Tokamak

Tokamak Fusion Reactor

Tokamak is Menlo’s internal software factory .

Aside: Real-Life Tokamak

A Tokamak is an experimental fusion reactor design that originated in the Soviet Union in the 1950s, that uses a magnetic field to provide a nearly limitless and clean source of energy.

Within Menlo, Tokamak is a metaphor for a Software Factory that provides a nearly limitless and clean source of “cognitive energy” to power through complex tasks. Menlo views Tokamak as our long-term competitive advantage, as a “Company of Agents” or “Datacenter of Geniuses” that accelerates Menlo’s vision, and gives us a competitive advantage.

What is Tokamak

Agent Orchestration

Tokamak orchestrates cloud coding agents, including Codex and Claude Code , to generate, modify, and iterate on code. Agents run in isolated cloud sandboxes, producing reviewable diffs and PRs as standard outputs.

Cloud-Based Execution Environments

Every task runs in a deterministic cloud environment. Tokamak spins up ephemeral sandboxes on demand, executes builds and tests, and captures complete artefacts for audit and review. This eliminates “works on my machine” failures and ensures consistent verification across every change.

Hardware-in-the-Loop Verification

For embedded and edge systems, Tokamak provides hardware-in-the-loop (HIL) testing on a custom hardware farm. Real physical hardware, provisioning, flashing, and exercising under test, validates firmware and system software in conditions that software-only testing cannot replicate.

Strategic Capability, Not a Product

Tokamak is a strategic capability we’re developing for Menlo:

Building Tokamak is closer to hiring than buying a tool. We’re adding capacity to the organization that:

Works asynchronously and never sleeps
Scales horizontally across any number of parallel tasks
Never forgets our conventions or best practices
Never burns out or switches contexts

This is fundamentally different from adding headcount. A human hire takes months to onboard, years to master our codebase, and eventually leaves. Tokamak is productive from day one and compounds over time.

Capturing Institutional Knowledge

Every successful pattern, every solved problem, gets codified into a playbook that Tokamak can execute again:

Playbooks: Our best debugging approaches, our testing strategies, all captured and reusable
Conventions: Code style, architectural patterns, applied consistently across every change
Process improvement: Each iteration captures what worked, so the next is faster

When an engineer figures out how to handle a class of task, they encode it into Tokamak. Now anyone can trigger that same workflow. Expertise gets encoded once and executed repeatedly.

Pulling Ahead

The gap between “using AI tools” and “building AI-native capability” is where real competitive advantage lives.

Menlo’s advantage isn’t just speed. It’s that Tokamak captures our institutional knowledge in executable form. Our best practices, our testing approaches, our debugging playbooks, all encoded into the factory itself.

This creates a compounding effect:

We ship faster because our processes are codified
We make fewer mistakes because verification is consistent
We iterate quicker because every lesson learned becomes permanent knowledge

The Factory Metaphor

Software factories combine generative (code creation) and verificative (testing and validation) capabilities in one integrated system. Tokamak embodies this vision:

Inputs: Intent, specifications, existing codebase
Process: Agent orchestration, cloud execution, hardware validation
Outputs: Verified code, test artefacts, telemetry, deployment packages

The factory metaphor also emphasizes reproducibility: every output can be traced back to inputs, and every verification is deterministic.