The Two-Week Apprentice · from $49

Let people experience your expertise before they hire you.

You have a way of assessing, deciding, or creating that took years to build. It works. It has never been written down, and right now it only works when you do.

We teach an AI apprentice one decision you make over and over, test it against real cases, and hand it to you as working software. Two weeks and $49. Keep it under a separate ownership agreement, or say no and we delete it.

Book a 30-minute call

30 minutes · we find the decision, who would use it, and how it would reach them · no pitch

Pressing play loads the player from YouTube, and Google receives your IP address at that moment. Nothing is requested before you do.

Built by a builderDesigned and delivered by a former EMT who ships production AI voice systems with crisis escalation.
Tested before it faces anyoneIt ships with test cases showing where it performs, and where it refuses and comes back to you.
Yours, portablyA documented file you keep. Not locked to any platform, any vendor, or me.
A new kind of personal AI

AI can read your work. Can it learn your practice?

Most personal AI is assembled from documents, posts, recordings and voice. Apprenticed AI is taught differently: through real cases, your corrections, the exceptions that break the obvious rule, and tests of where its competence ends.

It does not try to become the whole person. It learns one bounded part of your practice and turns it into a useful tool, with you remaining the teacher, the owner and the final authority.

A clonelearns how you sound.
A chatbotlearns what you have said.
An agentis given tasks and tools.
An apprenticed AIis taught how you handle a case, and when the case must return to you.

One part of a practice. Taught through cases. Tested before deployment. Owned by the expert.

What you end up with

One working tool and the method underneath it.

The $49 trial ends with working software you use on your own de-identified cases. Everything below — putting it in front of the people you serve, the portable skill, the rights — follows only if you keep it, under the separate Ownership Agreement. Personal-data or client-facing deployment requires a separate written privacy and legal review and, where required, a data processing agreement.

01

A tool the people you serve can actually use

One useful thing, deployed where the people you work with meet it: a guided assessment, a structured intake, a first-pass diagnosis, a document reviewer, a readiness check. It asks what you would ask, applies your criteria, explains its reasoning, and says plainly when a case needs you.

Working software, not a mockup. If your field is regulated or safety-sensitive, the first deployment is internal by default and stays there until the appropriate review is done.

The portable skillYour decision logic written down: the questions, in your order, the signals that change your answer, the cases you refuse. Yours to take anywhere.
Test cases and written limitsWhere it performs, where it stops, what it never answers, and when it hands off to a person. Proven before anything goes live.
Ownership termsAt $990 the separate Ownership Agreement grants exclusive rights of use in the bespoke deliverables, and the rights needed to operate and modify the background materials inside them.
They find youYour existing outreach, audience, referrals or network.
They use the toolAnd get something genuinely useful, straight away.
Fit becomes visibleTo them and to you, before anyone spends an hour on a call.
A better conversationThey arrive informed, qualified, and already in your frame.
How it works

Three stages, like an apprenticeship you decide after every step

Two weeks on the $49 trial. Twenty-one days on the human-led build, where the extra time is mine, not yours.

stage 1

Shadow

The apprentice watches how you handle real, de-identified cases you bring. About four hours of structured interviews, scheduled around your calendar. Not your documents - your actual reasoning, case by case.

GateIf your method does not reduce to something you would stand behind, we stop and say so.
stage 2

Practise and correct

It attempts new cases on its own. You correct its reasoning, its conclusions and its boundaries - the same way you would correct a capable junior who is nearly right.

GateYou approve every rule. Nothing you have not read goes live.
stage 3

Qualify and deploy

It has to pass the test cases you agreed, explain how it got there, and hand uncertain cases back to you. I try to make it overstep; it has to refuse. Then it goes live.

GateOn the human-led build: not delivered in 21 days and I keep working at no further cost until it is.

Your time, either way: roughly four to six hours in total.

Why not just use a platform?

Owned, not rented.

Hosted clone platforms

  • Subscription forever, and a share of your revenue if you monetize through them
  • You do the setup yourself, from what you have already written down
  • Stop paying, and it is gone

If a content-recall chatbot is genuinely what you need, a platform is the right buy, and I will tell you so on the call.

