02 / Labs

The lab is where opinions become evidence.

An OpenClaw presentation showing a live agent-driven maintenance tool
OpenClaw in the wild · systems made visible

A working shelf of agent architectures, local inference experiments, and creation tools. These are snapshots of learning in motion, not polished inevitabilities.

Lobster Agents / Published February 2026

A software organization made of 28 specialized agents.

Eight departments. Five model types. Cloud and local inference. Role-specific routing with fallback chains designed to keep the system operational.

DEPARTMENT 011 OF 8

Executive leadership

Sets direction, weighs tradeoffs, and keeps the system pointed at an outcome rather than a task list.

What should the organization do next, and why?
FEB 2026 MODEL MIX28 AGENTS
Strategic reasoningOpus
4
Tactical coordinationSonnet
5
Engineering executionCodex
7
Creative + specialistChatGPT
4
Local executionGPT-OSS 120B
8

A dated snapshot, not a permanent recommendation. Models, costs, and assignments keep changing.

The architecture is a dated case study. Marc explicitly expects assignments to evolve as models, economics, and evidence change.

Read the architecture ↗

Builder exchange

Compare architectures, not just answers.

The fastest way to find a weak assumption is to put it next to a very different system. Marc’s lab work stays connected to other builders exploring local models, new interfaces, and agent-native products.

Marc Mojica with Ben Guo of Zo Computer
Marc Mojica with Ben Guo · Zo Computer

Clawible

Build the site. Keep the choices.

A local-first AI website builder designed around control: choose the stack, database, cloud, and model that fit the work.

Clawible is actively evolving. This site does not characterize its release or licensing status.