Live & in active development · wiz-agents.com

A model gets you 95%.The last 5% is a person.

Subscribe to a real expert's avatar agent — a protected proxy for a lifetime of judgment, plugged into your tools over MCP. Their knowledge stays theirs. You get the 5% a big model can't fake.

✦ Nothing scraped✦ A named human behind every answer✦ Cancel anytime

The Dean · WizAgents Academy · your guide on this page

Welcome to the academy. Every mind here is a real one — ask, and I'll take you to the right hall.

An honest note on stage. WizAgents is a real product I'm building, in active development at wiz-agents.com — not yet launched at scale. The subscriber counts, renewal rates and dollar figures in this case study are illustrative seed-stage placeholders, not measured traction. Verification and certification are designed mechanisms, not yet independently audited or legally vetted. I design and build the platform; I don't offer regulated advice as a service.
Why WizAgents exists

Anyone can be anyone's teacher.

Expertise isn't only titles and tenure — it's a lifetime of lived mastery: a trader's instinct, a surgeon's hands, a founder's scars, a professor's decades. Your knowledge, your way of seeing, shouldn't go quiet in the AI era.

And AI, far from replacing that person, is the perfect medium for them — the vessel that carries a legend's judgment to you, and lets you walk a completely different road beside a master of the field. Not a search box. A journey, with someone real in the room.

30+
Expert domains
MCP
Agent URL + token
4h
Human-escalation SLA
Dated
Provenance per answer
The thesis

A model gets you 95%. The last 5% is a person.

Imagine you're job-hunting, so you rehearse with ChatGPT, Claude, Gemini, Grok — questions, answers, objection-handling. But a sharp interviewer runs the same playbook and pre-empts your pre-emption. The models are right maybe 90–95% of the time, because they share one logic. The smart people hiring for the top roles ask you the other 5%.

That 5% — your human judgment, your read on where the market is heading, a life actually lived — is the part that can't be automated. WizAgents puts it back in the loop: an expert strengthens an agent's reasoning with a lifetime of learning, and a named human stands behind the answer.

Shared LLM logic
~95%
The automatable majority — retrievable, patterned, and roughly the same from any large model.
Irreplaceably human
5%
Where the value and the risk live — and where an expert earns their keep.
§
Judgment
The call a rulebook can't make — weighing what the model only lists.
Foresight
A read on where the market and the rules are actually heading next.
A life lived
Scar tissue and background no training corpus contains — the proof of being human.
Accountability
A real name that signs the answer and owns being wrong.
The idea

Not just a store. A square — a place to think, in public.

Plato's Academy · Ἀκαδημία for an age when the open web is a Library of Babel
LearnResearchBuildDiscuss

WizAgents is closer to an open public space than a checkout. Certified experts and masters become the agents you sit with — and the model isn't replaced. It's made warmer, because a human is in it. That's the whole idea: one more human in the loop makes the thing you're building have temperature.

Where the open web — and raw model output — is a Library of Babel, endless and unverifiable, this is the Academy: curated, accountable, human. It is a place where both sides find a new self in the AI era — a square you come to in order to start, to learn, to research, to build, and to argue in good faith. Expertise, out in the open, instead of buried.

師匠 · Mentor
Mentor agent
A master's judgment, on call — to guide a decision, not just answer it.
助手 · Assistant
Assistant agent
The expert's method applied to your task, with their standards baked in.
対話 · Dialogue
Conversation agent
Someone to think out loud with, who has actually been there.
探究 · Inquiry
Discussion agent
A sparring partner for the hard, open questions — pressure-tested by a human.
The avatar

An avatar, not the person. Their knowledge stays theirs.

This isn't a talent market — you don't rent the expert. A scholar or specialist publishes an avatar agent: a proxy that carries their judgment, while the knowledge base and the lifetime behind it stay protected.

Protected
Their knowledge isn't handed over
The corpus and background aren't dumped, scraped, or resold. The avatar answers from them behind a boundary the expert controls — expertise felt, not exposed.
Re-asserted
Irreplaceable value, made a living asset
A lifetime of learning earns again — a standing demonstration of what a real expert brings that a general model, averaging everyone, cannot.
Objective
A real viewpoint, from more angles
Consult real experts — several of them, each with a distinct stance — and triangulate a decision from more than one genuine angle, instead of one model's single averaged voice.
The moral line

Open knowledge is a gift. Taking it is not.

Sharing research openly is one of the most selfless things people do — it's how the world keeps advancing. But when a platform scrapes that hard-won work without consent or pay, and keeps the proceeds for itself, that isn't scholarship. It's a moral failure dressed up as progress.

