Home / Articles / Show me the evidence: websites for people looking for a job, built by an agentic cell, with seven pages that prove instead of list, a page made for each employer, a pack for recruiters, eight personas to fork, a ladder from a free recipe to a recruiter's cell, and the brief for the team that builds it
Show me the evidence: websites for people looking for a job, built by an agentic cell, with seven pages that prove instead of list, a page made for each employer, a pack for recruiters, eight personas to fork, a ladder from a free recipe to a recruiter's cell, and the brief for the team that builds it
By Dinis Cruz · 2026-11-16 · article v1.2.0, 3 versions · newsroom v0.14.126 · careershiringagenticcellstemplatescellspersonasrecruitersarticle
Abstract: I want to help people I know get jobs, and the best thing I can give them is a website that shows the evidence. Now that anyone can generate a polished CV, hiring managers trust CVs less: in Robert Half's survey 65% said AI-generated applications make skills harder to verify, and in Greenhouse's 91% of recruiters had spotted deception. So the answer is not a better list; it is proof. Show me the work. Show me your thinking. Tell me what you do outside work, what makes you you, and what you think about my industry. And if you are meeting me, make a page for me. So this is the design of a job-seeker cell: a site of seven pages, home, evidence, thinking, me, a page for you, working with me and a recruiter pack, plus the files machines read; an agent team that interviews the person, finds the evidence, drafts in their words, designs a theme that looks like them, tailors a page per application and checks every fact, while the person approves and sends; eight synthetic personas, from a developer to a security engineer, a senior leader and a graduate, each a template anyone can fork onto GitHub Pages; a ladder from a free recipe to a recruiter's cell; and the brief for the team that builds it. The same cell serves virtual and fractional professionals too, a vCISO, a vCTO, a vDPO, a designer or an editor for hire, who sell their judgement and the agent team behind it. The mock-ups are below. They are also the references our agentic sites have been missing.
In short
- CVs are trusted less than ever, because anyone can generate a good one. Evidence is trusted more.
- A job-seeker site with seven pages, each answering a question a hiring manager actually asks.
- A page made for you: for each serious application or meeting, unlisted, with something made for that employer.
- A pack for recruiters, so the people who want to help can forward one link.
- Make it yours: it should look like its owner, not like every other AI site.
- An agent team builds it; the person approves and sends.
- Virtual professionals too: a vCISO, vCTO, vDPO, designer or editor, selling their judgement and their agent team.
- Thirteen personas to fork, a ladder from a free recipe to a recruiter's cell, and the brief for the team.
Why now
I know a lot of people looking for a job right now, some of them very good, and I want to help. I have written the advice before: act as the job you want, do the work in public, and give the people who want to help you something they can forward (act as the job you want). And the edge, when two candidates look the same, is to send something they have not seen, about their own problem (hired for the workflow).
What has changed is the other side of the table. Anyone can now generate a polished CV, a cover letter and even a portfolio project, and hiring managers know it. In Robert Half's survey of more than 2,000 US hiring managers, published in March, 65% said AI-generated applications make skills harder to verify, and 67% said they slow hiring down. In Greenhouse's, 91% of recruiters had spotted candidate deception, and 22% of hiring managers had caught prompt injections hidden in CVs. A better list will not fix that. Proof will.
So my message to anyone looking for a job is: show me the evidence. Show me what you have done, with links I can open. Show me your thinking. Tell me what you do outside work, what you are passionate about, what makes you you. Show me what you think about my industry. And if you are going to meet me, make a page for me.
Seven pages that prove
Every site the cell builds has the same seven pages, because each one answers a real question:
- Home: who are you? One line, in your own words, and three proofs with numbers, each linked to its evidence. Recruiters decide fast on a handful of facts, so the name, the role, the city and the links are at the top.
- Evidence: show me. Two to four case studies, each with the context, what you did, the result, and links you can open: a repository, a post, a talk, a live site, a reference. And one line under each: "ask me about", the conversation you want to have.
- Thinking: how do you think? What you write, your principles, and a short note on how you use AI, because honesty about it now reads as confidence.
- Me: who are you? Hobbies, side roads and a "now" section. When I interviewed people I always looked for what else they had done; it was often the most interesting part (connect your dots backwards).
- A page for you: why us? Made for one employer or one meeting, below.
- Working with me: a short user manual, so we save each other time.
- Recruiter pack: can I put you forward? One screen, below.
And for machines: the same facts as a JSON Resume file, schema.org markup and a PDF. An llms.txt is cheap to add, but I found no evidence that any screening tool reads one, so it is an option, not a promise.
The rules are the same on every page: city only, no address or date of birth; employers anonymised unless they agreed; the same dates and titles as LinkedIn, because recruiters cross-check; and every claim linked to something you can open.
