Both live in Slack.
Only one tells you how it learns.
Viktor is an AI coworker that does everything. Mavrick is an AI employee with one job — ship marketing work. Different category. Different game. Different math. We built Mavrick on a published, self-improving architecture and a contractual privacy posture. Viktor took a different road.
This page compares the two honestly. Where Viktor wins, we say so. Where Mavrick wins, we link to proof.
Viktor is a horizontal AI coworker. It executes code, runs persistent compute per workspace, automates browsers, deploys apps, and maintains workspace state across multi-week timelines. The product surface is wide and the use cases are general — eng, ops, support, marketing, finance.
If your team needs an AI coworker for everything, Viktor is in the category.
Mavrick is not in that category.
Mavrick is a vertical AI employee built for one job: ship marketing work. Narrowness and depth over horizontal coverage. Transparency over proprietary claims. A Privacy Charter you can reference in your MSA. A Learning Log that publishes every behavior the agent has acquired in production.
The rest of this page explains the bet — and where it doesn't apply.
The chart that matters in this category isn't features. It's velocity.
Mavrick is in week one of public learning. The first 30-day window closes June 1, 2026. Refresh this page weekly to watch the curve.
Last 30 days: 2 learnings shipped
Last 30 days: first one coming soon
Regression pass rate: 0.0%
Two categories. Two products built for them.
What matters most to you?
1.SOC 2 Type 1: Mavrick began continuous internal compliance logging on April 15, 2026. External auditor fieldwork scheduled to begin mid-2026, target report issuance Q3 2026. Privacy Charter contractually binds us to the controls SOC 2 audits — we operate to the standard before the certificate.
2.SOC 2 Type 2: Six-month observation period planned to begin immediately following Type 1 issuance. Fieldwork April–May 2027, target report issuance Q2 2027.
3.🟡 v1.x means committed to the v1 minor release cycle. "🟡 Phase 2" and "🟡 Phase 3" reference dated phases in the public roadmap. Vague "coming soon" entries are not used on this page.
4.3,200+ integrations — Mavrick uses a managed connector layer for integration breadth and ships native depth in CMO-vertical bundles. Viktor built theirs in-house. Both reach 3,000+ services. Different trade-offs; both valid.
5.Approval gate is architectural, not optional — Mavrick has no "always-approve" preference, by design. Procurement teams should be skeptical of any AI agent that lets its operator disable safety mid-task.
6.Code execution and persistent compute — Mavrick offers these at the Enterprise tier, where the per-tenant infra cost is justified by enterprise pricing. Below that tier, we own narrowness deliberately. Viktor includes these capabilities at every tier — that's a design choice with cost implications they're carrying on the buyer's behalf.
LAST REVISED: May 2, 2026
What Mavrick does that no other AI employee makes.
The model self-reports its own gaps mid-task. Not after.
When Mavrick can't fully satisfy a request, it calls a built-in tool that logs the gap with full context — before the user ever sees a no.
Refusals get one in-turn retry before a refusal lands.
A flagged refusal triggers a single retry with a meta-prompt that re-examines the request against available tools. Recovers an estimated 30–50% of capability-gap refusals before the user sees them.
Every learning is regression-tested in the originating workspace.
Not in a sterile test environment. In the workspace the gap came from, against real data, with real tools. Two failures in a row → auto-rollback.
Every fix is followed by a DM back to the person who asked.
The user who hit the gap last week gets a notification when the fix ships. Not generic. Specific to their original question.
We borrowed the right ideas (Hermes Agent for episodic memory, GEPA for trace analysis), adapted them for multi-tenant SaaS, added the four moves above, and published the whole thing.
Both products say they protect your data.
Only one says how. In writing. Versioned.
Standard SaaS — encryption in transit and at rest, no model training on customer data, SOC 2 Type 1. All real, all table stakes. The mechanism is not publicly disclosed.
- ⊢No background channel reads — published.
- ⊢Participated-only persistence — published.
- ⊢Approval gate is architectural, not optional — published.
- ⊢Credentials never reach the model — published.
These are commitments, not posture. They're versioned. Every change is a public diff with a date.
For procurement teams: Mavrick's Privacy Charter is the document you reference in your MSA. Viktor's is the policy page you trust the vendor on. That's a meaningful difference at enterprise scale.
Same entry price. Different math on what fifty bucks buys you.
Both products charge $50/month for the entry paid tier. Both include 20K credits. Both cover unlimited workspace seats. Pricing is parity.
An AI coworker that touches every function across your team. Code execution, persistent compute, app deploys, browser automation included. Horizontal coverage at scale.
An AI employee with one job — ship marketing work. Architecture published. System prompt published. Privacy charter contractually binding. Self-improvement public. Vertical depth in paid ads, outreach, attribution, content, and growth ops.
Same entry price. Different jobs. Different bets. Different math on what fifty bucks buys you.
Switching is direct.
Mavrick installs in 30 seconds and runs in parallel — you don't have to remove Viktor to evaluate. The free tier (10 missions, no card) is designed to give you enough surface to test Mavrick on real work before making any change.
If you switch:
- 1.Add Mavrick to the channels Viktor was already in.
- 2.Connect your accounts (about 30 seconds per integration).
- 3.Run the same missions you'd run with Viktor — paid ad pacing, attribution checks, briefings, anomaly alerts.
- 4.Compare the work product side by side.
- 5.Decide.
We're not building a Viktor migration tool. We're building a product that makes the choice obvious for marketing teams. If after a week of side-by-side use Viktor is the right call for your team, that's a real answer. We respect it.
We published our architecture. Our system prompt. Our Learning Log. Our decline log. Our privacy charter.
Viktor: do the same.
If we're wrong about how yours works — prove it.
[ Read the open letter → ]10 free missions. No card. No catch.
Ten missions is roughly a week of hard usage with multiple tools connected. By mission three you'll know if Mavrick belongs in your workspace. Decide from there.
Add Mavrick to Slack