The Judgment Build

  • No subscription and no revenue share. You are not renting your own judgment back
  • It learns the way an apprentice learns: by working through your real cases, not by reading your files
  • Tested competence boundaries, not conversational similarity
Where this works

One method many judgments.

Six examples. Yours is probably not on this list, and that is fine: the method captures judgment, not job titles. Work that turns on health, employment or legal decisions is considered only on the full build, and only under a separate review and agreement.

Design & product

Design studio lead

"Is this a brand problem or a product problem? Because the answer changes everything downstream."

Becomes
  • A diagnostic a prospect completes in ten minutes and gets a real read on what they are actually facing
  • Enquiries that arrive already scoped, instead of "can you do a website"
Content

Content strategist

"Should this piece build trust or drive action? Trying to do both is why most content does neither."

Becomes
  • A tool that reads a client's brief and calls it, with the reasoning shown
  • A standard her team applies without checking with her every time
Brand

Brand consultant

"Is this a naming problem or a promise problem? Renaming a company that cannot say what it does buys you nothing."

Becomes
  • A read a prospect gets before the pitch, telling them which of the two they actually have
  • Fewer projects that start as a rebrand and turn into positioning work halfway through
Publishing

Editorial director

"Does this manuscript need a structural edit or a line edit? Line-editing a broken structure is the most expensive mistake in publishing."

Becomes
  • A first-pass assessment an author runs before commissioning anything
  • A standard her freelance editors apply without asking her every time
Engineering

Technical architect

"Should this be one service or three? Almost everyone splits too early and pays for it for years."

Becomes
  • A design review a team runs on their own proposal before it reaches him
  • Reviews that arrive with the obvious objections already answered
Operations

Operations consultant

"Is this bottleneck the machine, the schedule, or the handover between two shifts? It is almost never the machine."

Becomes
  • A diagnostic a plant manager completes before booking a site visit
  • Visits that start at the real constraint instead of finding it on day two
Founding cohort

Try it for $49. Own it if it earns that.

Two weeks and forty-nine dollars gets your judgment captured and built into something you can use on real cases. At the end you decide whether it is worth keeping. If it is not, it is deleted rather than quietly parked on my servers, and you get a record of what was destroyed.

If the build works and you want to, I would like to write it up afterwards. That is optional and not a condition of the price, and it can be fully anonymised: no name, no employer, no identifying detail, nothing published that you have not approved line by line. Plenty of this work is confidential by nature. Say no and nothing changes.

What this does not do: it will not build you an audience. If nobody currently finds you, a tool sitting at the end of that silence will not change it. What it can change is what happens after someone does find you - whether they leave with something useful, and whether the conversation that follows is worth both your time.

The endpoint is a working asset you own. Where you take it after that is yours: kept private, given to your team, put in front of clients, or licensed into someone else's product. You keep the rights and the licensing template either way.

What you can keep, and on what terms, is set out in the separate Ownership Agreement.

The Two-Week Apprentice
$49two weeks

One decision, captured and built into working software you use on your own cases.

$990 to keep it, under the separate Ownership Agreement. The tool, the written method, the test cases and the boundary documentation come with it. The $49 comes off.
$2,900 for the human-led build. I run the capture personally over 21 days. Regulated or safety-sensitive work is considered only under a separate review and agreement.
Day 14 is a decision, not a renewal. There is no automatic renewal and no automatic Day 14 charge.

If you do not complete the ownership purchase by Day 14, active working copies are deleted by default and we send you a deletion record. Inaccessible backups may remain in rotation for up to 30 days. Invoices and minimal legal records are retained for the periods required or permitted by law.
Book a 30-minute call

The $49 trial opens with the first cohort. Until then, the way in is a conversation.
Prices are in USD. Applicable VAT or other transaction taxes will be added or accounted for as required by law. Offered only to businesses and self-employed professionals acting for a professional purpose. Consumer orders are not accepted and all orders are subject to verification.

Ownership. For USD 990, the separate Ownership Agreement grants exclusive rights of use in the bespoke deliverables and the rights needed to operate and modify incorporated background materials. Third-party components remain subject to their own licences. German copyright itself is not sold or transferred. The separate ownership agreement includes 12 months of hosting. Metered AI-provider usage and other variable third-party charges are not included. Personal-data or client-facing deployment requires a separate written privacy and legal review and, where required, a data processing agreement.