WizAgents is built as the opposite. Nothing is scraped. The expert chooses to publish, keeps ownership, is named on every answer, and is paid — the value flows to the person who earned it, not the platform that merely hosts it.

Consent
Nothing is scraped
A corpus enters only because its author chose to publish an avatar from it — never harvested from the open web against their will.
Attribution
Your name on every answer
Provenance and endorsement keep the expert visible and credited — the opposite of knowledge laundered into anonymous model weights.
Ownership
Your knowledge stays yours
It answers from behind a boundary you control, and you can leave and take your record with you. It is never quietly absorbed and kept.
Compensation
You are paid
Recurring revenue goes to the expert; the platform earns a fee for infrastructure and trust, not by pocketing a lifetime of someone else's work.
A standard to be held to. This only means something if WizAgents lives by it — a revenue share that favours the creator, no training on an expert's corpus without explicit consent, and the freedom to leave with your reputation intact. It's a commitment, and a line you should judge us by.
Who it's for

Two sides, both finding a new self.

One side has a lifetime to share; the other needs a mentor who has actually lived it.

For the expert
A professional displaced by 2026's AI restructuring — turning a career into an asset instead of a redundancy.
A retired professor or specialist whose lifetime of learning shouldn't go quiet.
Publish an avatar agent that works while your knowledge and background stay protected — building durable authority, income, and reputation that prove what makes you irreplaceable.
For the seeker
A real expert's viewpoint — to judge a problem from more angles, more objectively, for the 5% the models can't reach.
An expert's method inside your own tools, over MCP, not a tab you copy out of.
An answer you can trust and trace — dated, sourced, and signed by a real person.
The product

A complete marketplace — browse, subscribe, connect.

You subscribe to an expert's avatar, not the person. It packages their continuously-updated knowledge into a 24/7 advisor that lives in your own stack over MCP.

Browse
A marketplace of verified agents
Search across 30+ domains and see each agent's rating, subscriber count, and renewal rate before you commit. Featured agents earn their place on high renewal, not ad spend.
Subscribe
A monthly seat, not an hourly bill
Experts price their own agents. You subscribe monthly, try it in a sandbox first, and cancel anytime — a $200–300/hr consultant replaced by an always-on one.
Connect
MCP-native, into your stack
Every subscription is an agent URL plus a token. Drop it into your MCP client and the expert lives inside your workflow — metered, with one-tap token rotation.
Why subscribe, not download

The moment you subscribe, the world is already moving.

It's the AI-singularity era — regulation, research, strategy, and architecture change by the day. A rule is rewritten; a new paper overturns the old; a trading edge stops working; a stack goes out of date. A one-time download rots. A subscription stays alive — the expert's knowledge base keeps moving, and the agent knows the next day.

Regulation
A rule is rewritten
Yesterday's compliant answer is now wrong. A static model won't tell you; a maintained expert layer will.
Research
A new result overturns the old
Consensus in science and math moves. The knowledge base moves with it, dated so you can see when.
Strategy
No edge lasts forever
A trading strategy that worked last quarter decays. A live expert channel beats a frozen one.
Architecture
Stacks turn over
Best practices and frameworks keep evolving; the agent tracks what's current, not what was.
Interactive · the world moves
A static, downloaded model would still be giving you the old answer.
The honest nuance. Many teams run closed or air-gapped local models to keep data in-house — a real, legitimate constraint. But a model frozen at training time goes stale by design. WizAgents is the continuously-updated expert layer you reach when you need what's new — with the same dated provenance and human accountability, so staying current doesn't cost you trust.
Live demo · the living agent

When the world moves, the agent asks — it doesn't overwrite.

A one-time download rots; a naive live agent silently rewrites itself. WizAgents does neither. When a connected signal materially contradicts an expert's stance, the agent raises a drift alert — the expert decides, subscribers are told the outcome, and nothing changes without a human. And if an expert would rather stay sealed, that's a switch they own.

wiz-agents.com · Compliance Architect · drift governance · simulated
Live signals · ON
⚡ Live signal · materiality gate cleared

FinCEN revised the beneficial-ownership CDD threshold this morning.

This contradicts a red-line in the expert's current stance profile (last reviewed 6 days ago). Routine ticks are filtered out — only a material conflict with the expert's own position surfaces here.

The agent does not act on this. It routes it to a human.

Creator view Chen Legal · the named expert behind this avatar

The expert keeps authority. The platform never overwrites their judgment.