Eight people looking for a job
I made up eight people, each with a different job and a different story, so that we have references to build from. They are synthetic: invented names, invented facts, there to be forked.
- Tomás Reis, backend developer. His home page is a terminal, with his life as a
git log, from a football site for his uncle in 2017 to cutting a payments API's latency from 840 to 190 milliseconds. - Leila Haddad, agentic-era developer. A lab notebook, with her agent runs as a table: what each run did, what she rejected, what it cost. Her best case study is the migration an agent deleted, and the policy she wrote after it.
- Marta Sousa, project manager. Her CV is a kanban board: done, doing, next. Eight years in hotel operations, then four months managing releases for an open-source project, in public, with a reference from its maintainer. She is the person from act as the job you want, a year later.
- Kwame Mensah, agentic project manager. Mission control: 3 people, 34 agents, 410 articles in ten days, with the roster, the review gate and the bill.
- Inês Duarte, application security engineer. A findings report: six vulnerabilities disclosed responsibly, a ruleset others use, and a champions programme. Her rule is on every page: never test without permission.
- Robert Hale, engineering leader. A broadsheet. Instead of achievements, a decision log: what he decided, the options he rejected, and the four he would not make again, because that is what a board is hiring.
- Ana Lima, product designer. A studio, warm and visual, back after a two-year career break raising her daughter, and saying so plainly, with the charity redesign she did during it.
- Sam Okafor, graduate. A zine, loud and honest: a live dashboard of bus delays from his dissertation, a stock app for a food bank, and an open notebook of what he is learning.
Make it yours
Look at the eight again: none of them looks like the others. That is deliberate, and it is the part I care about most. Most AI-built sites look the same, and a hiring manager can tell in a second. A job-seeker site should look like its owner. A developer who lives in a terminal gets a terminal; a designer gets a studio; a project manager gets a board; a leader who writes gets a broadsheet.
So the cell's designer does not pick from three themes. It asks: what do you love, what do you use every day, what would make you smile when you open it? Then the person geeks out and makes it theirs: their photos, their hobbies, their jokes, their colours. That is also the best test of the work: if the person is not proud of the site, it is not finished.
A page for you
This is the most powerful page, and the one almost nobody makes. For each serious application, and for each important meeting, the site gets a page made for that employer or that person. It is unlisted, out of the sitemap, and expires.
It starts with a personal message: I read this, I noticed that, here is what I think your problem is. Then something made for them. Tomás found a bug in the employer's public SDK that matched one he had fixed, and sent a pull request. Marta booked a trip through a travel startup's app and mapped where she thinks they lose people. Kwame drafted their next case study in their own style, with the cost. Then, if it helps, a plan for the first 90 days.
It is the move that worked for me long before agents: an interview where they already knew my OWASP project, so the conversation started from what I could do, not from what my CV did not say (hired for the workflow). Agents make it possible for every application.
Two rules. Public material only: Inês's page for a health company says it in the first paragraph, she read their documentation and tested nothing, and never would without permission. And when the page is public, the employer is anonymised: "a logistics scale-up in Lisbon", not its name.
For recruiters
Recruiters are among the people who want to help, and they need something they can forward (the man behind the curtain). So every site has a recruiter pack: role fit, availability, right to work, notice, languages, references, three links to send, the downloads, and a line saying the client may see it.
And a recruiter can run a cell of their own: a shortlist page per role, made from the candidates' own pages, with the evidence each one sent, a note, an expiry date and consent recorded. In the mock-up, two candidates sent work about the client's own problem before the first call, and the third sent a PDF. Which two would you meet? Agency tools already make client-facing candidate presentations; what they lack is the candidate's own evidence, and that is what this adds.
Virtual professionals
Many of the people I talk to are not looking for one job. They want to be a virtual CISO, a fractional CTO, an outsourced data protection officer, a designer or an editor for several organisations at once. That market already exists. What is new is what they can sell: not only their judgement, but the agentic team they run behind it. A vCISO with agents that keep the audit evidence and draft the questionnaire answers can serve five scale-ups properly instead of two badly. That is a better offer, and it also proves they can run agentic teams, which their clients will soon want to learn from them.
So I added five more personas, synthetic like the others:
- Helena Brandt, virtual CISO, for five scale-ups. Her agents collect audit evidence, draft questionnaire answers from an approved library, keep the policies current and draft a one-page board report; she signs every answer and presents every report.
- Rafael Costa, virtual CTO, for non-technical founders. His agents review every pull request, watch dependencies and the cloud bill, and write a Friday digest; they never merge, deploy or spend.