Cancellation and refunds. You may cancel for any reason before the first capture session begins and receive a full refund. Once that session begins, the trial cannot be cancelled for convenience. This does not affect any rights arising from our non-performance or materially defective performance, or any rights that cannot lawfully be excluded. If we decline the proposed subject, or if for reasons not caused by you we do not make a materially conforming trial tool available by Day 14, we will refund the $49. Your decision not to purchase the ownership tier does not by itself entitle you to a refund.

Who is behind it

I'm Henrik Grünfeld. Former EMT. I build and run production AI voice systems for older adults, including a daily check-in service with crisis escalation that I operate myself. The capture method in this build is the one I used to turn my own emergency-services judgment into that product's escalation design - and the one I have since run on myself, to find out where it holds and where it does not.

Engagements contracted through HG Connect UG (haftungsbeschränkt), Germany.

A good fit if
  • You have a client-facing practice and a way of working refined over years
  • You can name cases where your judgment changed the outcome
  • People already find you somehow - referrals, network, audience, outbound
  • You can give four hours plus two review cycles
Probably not if
  • You want a general-purpose clone of yourself
  • You want guaranteed leads or passive-income promises
  • You would rather your method stay unexamined
  • You need this to replace human contact instead of routing to it
Subjects we decline

We decline high-impact, regulated and safety-sensitive uses that cannot be responsibly reviewed within this trial. Some may fall within prohibited or high-risk categories under the EU AI Act; others are excluded as a product-safety decision.

In practice that means a judgment turning on health, employment decisions, financial or credit matters, legal advice, children, biometric data, or anything with consequences for an identifiable individual. Work of that kind is considered only on the full build, and only under a separate review and agreement.

We ask what you want to build before you pay, so a decline costs you nothing. If we decline after payment, you are refunded in full the same day.

Straight answers

What people ask before they book.

Is this a digital clone of me?

No. A clone aims to reproduce your knowledge, voice or personality. This learns one bounded part of your professional practice, and nothing else.

Is the AI actually being trained?

It is taught at the system level: structured instructions, your cases, your corrections, examples, retrieval, tools and evaluations. That does not necessarily mean changing the underlying model's weights, and I will not claim otherwise.

How is this different from a custom GPT?

A custom GPT is configured with instructions and documents. This build starts from your cases, records the corrections and exceptions, writes explicit refusal boundaries, tests the resulting behaviour against cases it has not seen, and hands over the method and the tool with ownership terms.

Does it replace me?

No. It handles a bounded first layer and returns anything outside that boundary to you. You remain accountable for the professional work and the final decision.

What happens if I decide not to keep it?

On day 14 everything is deleted — the capture material, the tool, the deployment — and you get a record naming each thing destroyed and how that was confirmed. Two things survive and I would rather say so: backups expire on their own schedule within 30 days, and the invoice, which German tax law requires me to keep.

Who owns it?

For USD 990, the separate Ownership Agreement grants exclusive rights of use in the bespoke deliverables and the rights needed to operate and modify incorporated background materials. Third-party components remain subject to their own licences. German copyright itself is not sold or transferred.

Will it bring me clients?

It will not create an audience. It gives the people who already find you something useful, makes fit visible earlier, and improves the conversation that follows. That effect still has to be measured in the first builds, and I would rather say so than promise a pipeline.

Start with 30 minutes.

Bring one decision you make over and over, where your judgment makes the difference. You leave knowing whether it can be captured, who would use it, and whether the build fits.

Eight assessment conversations this round. Build slots scheduled one at a time.

Book a call

hg@kogsy.ai

The first cohort

Want the $49 trial when it opens?

Leave your details and I will tell you the day it opens — and call you back before then if you would rather talk it through. You hear from me directly, and from nobody else. No mailing list.

Controller: HG Connect UG (haftungsbeschränkt). We use your name, email, optional phone number, role and message to respond and contact you about the Judgment Build under Article 6(1)(b) GDPR. Netlify receives and stores the form submission for us, including under the international-transfer safeguards described in our Privacy Notice. We delete it when the inquiry ends unless a legal retention duty applies. Name and email are required so we can respond; the other fields are optional. Do not include client or other third-party personal data. See the Privacy Notice for your rights and full details.