Free AI setups, built for real people
A custom ChatGPT, Claude, Gemini or Grok, built for a specific person's job, without them asking. Every one is public, nothing is gated, and nothing is tracked.
So rather than argue about it, I pick a real person with a real job, build them the thing properly, and publish it. If it is useful they keep it. If it misses, they say so and it gets recut. Either way you get to judge the work instead of the claim.
What every one contains
- One setup file carrying the person's job, market, constraints and writing voice. Loaded once. Every helper uses it. It also carries the router: what each helper owns, and how it picks the right one for whatever you paste.
- Eight helper files for the work they repeat every week, each one laid out the same way: what it needs, how it thinks, what it gives back, its rules, the tricky cases, and a worked example.
-
Two formats.
.mdfor Claude Projects, Gemini Gems, Copilot agents or Grok..txtfor locked-down work chat that won't take uploads. Plus a one-file version for ChatGPT Free. - Brackets where a fact was not public. Team size, targets, authority levels. A gap they can fill in ten seconds beats a number invented to look complete, and a single made-up figure would make the other eight files untrustworthy.
What the files do that a one-line prompt does not
- Set up once. Say what happened and only the right helper answers.
- A missing number means one question, then it waits. It never makes one up.
- The answer comes back in a fixed shape, not as an essay.
- One reason, not three. Three reasons read as three excuses.
- Drafts come back unsent. It cannot send anything in your name.
- Stopping is allowed. "Walk away" and "hold" are good answers.
The builds — newest first
Nicolas C.
Digital writer and ghostwriter · co-founder of Ship 30 for 30
On 31 August he attacked the culture of AI detection as futile. Three days earlier he attacked AI dependence. He needs both, because he co-writes a paid newsletter about writing with AI and teaches the craft for a living. Helper 01 checks for a point of his own before a word is drafted — no claim of his own in the input, no draft.
Open it →Haley S.
Finance educator behind Mrs. Dow Jones · Finance Is Cool · Brooklyn
She teaches millions of people about money decisions they are actually facing, and holds no licence that lets her tell one of them what to do. Every sentence with the word “you” in it runs three checks — no product, no instruction, no promised outcome — and Helper 01 names the instruction it deleted.
Open it →Shawn W.
Writer of Latent Space · organiser of the AI Engineer conferences · San Francisco
In January he wrote that the most important problem in media is scaling without slop, while running a daily newsletter that is 99% agent-generated. He will not blame the machine and will not cut his own output. So every fix has to name the change to the generator, not just the paragraph — and say plainly when the taste does not specify.
Open it →Rob W.
Host of Startups For the Rest of Us · MicroConf · TinySeed
In July he told his audience the future of SaaS is SaaS, and that “X is dead” is a way to get free clicks. On 4 September his fund skipped an entire accelerator batch because AI is reshaping what defensible means. Helper 07 finds the assumption under old advice and checks whether it is still alive in 2026, then dates the correction.
Open it →Chris D.
Founder, Searchable · writer of Step by Step · UK
He posted “Google SEO is dead” in September 2025, then wrote the opposite on his own company blog the day he announced a $14M seed. The audience that used to be the business is now the funnel for it. Helper 01 checks the hook can be found by search and AI before a word gets drafted, and allows exactly one ask.
Open it →Wes K.
Executive coach · teaches Executive Communication and Influence on Maven
In July she went after work that is polished, professional and adds nothing. In September she argued the operator who can go end to end alone now beats the specialist. The same tools made defensible output free. Helper 01 kills the merely defensible drafts first, and 08 names the task she is keeping for the wrong reason.
Open it →Dan S.
Co-founder & CEO, Every · New York
In February he argued a team should shrink to one person per product. In May he published that the company had automated everything and tripled headcount, and said he does not know whether that is a lag or a law. Helper 01 makes him name the choice on every decision — hand it to the AI, or give it a person — and records the call with the thing that would reverse it and a date.
Open it →Shreyas D.
Product leadership teacher and advisor · previously Stripe, Twitter, Google
Spent 2026 arguing that judgement on top of AI output is the last moat, then wrote in May that judgement from below the level of the work is worse than useless. Helper 01 makes every run say which one it just did, and permits the answer he has never printed: this edit was decoration.
Open it →Simon O.
Writer of a media industry newsletter · Washington DC
He audits everybody else's audience numbers, then published his own: 3.5 million views in a month, 35 net new followers. Helper 03 keeps a ledger where reach does not count as success, because what the platform actually pays him in is reporting, and that appears on no dashboard.
Open it →Jay C.
Founder of Creator Science · Ohio
Posted on 2 September that brands overvalue follower counts and that the platform rewards repackaging other people's work, then called the mismatch an opportunity in the same breath. Helper 02 draws the line between borrowing a format and taking the work, and 03 turns his own sponsorship test into a written no.
Open it →Katelyn B.
Buyer psychology writer and trainer · Halifax
She teaches the attention mechanics for a living, and in 2026 every one of them is one prompt away from anybody with a laptop. Helper 03 sorts every draft into an argument that earns attention or a trick that grabs it, and a trick that cannot name its argument does not go out.
Open it →April D.
