Marvik and WorkingAgents occupy adjacent positions in the enterprise AI stack. Marvik is an AI consulting firm that builds and deploys production AI systems for enterprises. WorkingAgents is a governance platform that controls what AI agents are allowed to do once deployed. One builds the agents. The other governs them. The overlap is narrow. The partnership surface is wide.
Marvik: The AI Implementation Partner
Marvik is a Montevideo-based machine learning consulting firm founded in 2016. They were acquired by Moove It in 2020, which merged with December Labs in 2023 to form Qubika — though Marvik continues operating under its own brand. Their numbers: 200+ AI specialists, 300+ production AI projects, clients including Stanford University, Rappi, Mercado Libre, Procter & Gamble, and UNICEF.
They are not a SaaS product company. They sell services — end-to-end AI project delivery from strategy through production deployment.
What Marvik Does
Four service lines:
-
Find Your AI Opportunities — Strategy consulting to identify high-value AI implementation areas in an enterprise.
-
From Data to Production-Ready AI — End-to-end solution development: data engineering, model training, annotation, full-stack development, deployment.
-
On-Demand AI Talent — Staff augmentation with embedded senior AI specialists ($100–149/hr, $25K+ minimum project size).
-
AI Leadership — Fractional Chief AI Officer (CAIO) advisory for enterprises that need executive-level AI governance guidance.
Technical Capabilities
Marvik’s expertise spans the full AI engineering stack:
| Domain | Capabilities |
|---|---|
| AI Agents & LLMs | Multi-agent orchestration (LangGraph), domain-tuned models, agentic workflows |
| Generative AI | LLM applications, text-to-SQL, conversational AI |
| Computer Vision | Object detection, image classification, gesture recognition |
| Predictive Analytics | Demand forecasting, fraud detection, churn prediction, credit scoring |
| Data Engineering | Pipelines (Airflow, Spark, dbt), warehousing (Snowflake, BigQuery), real-time analytics |
| Robotics | Physical AI, NVIDIA Isaac Sim (presenting at GTC 2026) |
Their agent framework stack includes LangGraph, LiteLLM (100+ LLM unified interface), Pipecat (real-time voice agents), Vertex AI, and both MCP and A2A protocols.
Marvik’s Multi-Agent Architecture
Marvik calls their approach “The Agentic Orchestra”:
- Orchestrator agents manage conversation flows and delegate to specialized agents
- Specialized agents organized as modular subgraphs — independently scalable and reusable
- Single-step orchestration for customer-facing speed, multi-step for back-office complexity
- State management via LangGraph with structured state variables
- Standardized BaseNode class for LLM calls, JSON parsing, and schema validation
- Provider-agnostic via LangChain chat models
- Production infrastructure: Redis for prototypes, DynamoDB or PostgreSQL for production
MCP Knowledge
Marvik has published detailed technical content on MCP — architecture, transport mechanisms, framework support, and implementation with OpenAI Agents SDK and FastMCP. They’ve built MCP-powered agents (including a financial agent demonstration) and written about remote MCP servers on AWS and Azure.
This is implementation expertise, not a product. They know how to build MCP integrations for clients — but they don’t sell an MCP server.
Notable Deployments
| Client | What Marvik Built |
|---|---|
| Stanford University | Patient emergency prediction using 300M+ medical records |
| Rappi | AI-powered search, agentic AI for merchant growth, CV for delivery optimization |
| Mercado Libre | AI for product listing quality at scale |
| Celistics | Customer service transformation with AI agents |
| Microsoft AI Lab | Rapid ML prototyping (200 prototypes, 6-10 Marvik engineers) |
Partnerships
NVIDIA (first LATAM company at GTC Inception pavilion, GTC 2026 sponsor), AWS, Google Cloud, Microsoft, Oracle, ElevenLabs. Program memberships: Google For Startups, NVIDIA Inception, AWS Activate.
Where They Compete
AI agent implementation
If an enterprise needs AI agents built, both companies can participate in the conversation. Marvik builds them from scratch as a consulting engagement. WorkingAgents provides the governance infrastructure that agents run through. For a client evaluating “who helps us deploy AI agents,” both names could appear on the shortlist — but solving different parts of the problem.
Fractional CAIO vs. governance platform
Marvik’s CAIO service provides executive-level AI governance advisory. WorkingAgents provides the technical enforcement layer for that governance — permissions, guardrails, audit trails. An enterprise could hire Marvik’s CAIO to define the governance strategy and deploy WorkingAgents to enforce it. Or they could view Marvik’s advisory as competing with WorkingAgents’ platform approach to the same governance problem — policy-as-advice vs. policy-as-code.
MCP expertise
Both organizations understand MCP deeply. Marvik implements MCP integrations for clients. WorkingAgents is an MCP server with 86+ governed tools. If a client asks “who can help us with MCP,” both are credible answers — but at different layers.
Where They Don’t Compete
Services vs. product
This is the fundamental distinction. Marvik sells engineering hours and expertise. WorkingAgents sells a platform. Marvik builds custom AI solutions from scratch for each client. WorkingAgents provides a reusable governance layer that sits between any agent and any system.
Marvik’s 200+ AI specialists build, train, and deploy models. WorkingAgents doesn’t build models — it governs them after deployment.
No governance product from Marvik
Marvik’s public materials describe message validation and state safety within agent workflows, but there is no proprietary access control system, no per-tool permission model, no structured audit trail platform, and no guardrail engine. Their governance is advisory (CAIO consulting), not a technical enforcement layer.
WorkingAgents’ entire value proposition — capability-based permissions, per-agent identity, three-checkpoint guardrails, cross-system audit trails, Virtual MCP Servers — has no equivalent in Marvik’s offering.