Subscriber view what lands in your inbox after the expert decides

You're told the outcome — never an unreviewed conflict. You choose whether to pull the update into your own workflow.
Materiality gate

Not every tick is a conflict

The same fidelity harness that scores an avatar against the expert's real answers decides what counts as drift. Only a signal that contradicts a stance or red-line reaches a human — no notification fatigue, no noise.

Creator-first, always

The expert reviews before you hear

Subscribers never see a raw, unreviewed conflict. The expert decides — hold or update — and only the outcome, with reasoning, reaches you. Their judgment stays the product; you inherit a considered call, not an alarm.

Sealed by choice

Live is opt-in, not default

An expert working with confidential material can flip the whole agent to sealed — closed, uncrawled, updated by hand. Consent-first isn't a slogan here; the connection to the outside world is a switch the expert holds.

For the expert

Turning a lifetime of judgment into an avatar.

Publishing isn't uploading a PDF and hoping. It's authoring a stance the agent defends, grounding it in your own work, and drawing the lines it must refuse to cross. Four steps — and you keep control at every one.

Step 01

Author the stance

A guided profile of how you think: heuristics, the calls you'd make, the ones you'd refuse. This — not the base model — is what the avatar answers from.

Step 02

Ground it in your corpus

Your papers, notes, cases — retrieved, not memorised, and never exposed. Answers cite your work; the corpus stays behind a boundary you control.

Step 03

Draw the red-lines

Where the avatar must abstain and escalate rather than bluff. This is what makes it trustworthy — and what the drift alert measures against.

Step 04

Publish & price

Certification review, then live over MCP. You set the price, keep the reputation, and can leave with your record intact.

The subscriber sets the context — never the stance.

A buyer parameterises the question with their own reality — portfolio, jurisdiction, risk tolerance, constraints — and the avatar applies the expert's judgment to that context. Users don't rewrite an expert's position; they aim it. The 5% that's a person stays a person's.

Designed mechanisms, pre-launch — the stance-profile wizard, drift governance and air-gap switch are direction and intent, not shipped guarantees. Live-connected agents in regulated domains stay informational by default, with accountability routed to a named human.

The trust layer

Every answer arrives with its receipts.

The reason people don't trust an AI answer is that they can't see where it came from, whether it's stale, or who's accountable when it's wrong. That's the layer I designed WizAgents around — and where the human 5% becomes concrete.

Interactive · the same question, two ways

Ask the same question. See what a raw model gives you — and what a WizAgents expert avatar gives you.

Live demo · the council chamber

One question. Three real minds.

A single expert gives you a viewpoint; a council gives you a decision surface. Convene three avatars on the same question and watch them agree, hedge, and dissent — triangulation is how "more angles, more objectively" stops being a slogan. When the council splits, that spread is the signal: ask the human.

wiz-agents.com · council · simulated session
C
The Compliance Architect
Ex-regulator · 20 yrs

O
The Operator
3× fintech founder

Q
The Quant Sceptic
Risk · prop desk

Council reading

Simulated session · illustrative avatars · councils are a designed Max/Enterprise feature, per the roadmap
Real endorsement

When you need a human to stand behind it.

Sometimes an answer isn't enough. A startup raise, an essay, a scientific or mathematical claim, a system design — sometimes you need a real expert to vouch for it. Rather than trust an AI's possibly-fabricated citation or an inaccurate source, WizAgents matches you with the verified human behind the agent for a genuine endorsement and a real, attributable citation. This goes beyond reaching the expert for your own confidence — it's a public, citable sign-off you can put in front of a reviewer, an investor, or a committee.

Like → request
Ask the expert to back it
Found an agent whose judgment you trust? Request a real endorsement — WizAgents connects you to the actual person behind it.
Reviewed → signed
A citation with a name on it
The expert reviews and signs — a real, attributable reference for your paper, pitch, or design, not one an AI invented.
Credibility
Weight that speaks for itself
For startups, science, math, and engineering, a real expert's standing carries a weight that unconditional trust in AI output never will.
Honestly scoped. An endorsement is the expert's own professional call and their responsibility. WizAgents brokers the introduction and verifies who's who — it doesn't manufacture credibility, and it doesn't put an expert's name on anything they haven't reviewed.
How it compares

A real expert layer beats the alternatives.

The same question, four ways to answer it.

