less useful content than the others.\n\neditor can approve, request a revision, or reject an asset. it cannot publish.\n\nthe human review queue should show the final copy, supporting source, intended platform, media, and the decision required. you should not have to reconstruct how the team reached the output before approving it.\n\n![](https://pbs.twimg.com/media/HQ0BBeoasAAX1zz.jpg)\n\n## make every handoff inspectable\n\na multi-bot team fails when one bot returns prose and the next bot has to guess which parts matter.\n\ngive every campaign one record that travels through the system:\n\n- campaign: hermes-media-company\n\n- status: research\n\n- signal: event, source, urgency, and audience question\n\n- research: verified claims, sources, authority clips, contradictions, and unknowns\n\n- angle: reader, outcome, tension, thesis, and reusable object\n\n- flagship: format, path, and approval state\n\n- distribution: x, linkedin, newsletter, video, and carousel assets\n\n- review: issues and final decision\n\n- performance: observations and proposed rule changes\n\nthe record does two jobs. it gives the next bot a predictable input, and it lets you inspect the history of the campaign without reopening six conversations.\n\nif a required field is missing, the bot should return the task to the previous stage. it should not quietly fill the gap with a plausible assumption.\n\n![](https://pbs.twimg.com/media/HQ0BBo2aYAAZ23Y.jpg)\n\n## build the team in hermes\n\nopen the latest version of hermes desktop and create one isolated profile for each role. clone the same base configuration into all six profiles so they share your model and core capabilities while keeping separate sessions and memory.\n\nput each role contract in that profile's SOUL file. point each profile's working directory at the same media-company vault, but restrict the files each role is expected to change.\n\nthen open bot mode and add the six profiles to the roster. give them recognizable names and keep one persistent room for the media company so you can see questions and interventions without mixing them into the durable campaign record.\n\nuse conversation for coordination. use files and kanban for state.\n\ncreate one kanban board for the media company, start the dispatcher, and assign the first campaign to signal scout. the board becomes the visible production path from discovery through final review.\n\nhermes kanban stores tasks and handoffs in a durable sqlite-backed board. a task can wait for dependencies, move into review, survive restarts, carry comments, and return to the correct profile when changes are required.\n\nthat makes it a better production desk than asking one bot to message the next and hoping the context survives.\n\n## run the first campaign through the complete team\n\nuse the system itself as the first assignment.\n\ngive signal scout this prompt:\n\n> find the strongest practical content opportunity created by the latest hermes bot mode release.\n\n> \n\n> prioritize a specific workflow a solo operator can build now. return the official source, current audience interest, useful authority clips, the question the finished piece should answer, and reasons to reject weak angles.\n\n> \n\n> do not draft content.\n\nsignal scout should return bot mode as the event and the one-person media company as a candidate workflow.\n\nresearcher then checks the official release, hermes documentation, relevant walkthroughs, and community tests. it records what bot mode, profiles, and kanban actually do, along with the distinction between visible collaboration and durable task execution.\n\ncontent strategist receives that evidence and makes the editorial decision:\n\n- reader: solo creator or operator publishing across several platforms\n\n- outcome: build a six-bot content operation around one shared brain\n\n- tension: faster writing does not solve weak ideas, duplicated distribution, or lost learning\n\n- thesis: a one-person media company becomes possible when specialized bots share accepted knowledge and pass structured work through one feedback loop\n\n- reusable object: six-role operating model, obsidian graph, and handoff record\n\nlong-form writer builds the guide you are reading.\n\ndistribution bot then creates several distinct entryways into it:\n\n1. capability: hermes bot mode can turn six isolated profiles into one visible media team.\n\n1. architecture: the bots are the people, obsidian is the company brain, and kanban is the production desk.\n\n1. x growth: one flagship idea can support a week of x posts without repeating the same hook.\n\n1. research: signal scout and researcher stop weak or unsupported topics before writing begins.\n\n1. compounding: performance updates the shared playbooks instead of disappearing into analytics.\n\neditor reviews the complete package, compares the hooks, checks every factual claim against the source packet, and creates the approval queue.\n\none idea has now travelled through the same system the finished piece teaches.\n\n## turn the media company into an x growth engine\n\nthe larger system can run every platform, but x is the easiest place to see why specialized distribution matters.\n\ndo not ask distribution bot to summarize the flagship piece seven times. give each post a separate reason to exist.