- Priya Shah, virtual DPO, appointed for eleven organisations. Her agents keep the records of processing current, triage access requests, draft impact assessments and start the 72-hour clock on a breach; she makes every judgement, and the agents see redacted copies by default.
- Joana Matos, fractional product designer, a studio of one person and four agents: synthetic users who test every prototype, a prototyper, a keeper that holds the design system in step with the code, and a researcher.
- Elliot Grant, fractional editor and librarian, a newsroom for hire: a researcher, a writer, a fact-checker, a librarian and a translator, and nothing published without his edit and the client's approval. He is for the organisations that want to hire a journalist, and get a desk.
Their sites have the same seven pages, with two changes. The home page shows the team behind the person: each agent, what it does, and the human check on it, with a line on what it can touch (where assist becomes execute). And the recruiter pack becomes an engagement pack: the model, how many clients at once, insurance, where the client's data stays, and the agents' written behaviour policies, shared before anything starts. The page for you works even better here: Helena's sketches the first ninety days of ISO 27001 for a health-tech company, Joana's runs fifty synthetic users through a prospect's public sign-up, Elliot's is a sample weekly briefing built from a prospect's members' public news.
Each client gets its own cell, with its own vault, policies and digest (the cell is the unit). That is also how these professionals scale without cutting corners: the agents add capacity, the person stays accountable.
The pipeline and the team
The cell runs in seven steps. It starts with a conversation, not a form: the look-back interview from connect your dots backwards, one question at a time, asking for each dot what you can show. Then it collects the evidence, drafts in your words, designs the site with you, tailors pages as applications come, checks everything and publishes.
Behind it is a small agent team, each role with a brief, a deliverable and a behaviour policy (the cell is the unit): an interviewer, an evidence-hunter, a writer, a designer, a tailor, and QA. QA is the one that matters. It checks every fact against its evidence, every link, privacy, consistency with LinkedIn and the CV, accessibility and the phone layout, and it blocks a page that invents anything. The policy for the whole team is short: public material only, never invent a fact, never send, post or apply; the person approves every page and sends every message.
From a recipe to a recruiter's cell
The same cell comes in six sizes:
- The recipe, free: a prompt and a checklist to run with your own Claude or ChatGPT.
- Fork a template, free: pick the persona closest to you, fork it, point your assistant at your copy, live on GitHub Pages (fork it).
- Set up for you, £500: the interview, the evidence pack and the site, live on your domain, made by the site factory (the site factory works).
- Coached, £1,500: two sessions with a person, and tailored pages for your next applications.
- A team for your search, £5,000: research, tailoring, QA and a weekly debrief for the whole search.
- A recruiter's cell, £20,000: many candidates, shortlists for clients, consent and expiry built in.
The first two matter most. Most people should start there, and many will stay there, which is fine.
For us first
We should use this ourselves first. I made a version for me, from what is already public on this site, as a "work with me" page with a map of my work. I will help people on our team who want one to build theirs, starting from a draft made only from what we have published about our work together, and nothing goes live until they have rewritten it in their own words.
And there is a second reason. Our agentic sites have not looked good enough, because we had no good references. These mock-ups are a pile of them, and the pile should grow: every persona is also a template for the hub.
The brief for the team
This is the brief for the agents and people who build the cell. The mock-ups are a single HTML file the team has; build from them, and as the cell is built, compare it with this article and fix whichever is wrong.
- The site template. Seven pages and the machine files, from one content file per person: profile, proofs, case studies, posts, hobbies, manual, pack. Static, fast, accessible, easy to read on a phone, deployable on GitHub Pages, and fork-ready (fork it).
- Eight themes, then many more. The eight personas as forkable templates, each a full site with sample content clearly marked. Then a theme per job family: nurse, teacher, lawyer, chef, data scientist, sales, architect, researcher.
- The look-back interview as the front door: voice or text, one question at a time, each dot marked remembered or evidenced (connect your dots backwards).
- The evidence-hunter: finds and checks the person's public work, and asks before it uses anything about a former employer.
- "Make it yours": a designer agent that interviews the person about their tastes and proposes three directions, never the same three twice.
- Pages for you: a tailor agent that reads an employer's public material and drafts the note, the thing made for them and a 90-day plan; unlisted, expiring, anonymised when public; and a version for a meeting, not only an application.
- QA as a gate: facts against evidence, links, privacy, LinkedIn match, accessibility, phone layout, no hidden text, no invented facts. A failed check blocks publishing.
- The recruiter pack and the recruiter's cell: a one-screen pack on every site, and shortlist pages for recruiters, with consent, expiry and client decisions flowing back.
- Synthetic users who review every template: a hiring manager with ten seconds, a recruiter on a phone, a sceptical CTO, a screen-reader user (synthetic users for your website).
