human://vegard
Vegard — accountable for everything
Talks to you, makes every decision, approves every claim and signs off every deploy. One person from first letter to handover.
How I work · One man. An army of agents.
How my AI agency works: agents propose, I approve.
Agents do the volume: research, drafts, code, tests, checks. I make every decision, set every approval rule and sign off every deploy. You talk to one person from the first letter to handover.
- agent://researcher sources and competitors gathered pass
- agent://writer draft ready for checks pass
- agent://fact-checker unsupported claim flagged · held rejected
- agent://tester all checks green pass
- human://vegard approved · ready to ship approved
Teal is agent work. Ember is a rejection. Gold is a human decision.
The crew
Seven agents. One accountable human.
Each agent has one job, a list of things it’s never allowed to do, and a running cost you can see. All of them report to me.
agent://researcher
Reads your market, competitors, sources and analytics before anything is planned.
Never decides what gets built.
agent://architect
Drafts structure: sitemaps, data models, page outlines, internal links.
Never ships a plan I haven’t approved.
agent://writer
Drafts copy and articles in your voice from your real expertise.
Never invents facts, prices or testimonials.
agent://fact-checker
Checks every claim against the source and rejects what it can’t prove.
Never lets an unsupported claim through.
agent://builder
Writes code, components and integrations to my spec.
Never deploys to production on its own.
agent://tester
Runs hundreds of automated tests, accessibility and speed checks on every change.
Never marks a failing check as passed.
agent://watcher
Watches crawlers, uptime, budgets and errors after launch.
Never spends past the budget cap you set.
The operating principle
Agents propose. A human approves.
On this site, gold always means a human act: a decision, an approval, a seal. Here’s where the line sits.
What agents do
- Research sources, competitors, analytics and your existing content
- Draft pages, articles, code and tests to my spec
- Check every factual claim against its source
- Run accessibility, speed and regression checks on every change
- Watch crawlers, uptime, errors and budgets after launch
What only I do
- Talk to you, from the first letter to handover
- Decide scope, architecture and price
- Set the approval rule for everything that gets published
- Sign off every deploy to production
- Say no, when something shouldn’t be built at all
Verification gates
AI is fast. It also makes things up. So nothing ships until it passes the gates.
When I built a law firm’s knowledge hub, an AI pass suggested 1,175 internal links across 230+ articles. The gates threw out 72 that were invented outright and 179 that didn’t belong. 924 survived and went live.
- 1,175 AI internal-link suggestions
- −72 hallucinated — 36 invented anchors, 36 links to pages that don’t exist
- −179 irrelevant — failed the relevance check
- 924 approved and published · avg 3.97 per article
- Gate 01
Deterministic checks
Code, not a model, checks the boring things first: does the quoted anchor exist word for word, does the linked page exist, is the schema valid, are banned phrases absent.
- Gate 02
Fidelity and relevance
A second model checks every factual statement against the source dossier and marks it supported, partial or unsupported. Unsupported claims don’t ship.
- Gate 03
Automated tests
Every change runs the test suite. The law-firm engine alone has 649 of them.
- Gate 04
Budget breakers
AI spend has a hard monthly cap. On that engine, everything stops at €100 until a human raises it.
- Human gate
Human approval
I approve what ships. For client-facing content you set the rule: one-click approval per item from a daily digest, or automatic publishing for categories you trust.
The process
From first letter to handover. Five steps. No surprises.
-
Your letter
2 minutes
Five short steps: what you need, who you are, your budget and how you like to talk. I read it myself and look at your site before I reply.
-
20-minute call
this week
Your goals, your constraints, and whether I’m the right builder for them. You leave with a clear next step, whether or not we work together.
-
Fixed price
within 48 h
A written proposal: scope, price, timeline and exactly what you’ll own. No hourly meter. Nothing starts until you say yes.
You approve -
Build
weekly previews
Agents do the volume; I check every claim and sign off every deploy. Every week you get a preview link and tell me what to change.
I sign off -
Handover
into your accounts
Code, content, keys, docs, runbook and tests, in accounts you own. Bugs found in the first 30 days are fixed at no charge.
You own it
What you own
Everything. In your name, from day one.
I work inside your accounts with access you grant. When the project ends, there’s nothing to hand back, because it was never mine.
- Domain and DNSRegistered to you.
- Hosting accountUsually Cloudflare, billed to you at cost.
- Code repositoryWith the full history, not a zip file.
- Content and dataExportable, in open formats.
- AI provider accountsYour keys, your billing, your caps.
- AnalyticsCookie-free, in your account.
- Docs and runbookHow to run, change and recover it.
- TestsSo the next developer can change it safely.
Trust
How I handle your data and AI. Plain rules, agreed before we start.
These rules go into every proposal. If your industry needs more, such as a law firm’s privilege or a clinic’s patient data, we write the extra rules down together.
Providers chosen per project
Your proposal lists which AI providers the project uses and for what. You can veto any of them.
No training on your data
Client work runs on providers’ business API terms, under which your inputs aren’t used to train their models. No consumer chat apps.
Keys in your accounts
AI keys and billing sit in your provider accounts, with spending caps you set. When we part ways you revoke my access and nothing breaks.
Sensitive data stays out of prompts
Privileged, patient and personal data is kept out of prompts. Agents work from public content, your sources and anonymised examples unless we agree otherwise in writing.