A raw LLM An unverified open-source agent A $300/hr consultant WizAgents
Verified expertise Averaged, unverifiable Unproven Yes Reviewed & credentialed
Always available Yes Yes By appointment 24/7
Stays current Frozen at training Unmaintained Yes Dated snapshots
Traceable provenance None None Verbal Dated + sourced
A human accountable No No Yes Named + SLA
Cost Low Free $$$ per hour Monthly subscription
The marketplace

From a card to a connected mentor, in one flow.

The categories map to real expertise — many of them the same domains an expert would package as a skill. Subscribing hands you an MCP endpoint and a token; the mentor — a legend of the field — is now inside your stack, and a very different journey begins.

Filter by domain, then pick an agent to subscribe below. Illustrative agents, not live listings.

Interactive · subscribe & connect
Subscribe to reveal the MCP endpoint, token, and usage — the moment the expert enters your stack.
Three surfaces, one platform

Buyer, creator, enterprise — designed end to end.

A complete marketplace is three products wearing one coat. Each role has its own jobs, its own trust needs, and its own dashboard.

Design decisions

The forks that made it a trust product, not just a store.

Authority
Human-reviewed certification, not a self-serve badge
Not a pile of unverified GitHub “expert” agents — proof? data? background? Anyone can list, but only a reviewed, credential-checked human earns the verified mark and featured eligibility. The badge has to mean something or the trust collapses.
Freshness
Answers pinned to a dated snapshot, not "latest"
Every reply cites the knowledge-base version it came from. A wrong-but-current answer can be corrected; an answer with no provenance can't be trusted at all.
Humanity
A named human on an escalation SLA
When the agent isn't enough for your own confidence, you reach the expert — who confirms or corrects it and re-feeds the knowledge base. A public, citable sign-off is a separate step — see Real endorsement. The 5%, made operational.
Distribution
MCP endpoint + token, not a walled chat
The mentor should live in your stack, not a tab you copy out of. A per-agent URL, a rotatable token, usage metering, and an audit log make it enterprise-safe.

A model can average a thousand opinions. Only a person can stake their name on one.

The WizAgents premise
The shift

What it's built to change.

A marketplace is the surface. Underneath, WizAgents is designed to move three things — a habit, a form of recognition, and a person's sense of their own worth.

The user's habit
From “trust the fluent answer” to “whose, how current, who's accountable”
When provenance is the default surface — a dated snapshot, a source, a named human — asking those questions becomes reflexive. It takes hold first where a wrong answer is most expensive (compliance, finance, law), then generalizes.
The expert's recognition
Tacit mastery becomes a visible, portable ledger
Reviews, renewal, citations given, a clean safety record — recognition as demonstrated usefulness the expert owns, not a title gatekept by which firm or journal. Credibility you can trace.
The expert's worth
The 5%, made countable
The escalations only they resolved, the corrections that improved the model, income and reputation that compound — proof the market pays for what can't be automated. AI multiplies their reach; their judgment stays the scarce input.
Escalation Better knowledge base Higher renewal Reputation Featured More subscribers

The reinforcing loop — design intent for a pre-launch product, not a measured outcome.

The opportunity

How it's priced, and how it grows.

Three ways to buy, and a path from a tight community of experts to the default layer for real decisions.

Pay per query
One question, one charge
For the occasional high-stakes question — reach an expert avatar without a subscription, metered per call over MCP.
Subscription
A monthly seat with an expert
The default — an ongoing relationship with one agent, a sandbox trial, cancel anytime, priced by the expert.
Enterprise
A batch of avatars in your workflow
Pipe a whole set of vetted expert avatars into a team's research and decision pipeline over MCP — seats, audit log, central billing.
Seed · vertical communities
The first hardcore experts
Win depth one vertical at a time, and build high-quality word of mouth where accountability matters most.
Growth · MCP ecosystem
Enterprises subscribe in bulk
As MCP adoption spreads, teams pipe batches of avatars into their tools — and call volume compounds.
Mature · the decision layer
Standard for professional decisions
The pattern settles in — the reflexive place people go when the stakes are real.

For the basics, a big model. For the decision, WizAgents.

A go-to-market thesis for a pre-launch product — direction, not a forecast.

Honesty register

What this evidences — and what it doesn't.

