How Metropolis works

The founder leads. AI-native teams execute.

Metropolis is a venture studio operated by hierarchical teams of specialized AI agents. The founder sets strategy, allocates resources, and makes the consequential decisions; the system turns that direction into coordinated work.

Where this comes from

An evolution of the platform-VC thesis, for the AI generation.

a16z argued in 2011 that software would eat the world, then built an in-house platform of experts—talent, recruiting, marketing, business development—to give its portfolio an advantage capital alone could not. Both ideas change shape when that platform can be software. The personal computer amplified one person's ability to work; Metropolis aims to amplify a founder's ability to lead an institution.

Investment thesis → build thesis
Build the companies, don't pick them

A venture firm forms a view about where value will be created and funds the teams creating it. Metropolis commits its own resources and builds there instead. The cost of creating software has collapsed; the constraint is now getting it through an organization.

Eating the world → composing the company
The org chart is what software eats next

Software disrupted industries from the outside, but the disrupting company still ran on people, meetings, and handoffs. Governed agent workflows can now perform and coordinate the functions of the company itself.

Platform of experts → agents on a knowledge graph
The shared advantage compounds

A platform staffed by people scales by hiring, and its knowledge lives in individual heads. Implemented as agents over an explicit knowledge base, methods, decisions, and lessons accumulate through use. The machinery itself is not the moat—what compounds on top of it is.

What it is built from

A company's machinery, built once and inherited.

Two layers run on the same platform. An operating system that decides and executes work, and the business systems any company needs but none should rebuild from scratch.

Knowledge
A shared base that grounds every decision

Research, decisions, methods, and plans accumulate in a common base that specialists retrieve from before they act. New knowledge enters through a human-reviewed proposal—merge, split, or supersede—never silently, and never outside the boundary of the business that owns it.

Planning
Explicit workflows, not one-shot prompts

Work moves through config-driven workflow graphs with defined inputs, outputs, and review gates. One of the first reads the studio's own code and produces a grounded account of what each part does and how it connects—the studio explaining itself to itself.

Execution
A hierarchy of agents that delivers the work

Coordinating agents translate goals into bounded work, delegate to specialists, monitor progress, review results, and escalate when human judgment is required. A local fleet runs this today; a hosted layer with live session visibility is next.

Business systems
Customers, money, and measurement

Customer and lead records with traceable lifecycle and outreach history, where consequential outbound action needs human approval. Product events stored at volume, turned into canonical metrics, experiments, and signals. Payments and billing state. One canonical customer identity correlating all of them.

Model training
From a training table to a served model

A company supplies a tabular training set and declarative model configuration; the platform validates it, trains and honestly evaluates candidate models, versions the resulting artifact with its lineage, and serves predictions through a stable interface. Domain-specific feature engineering stays with the company.

Portfolio companies
The same machinery at a narrower scope

A portfolio company runs this system for one business instead of the whole studio—the same knowledge discipline, the same planning workflows, the same execution model, the same business systems—while keeping its own market, product, and private context.

Interchangeable parts

Start with the work—not the job title.

A conventional role is a historical bundle of responsibilities. Metropolis decomposes that bundle into explicit capabilities with their own knowledge, tools, permissions, evaluations, workflows, and authority boundaries, instead of asking one synthetic employee to imitate a title. Reusable machinery is leverage, not judgment—real markets still need people who know what outsiders miss, which is why Metropolis works with domain experts to build that knowledge into the system.

Understand
Research and analyze

Assemble evidence, build institutional context, identify uncertainty, and challenge assumptions.

Decide
Model and recommend

Prepare scenarios and tradeoffs, define authority boundaries, and return consequential choices for human judgment.

Operate
Execute, evaluate, and learn

Run bounded work, test its quality, surface exceptions, and feed evidence back into the institution.

Founder governance

Autonomy concentrates human attention where it matters most.

The founder sets objectives and strategy, chooses where to act, allocates capital and attention, defines the quality bar, and remains responsible for consequential decisions. The system executes between those decisions without making responsibility disappear.

Human gate one
Plan approval

Before consequential work begins, the founder reviews the direction, scope, dependencies, and tradeoffs—or asks for another iteration.

Human gate two
Delivery and decision approval

Material product changes, strategic decisions, and sensitive or irreversible actions return to the founder for judgment.

Where it is going

The studio is its own first customer.

Metropolis is not designing an operating model in the abstract and asking a future company to trust it—it is building the studio by using the studio to build itself. Engineering is the first working team: agents plan, build, review, and deliver software under founder governance. Product and planning follow, then the remaining business functions, each leaving behind reusable responsibilities, workflows, evaluations, and knowledge. The shared business systems are being proven against the first portfolio company now, and the same machinery goes to outside companies through consulting—the fastest way to find out where it holds up against businesses Metropolis did not build.

How consulting engagements work →
The destination

Reinvent the company—not just its tools.

One working team, expanding one governed capability at a time, into a company-creation institution whose organization, workflows, and knowledge improve through everything it builds.