Positioning consultant · author of Obviously Awesome and Sales Pitch · Toronto
In May she told B2B tech to ignore its future competition in positioning. Three weeks later she wrote that a point of view about the future is no longer optional, and named the very thin line between them without saying where it sits. Helper 02 makes every claim about the future carry proof from today or get cut.
Open it →Peter Y.
Writer of Behind the Craft · host of its interview show
Sells a personal AI operating system, then says a skill only gets the output 80 to 90 percent of the way there. Helper 01 splits a launch into what anyone would have by tonight and what only came out of his own work, and won't let the two share a sentence.
Open it →Lenny R.
Writer of Lenny's Newsletter · host of Lenny's Podcast · Bay Area
Roughly two hundred inbound requests a week, and he declines about 99.9% of them. Helper 01 sorts the pile by shape rather than sender, pulls out the two he owes a real answer to and leaves them unscored. It never drafts in his voice.
Open it →Matt N.
Independent social platform analyst · Geekout · UK
Six to twelve news items a day and a Friday newsletter, run by two people. Nothing runs without a link to the original source — the demo held an item because it was a competitor's reporting, and caught a repeat against an issue eleven days old.
Open it →Ann H.
Chief Content Officer, MarketingProfs · writer and keynote speaker · Boston
Her line on AI has been settled since 2023: a tool yes, an author no. So the setup researches, argues back and structures, and never writes her prose — not an opening, not a transition, not an example. The one email it drafts labels itself as admin.
Open it →Sam J.
Founder & CEO, Pavilion · New York
Said on his own show that growing past 10,000 members diluted the mission, then gated the top tier. No unpublished revenue, churn, price or member figure gets printed — including the version with the number filed off.
Open it →Ross S.
Founder, Foundation & Distribution.ai · Halifax
Has argued since 2014 that production is free, so distribution is the constraint. Helper 01 checks every piece against his own rule about what stays quiet, and holds a client result back rather than publish a win the client never did.
Open it →Arvid K.
Solo founder, Podscan · writer of The Bootstrapped Founder
Built a career on radical transparency, then published the argument against it. Helper 01 checks whether a post would let someone copy the business before drafting a word — lesson out, blueprint in, and the post says what it held back.
Open it →Lia H.
Creator economy expert · author of the ICYMI newsletter
A one-person media business read by nearly 40,000 people, from students to CMOs. The demo answers independent media's hardest email — the sponsor who wants the ad unlabelled. The file sells the label, never the disguise.
Open it →Stuart C.
Chief Revenue Officer, Clarus Networks Group · UK
Founded and ran his own satcom business before this seat, so his setup bans talking down to smaller operators. The demo answers “why not buy a single LEO provider direct” with a 90-day test priced in the buyer's own outage minutes.
Open it →Rand F.
Co-founder & CEO, SparkToro · Alertmouse · Snackbar Studio
The zero-click playbook this whole campaign runs on, set up for its author. Three checks before any data post ships: totals only, the sample stated in the body, and the claim never goes further than the data.
Open it →Steli E.
Co-founder & CEO, Close
The follow-up doctrine half of SaaS learned, turned into files. A bump is a receipt, not a follow-up — and the close-the-file email is real. Fake breakups are banned in writing.
Open it →Nadav P.
Chief Revenue Officer Americas, Ogury
Days into the seat, in a market where every deck says privacy-first. The demo concedes the noise completely, picks ONE differentiator, and hands the buyer a test she can run without trusting a word of it.
Open it →Aaron S.
Chief Revenue Officer, Mitek Systems · San Diego
Owns five functions under one CRO, inside a listed company. The setup carries a standing public-company rule — no forward-looking commitments, no quoted financials, no unmeasured performance number in a customer letter.
Open it →Chris O.
CEO, Caliber
Selling a category that does not exist yet. First rule in the file: never argue against a tool the buyer already owns and likes. Concede the point, then name the question their stack cannot answer.
Open it →Lauren C.
Head of Business Banking, Triodos Bank UK · Bristol
At an impact bank every deal has two cases, not one. The hardest thing she writes is a decline where the numbers were fine — so the file sorts the reason into mandate, credit or structure before drafting a word.
Open it →Morgan I.
Founder & CEO, AMP Social
Coaches, posts and still prospects himself. Helper 02 rewrites a rep's message and explains what changed, so the rep can do it themselves next time. No signal, no message — it refuses to invent a reason to write.
Open it →Mats E.
Head of Sales, Credit Benchmark · London
Sells consensus credit data into bank risk teams. Helper 04 will not draft a reply to a price challenge until he gives it his floor. No floor, no draft — it stops and asks.
Open it →The rules I build under
- Public information only. Their own posts, their company site, the appointment announcement.
- No invented statistics, client names, pricing, team sizes or revenue figures — theirs or their employer's.
- Surnames shortened. They did not ask to appear here, so full names do not go into pages a search engine indexes.
- Say which helper missed and why, and it gets recut. No charge, no pitch attached.
Want one for your own job?
Send your job and company. It comes back inside 24 hours: a custom ChatGPT, Claude, Gemini or Grok, built from your role and your market, in both formats.
Custom AI for Your Job — $49