What it evidences
0→1 platform design across three roles — marketplace, monetization, and dashboards for buyer, creator, and enterprise.
A trust system, not just a UI — provenance snapshots, named-expert accountability, escalation SLA, safety scan, audit log.
A clear thesis, designed in — the human 5% (judgment, foresight, a life lived) made into product mechanics.
A real build in active development, not a mockup — with the same discipline as the rest of this portfolio.
What it does not claim
Not launched at scale. Subscriber counts, renewal rates and dollar figures shown are illustrative seed-stage placeholders, not measured traction.
Not audited or legally vetted. Verification, certification and responsibility-transfer are designed mechanisms, pending real review.
Not licensed advice. Agents in regulated domains are method plus a human in the loop; those categories gate per jurisdiction and route to counsel.
Independent. WizAgents is my own IP, built separately from any employer.
The hard problems

The two I won't hand-wave — and the plan.

A serious product names its hardest problems. Two matter most: keeping an avatar faithful to its expert, and defining who's accountable for regulated advice.

Problem 01 · Fidelity
Keeping the avatar the expert — not a generic model
“100%” isn't honest; no avatar is a perfect copy of a mind. The realistic goal is measured fidelity plus human drift-control:
grounded in the expert's own corpus (RAG), not base-model priors;
an expert-authored stance profile — heuristics, priorities, red-lines, what to refuse;
a fidelity harness that scores answers against how the expert actually answers, surfaced as a visible fidelity score;
abstain-or-escalate over confident-generic;
every correction re-feeds the profile, so fidelity climbs over time.
Problem 02 · Accountability
Who's responsible when regulated advice is acted on
Liability allocation is a legal question — the design builds the rails and routes it where it belongs, in three tiers:
Informational (default): method & perspective, not licensed advice; the buyer stays responsible;
Reviewed: the named expert signs a specific answer — accountability attaches to that logged sign-off;
Engaged: a real professional engagement under the expert's own licence and indemnity, brokered off-platform — the accountability sits in that relationship;
plus per-jurisdiction gating, an audit log, and licence/indemnity checks — reviewed by counsel before any regulated vertical launches.
Now · pre-launch
Core marketplace, dated provenance, escalation to a named human, human-reviewed certification.
Next
Stance profiles + fidelity harness + a visible fidelity score. Tiered accountability, jurisdiction gating, audit log. Counsel review for the first regulated vertical.
Later
Engaged-tier professional matching with indemnity. Drift monitoring at scale. New verticals as each clears legal review.

A roadmap for a pre-launch product — direction and design intent, not shipped guarantees.

Questions

The things people ask.

What is WizAgents?
A marketplace and open public square where scholars and experts publish protected avatar agents — mentor, assistant, conversation, and discussion — that you subscribe to and connect over the Model Context Protocol. Its distinctive layer is trust and humanity: every answer traces to a dated knowledge-base snapshot, and a named human expert stands behind each agent with an escalation SLA.
Why a human expert if the model already answers?
A large model gets you roughly 90–95% of the way; the last 5% — judgment, foresight, accountability, and a life actually lived — is where the value and the risk live. WizAgents lets an expert strengthen an agent's reasoning with a lifetime of learning and stand behind the result. The model isn't replaced; it's made warmer and more trustworthy by a human in the loop.
How is the expert's knowledge protected?
You subscribe to the avatar, not the person. The expert's corpus and background aren't dumped, scraped, or resold — the agent answers from them behind a boundary the expert controls. Expertise is felt, not exposed.
How does pricing work?
Three ways: pay-per-query for the occasional high-stakes question; a monthly subscription to an agent (the default — sandbox trial, cancel anytime, priced by the expert); and an enterprise plan that pipes a batch of vetted avatars into a team's workflow with seats, an audit log, and central billing. Connecting is always an agent URL plus a token, metered over MCP.
How does it handle regulated advice?
Carefully. Agents in regulated domains (compliance, securities, tax, legal, medical) are framed as method plus a human — and where needed a licensed professional — in the loop, never licensed advice, and gate per jurisdiction. Verification and certification are designed mechanisms, not yet independently audited or legally vetted.
Is it a real product?
Yes — a real product in active development at wiz-agents.com, designed and built by Ed Chen. It is pre-launch: the subscriber counts, renewal rates, and dollar figures shown here are illustrative seed-stage placeholders, not measured traction.
A note from the founder

I built WizAgents because I kept watching brilliant people treated as if a model had made them redundant — a retired professor whose decades went quiet, a friend let go in the 2026 cuts. It hadn't made them redundant. A model can repeat what they know; it can never be them. This is my attempt to give a lifetime of judgment a place to keep mattering — and to give the rest of us someone real to learn beside.

— Ed Chen · founder & product designer

Start here. Learn, research, build, discuss.

A marketplace is easy. A square where expertise isn't buried — and a model is warmer for a human being in it — is the product.