Human approval before anything publishes
Nothing an agent drafts goes live without passing the gates and the approval rule you agreed (per item in pilot mode, per category in auto mode).
Logged, not hidden
What an agent proposed, what the gates rejected and who approved it is logged, so you can audit it later.
Continuity
What if something happens to me? The honest answer to the bus-factor question.
There’s no hidden team behind me, so I won’t promise you a stand-in. I promise something more useful: everything a competent developer needs to take over is already yours.
Your accounts, from day one
Domain, DNS, hosting, code repository, AI provider keys and analytics are registered to you. I work inside them with access you grant, and can revoke, at any time.
Docs written for the next developer
A README and architecture notes that explain how the pieces fit and why, written for a stranger, not for me.
Runbooks for the things that matter
How to deploy, roll back, rotate keys and restore data, step by step, kept next to the code.
Tests that guard the behaviour
Automated tests run on every change. If someone else edits the code later, the tests say exactly what broke.
A mainstream stack
TypeScript, Astro, Cloudflare Workers, and WordPress where your editors need it. No proprietary framework only I understand.
60 seconds
Rather watch than read? The whole method in one minute.
Film · 60 s · captions on
One man. An army of agents.
60 seconds with WebRevolution.si
Thanks for watching
Be the business AI recommends.
Found by AI. Chosen by people. Owned by you.
webrevolution.si · Personal reply within one business day.
Transcript and scene descriptions. Captions are always on in the player.
-
Scene 1, The question. An AI chat window. A buyer types: “Who can rebuild our website so AI recommends us?” The assistant replies that it can only recommend businesses whose sites it can read and verify, cites two competitors and marks yourcompany.com as having no readable text. Beside it: 51% of B2B software buyers start research in an AI chatbot (G2 buyer survey, March 2026).
Your next client asks an AI before they ask you.
51% of B2B software buyers start their research in an AI chatbot.
-
Scene 2, Through a crawler’s eyes. A website wireframe. A crawler’s scan line passes over it and the page dissolves, leaving only <div id="root"></div> and the count: 0 words. That was my own old site, fetched as raw HTML.
Through a crawler’s eyes, a JavaScript-built site can be an empty page.
My own old site served crawlers 0 words. So I rebuilt it.
-
Scene 3, One man, an army of agents. My portrait in a gold ring. Seven teal agent icons drift in, orbit it, then line up beneath it: researcher, architect, writer, fact-checker, builder, tester, watcher.
One man. An army of agents.
Seven specialised agents do the legwork.
I direct every one, and I answer for the result.
-
Scene 4, Found, cited, chosen. Found, cited, chosen, with four dated receipts: 1,100+ pages live on a law firm’s own domain (Client · law firm); 103 → 0 server invocations from crawlers, before and after the static rebuild (Own product · SitShape); 72 hallucinated links rejected before publishing (Client · law firm); 3 weeks from empty repo to a live, monetised SaaS (Own product · Filyhost). A gold seal stamps: human approval.
Found by AI. Cited with sources. Chosen by people.
Real numbers from live systems, each with a date and a method.
Agents do the work. Nothing ships until I approve it.
-
Scene 5, What I build. Five service cards with starting prices: AI-ready website rebuild from €6,500; AI content & distribution engine from €3,500; AI agents & automation from €7,500; Custom tools & agent endpoints from €8,900; Conversion landing pages from €2,400.
What I build: AI-ready websites, content engines, agents and tools.
Clear scope. Published prices. You own everything.
-
Scene 6, Built by me. My portrait again: “I build every project myself.” Signature line: Agents propose. A human approves.
I build every project myself. No account managers, no hand-offs.
Agents propose. A human approves.
-
Scene 7, Your next step. End card: Be the business AI recommends. Start a project, scan your site for free, book 20 minutes, or send a WhatsApp message.
Start a project, scan your site for free, or book 20 minutes with me.
Questions
How it works, in straight answers.
Do you outsource work to other people?
No. Agents do the volume; I do everything that needs judgement. If your project needs a specialist I don’t replace, such as a photographer or a lawyer, we agree on it in writing first.
Which AI models do you use?
The right one for each job, chosen per project and listed in your proposal. On recent projects that has mostly meant Anthropic’s Claude models for research, writing and code, with OpenAI, Google and xAI models where they fit better. Cuckoo, my own distribution engine (in pilot), works with 4 providers side by side.
What happens when an agent gets something wrong?
That’s what the gates are for. Most mistakes die in the deterministic checks; the rest get caught in review. If something slips through after launch, it’s fixed under the 30-day warranty, and the gate that missed it gets tightened.
How do the weekly previews work?
Every week you get a private link to the real thing, not slides. You comment, I adjust. Nothing reaches production until you’ve seen it.
Can my team keep using the agents after handover?
Yes, when they’re part of what we built. They run in your accounts, with your keys and your spending caps, and the runbook explains how to adjust them.
Why should I trust a one-person studio with this?
Because you can check everything: published prices, receipts with dates and methods, live products you can open, and a handover that leaves you holding every account. Then talk to me for 20 minutes and judge for yourself.
Want to see the method on your project?
20 minutes with me, not a sales rep. You leave with a clear next step — whether or not we work together.