\n\nuse this weekly sequence:\n\nday 1: publish the flagship argument\n\nlead with the largest outcome and attach the complete guide.\n\n> i built a team of six hermes bots that gives one person the operating capacity of a complete media company.\n\nday 2: teach the architecture\n\nexplain one useful distinction completely:\n\n> the bots are the people.\n\n> \n\n> obsidian is the company brain.\n\n> \n\n> kanban is the production desk.\n\nthen show what breaks when those responsibilities are mixed.\n\nday 3: use an authority clip\n\nattach a relevant demonstration or creator clip and develop one mechanism it reveals. the post should be useful without requiring the reader to open the guide.\n\nday 4: publish the practical build\n\nshare the six roles, the obsidian tree, or the handoff contract as a standalone implementation post.\n\nday 5: challenge the common workflow\n\nexplain why one ai chat writing every format creates repetitive distribution, even when each individual draft sounds polished.\n\nday 6: show the feedback loop\n\nbreak down which performance signals should update hooks, angles, platform rules, or audience assumptions.\n\nday 7: compress the system\n\nturn the complete workflow into one visual:\n\nidea → research → angle → long-form → distribution → review → performance → updated playbooks\n\nthe result is a week of connected distribution with seven different reader entryways. someone can discover the system through the headline, architecture, clip, build, critique, feedback loop, or visual.\n\nthat is much stronger than posting the same link seven times.\n\n![](https://pbs.twimg.com/media/HQ0BBy_aEAA6V70.jpg)\n\n## keep the human at the editorial boundary\n\nthe first version should prepare everything and publish nothing.\n\nyou still approve:\n\n- the central angle;\n\n- the flagship draft;\n\n- every factual claim carrying real consequence;\n\n- every public post;\n\n- changes to voice, audience, offer, or editorial policy;\n\n- performance lessons that become permanent rules.\n\nthis gives you a fast way to train the system. every approval, revision, and rejection becomes a concrete example of your judgment.\n\nwhen the same decision becomes predictable, move it earlier into the playbook. the editor can learn that you always reject unsupported superlatives, duplicated hooks, generic ctas, or carousels that merely quote the flagship piece.\n\nkeep publishing approval human until the cost of a mistake is genuinely low and the review queue has been consistently boring.\n\nautonomy should remove repeated decisions, not remove your taste from the operation.\n\n## make performance improve the next run\n\nmost content analytics stop at reporting numbers. a useful learning system changes how the next campaign is built.\n\nafter each campaign, record:\n\n- the signal and subject;\n\n- the reader outcome;\n\n- the central angle;\n\n- the hook type;\n\n- the format;\n\n- the platform;\n\n- impressions or reach;\n\n- meaningful engagement;\n\n- clicks, follows, replies, or conversions tied to the goal;\n\n- what the editor approved or revised;\n\n- what should be tested again.\n\nthe performance bot does not need to become a seventh permanent role. give editor a weekly review task that compares recent campaigns and proposes changes to the hooks, angles, x platform, and other playbooks.\n\nrequire the proposal to name the posts supporting it. one strong result should create a hypothesis, not a universal rule.\n\nthe weekly review should return three short lists:\n\n- keep: patterns that worked repeatedly and still match the strategy.\n\n- test: promising patterns that need another controlled attempt.\n\n- stop: repeatedly weak patterns, duplicated formats, or expensive work with no useful result.\n\nyou approve the changes before the shared brain updates.\n\nthis closes the loop. the team no longer begins every campaign from the same generic prompt. it begins with the accumulated judgment of everything you chose to keep.\n\n## build the smallest version this week\n\nyou do not need six fully autonomous bots on day one.\n\nbuild the system in this order:\n\n1. create the obsidian content brain and fill the minimum voice, audience, proof, and platform files.\n\n1. create signal scout, researcher, and editor first.\n\n1. run three real ideas through signal, research, and review.\n\n1. add content strategist once the evidence packages are consistently useful.\n\n1. add long-form writer when the angle briefs are strong enough to constrain a draft.\n\n1. add distribution bot after one flagship format is working.\n\n1. start recording performance and update the playbooks once a week.\n\n1. keep approval human while the team learns your standards.\n\nthe first useful version can simply return a researched opportunity, one angle brief, and one approval-ready x post.\n\nthen add the flagship piece. then the platform package. then the performance loop.\n\nthe destination is a one-person media company. the build still begins with one piece of work you can judge.\n\nhermes supplies the team. obsidian supplies the shared brain. your decisions teach both what deserves to compound.\n\nfollow @vibemarketer_ for more practical ai systems you can build and use inside a real business."},"adhxContext":{"savedByCount":1,"publicTags":[],"previewUrl":"https://adhx.com/VibeMarketer_/status/2093330541177352217"}}