- The ladder on the cell's page: the recipe and the fork free, then set up, coached, a team, a recruiter's cell, with the side panel from the cell is the unit.
- A weekly debrief for the search: applications, replies, what worked, what to tailor next, for the person and nobody else.
- A hub of examples: every template and every real site whose owner agrees, listed on the hub (a hub for agentic sites).
- Virtual professionals: the agent-team block and the engagement pack as options in the template; per-client cells with their own vault, policies and digest; themes for vCISO, vCTO, vDPO, designer and editor first, then legal, finance, HR and marketing.
- Ourselves first: sites for me and for team members who want one, each reviewed by its owner before it goes live.
What this does not claim
- The thirteen personas are invented, with invented facts; any resemblance to a real person is accidental.
- The mock-ups are designs, not a product; the cell, the templates and the recruiter view do not exist yet.
- The evidence for personal sites themselves is thin: the surveys show that trust in CVs is falling, not that a site gets you hired.
- Prices follow the AgentiCCells ladder, which its site calls first proposals.
- Nothing here is automatic about applying: agents draft; a person approves and sends.
Where this comes from
My own thinking, 16 November 2026, and the mock-ups made for this article. Research read that day: Robert Half, AI-generated applications are slowing hiring (10 March 2026); Greenhouse, an AI trust crisis (19 November 2025); JSON Resume; reports on llms.txt requests; and personal sites I admire, Julia Evans, Simon Willison, Troy Hunt, Vicki Boykis and Will Larson. Earlier on this site: act as the job you want, hired for the workflow, connect your dots backwards, the man behind the curtain, the cell is the unit, fork it, the site factory works, synthetic users for your website, a hub for agentic sites.
Peer review Not reviewed yet. Review it
Threads
Builds on
- Act as the job you want: if you are looking for a job, you have time, so do the work in public, make it linkable, and let the people who want to help you help you If you are looking for a job you have time: act as the person in the job you want, in public, and give the people who want to help you a link.
- Hired for the workflow: why October 2026 is the moment to get into agentic work, why you are more technical than you think, and how the edge starts with the application October 2026 is the moment to get into agentic work: companies need AI operators, not model builders, and the edge starts with the application.
- Connect your dots backwards: a look-back interview with your LLM, for founders, to find the non-linear path, the doors that were not what they seemed, and where a startup may be hiding Let your LLM interview you about your path: find the side roads, the misread doors, what memory rewrote, and where a startup hides.
- The man behind the curtain: the emails from my agents are from me, there is an article behind each one, and every line has to buy the next Emails from my agents are from me: the Wizard of Oz in reverse, an article behind every answer, and an email where every line buys the next.
- Where assist becomes execute: tiering AI agents by what they can reach and whether it can be undone, how Agent Behaviour Policies drew that line for more than a hundred agents at Web Summit, and the rule that failed because it was only a rule Where assist becomes execute: tier AI agents by reach and undo class, enforce each tier with a boundary, and learn from the rule that failed.
- The cell is the unit: positioning AgentiCCells around small, focused cells, each with an objective, a deliverable and a behaviour policy, each with a page that sells it, a free recipe and three ways to get help, a catalogue that grows one need at a time, in each industry's own words, and why we should run our own work in cells first The cell is the unit of AgentiCCells: one focused job with an objective, a deliverable and a behaviour policy, each with a page that sells it.
- Fork it: making forking the way people adopt our cells and template sites, as GitHub made it the way people adopt code, with five ways to make a cell yours, what a fork-ready repository needs, paths from one repository to the next, a hub that lets usage decide what we build, and the brief for the team Forking as the way people adopt our cells: fork a template, point Claude at it, and it is yours. Five ways, fork-ready repositories, and a hub.
- The site factory works: three AgentiCCells sites made by hand, then one made by a commit in two minutes and thirty-nine seconds, live over https with a first version written by Claude Code, and the long list of what it took to make something that hard look easy Three AgentiCCells sites by hand, then one by a commit: live over https in 87 seconds, a first version by Claude Code at 159.
- Synthetic users for your website: how to set up five invented visitors who read your site one screenshot at a time, with Claude, a sub-agent per persona, and an encrypted vault you can publish anywhere Onboarding for synthetic users: invented personas, a sub-agent each, a real browser one screenshot at a time, all in a vault you publish.
- A hub for agentic sites: every site AgentiCCells builds in a repository and an sgit vault, served at its own address, and listed on hub.agenticcells.site, where it can be found, promoted and copied as a template; with an orange grower as the first example hub.agenticcells.site: every agentic site in a repository and an sgit vault, served at its own address, found, promoted and copied as a template.
Article No. 438 · start a reading list with it
Reviews
No reviews yet.