No AI engineering from WorkingAgents
WorkingAgents doesn’t build custom ML models, train computer vision systems, design predictive analytics pipelines, or provide data engineering services. WorkingAgents provides the governance infrastructure. Someone else builds the agents that run through it.
The Partnership Case
The competitive overlap is minimal. The partnership surface is significant.
1. Marvik Builds, WorkingAgents Governs
Marvik’s 300+ production AI projects need governance once deployed. Every agent system Marvik builds for an enterprise — customer service agents, predictive analytics pipelines, multi-agent orchestration systems — needs access control, audit trails, and guardrails in production.
Today, Marvik builds governance into each project as custom code. With WorkingAgents, they could plug into a governance layer that’s already built:
Marvik builds the agent system
→ Agents connect through WorkingAgents' MCP Gateway
→ Per-agent permissions enforced automatically
→ Guardrails on every tool call
→ Audit trail for every action
→ No custom governance code per projectThis reduces Marvik’s delivery timeline (governance is infrastructure, not custom build) and increases the client’s confidence (battle-tested governance vs. bespoke implementation).
2. The Agentic Orchestra Needs a Conductor’s Rules
Marvik’s multi-agent architecture — orchestrator agents delegating to specialized agents — is exactly the pattern where governance matters most. When an orchestrator agent decides which specialized agent handles a task, someone needs to enforce:
- Which specialized agents the orchestrator can delegate to
- What data each agent can access
- What actions each agent can take
- What happens when an agent tries to exceed its scope
- Who approved high-risk actions
WorkingAgents’ Virtual MCP Servers map directly to this: define permission boundaries per agent role, enforce them at the gateway, log everything. The orchestrator sees only the tools it’s authorized to delegate. Each specialized agent sees only the tools within its domain.
3. MCP Integration Accelerator
Marvik already implements MCP integrations for clients — they’ve built MCP-powered financial agents and documented AWS/Azure MCP server deployments. WorkingAgents is a production MCP server with 86+ governed tools.
Marvik could use WorkingAgents’ MCP Gateway as a pre-built integration layer for client projects instead of building custom MCP servers per engagement. The client gets governed tool access on day one. Marvik focuses engineering time on the agent intelligence layer, not the plumbing.
4. Regulated Industry Deployments
Marvik serves healthcare (Stanford — 300M+ medical records), fintech (credit scoring, fraud detection), and government (UNICEF). These are regulated environments where every agent action faces scrutiny.
Marvik builds to GDPR, HIPAA, and ISO 27001 compliance frameworks — but as a consulting capability, not as platform certifications they hold. WorkingAgents is designed for SOC 2, HIPAA, GDPR, and FedRAMP compliance with built-in audit trails, PII detection across 20+ categories, and encryption.
Joint deployments in regulated industries: Marvik builds the domain-specific AI (patient risk prediction, fraud detection, document processing), WorkingAgents provides the compliance-ready governance layer (audit trails, permission enforcement, PII redaction). The client gets domain expertise and governance infrastructure without building either from scratch.
5. Fractional CAIO + Governance Platform
Marvik’s CAIO advisory defines AI governance strategy for enterprises. WorkingAgents provides the technical enforcement. This is policy and enforcement — the CAIO says “engineering agents should not access financial data,” and WorkingAgents’ permission model makes it technically impossible.
The CAIO engagement becomes more valuable when it maps directly to a platform that enforces the policies. Marvik’s advisory produces governance requirements. WorkingAgents implements them as configuration.
6. LATAM and Global Distribution
Marvik operates across LATAM, Europe, and the US with strong presence in Uruguay and Brazil. WorkingAgents targets enterprises globally. Marvik’s 300+ client engagements represent a distribution channel — every production AI project they deliver is a potential WorkingAgents deployment.
Marvik engineers recommending WorkingAgents as the governance layer for their agent architectures creates a natural referral path that doesn’t compete with Marvik’s services.
Head-to-Head Comparison
| Dimension | Marvik | WorkingAgents |
|---|---|---|
| Business model | Consulting / services | Platform / product |
| What they sell | AI engineering hours | Governance infrastructure |
| Builds AI agents | Yes — custom per client | No — governs them |
| Governs AI agents | Advisory (CAIO) | Technical enforcement |
| MCP role | Implements MCP per project | Production MCP server (86+ tools) |
| Access control | Custom per engagement | Capability-based, per-agent, per-tool |
| Audit trails | Custom per engagement | Built-in, structured, cross-system |
| Guardrails | Message validation, state safety | Pre/during/post execution, PII, injection |
| Compliance | Builds to frameworks | Designed for SOC 2, HIPAA, FedRAMP |
| Team | 200+ AI specialists | Platform engineering team |
| Revenue model | Project fees, hourly rates | Platform licensing |
| Client relationship | Embedded engineering partner | Infrastructure provider |
Strategic Assessment
Marvik and WorkingAgents are not competitors. They’re different layers of the same stack.
Marvik is the AI engineering layer — they build the intelligent systems. WorkingAgents is the AI governance layer — it controls what those systems are allowed to do. An enterprise deploying AI agents in production needs both: someone to build the agents and something to govern them.
The most natural partnership model: Marvik recommends and integrates WorkingAgents as the governance layer in their client deployments. Marvik’s engineering expertise builds the agent intelligence. WorkingAgents’ platform provides the permissions, guardrails, and audit trails. The client gets production-ready AI that’s also production-governed.
Every new Marvik client engagement that deploys agents in production is a potential WorkingAgents deployment. Every WorkingAgents customer that needs custom agent development is a potential Marvik referral. The relationship is symbiotic, not competitive.
WorkingAgents is an AI governance platform specializing in agent access control, orchestration, and security for enterprises deploying AI at scale.