The Playground

The Playground

Half of this studio builds systems for clients. The other half breaks things on purpose to find out what actually works.

The Playground is our research arm. We run experiments on the frontier of AI and the web, and the findings flow straight into the systems we ship.

Bring us the bottleneck
Stick figure researcher at a workbench, blue threads radiating to a gear, three linked dots, a wireframe cube and an eval chart on one side and a brain in a head profile, a small figure sketch and a ring of people on the other

Client work demands answers nobody has written down yet. How many agents should share one task before coordination costs eat the gains? What makes an autonomous system safe to leave running overnight? How heavy can a 3D scene get before phones give up? Vendor blogs will not tell you. Experiments will.

So we run them. Every technique we recommend to a client has been tested here first, on our own systems and our own dime. The Playground is why our advice comes with evidence attached.

What We Are Researching

Seven threads, all active, all feeding production work. Four sit on the machines. Three sit on the people who have to live with them, because a system nobody trusts or adapts to is a system that fails quietly.

Agentic AI
What it actually takes for an AI system to finish real work without supervision. We test task completion, escalation design, and failure recovery before recommending any autonomous build.
Multi-agent systems
When teams of agents outperform a single one, and when the coordination overhead eats the gains. We benchmark both before scoping client systems.
3D and WebGL web experiences
Scenes that stop scrolls without destroying mobile frame rates. We measure polygon budgets, shader cost, and load time on real devices, not spec sheets.
Harness engineering
The verification scaffolding that makes AI agents reliable enough to trust. Static checks, integration tests, and end-to-end probes run before anything reaches a client build.
See the service
The psychology of AI
How people actually trust, distrust, and anthropomorphize AI, and how that shapes whether a system gets used. We study where confidence is earned and where it is dangerously misplaced.
Human adaptation & the future of work
What changes for a team when the routine work leaves. We track how roles reshape around AI, which skills gain value, and how people stay in the loop on the decisions that still need them.
AI's impact on work, culture & society
The second-order effects past any single tool: how automation reshapes incentives, trust, and the way organizations make decisions. We watch this so the systems we build account for it.

We Publish What We Learn

Research that stays private rots. We write up our experiments, including the failures, and publish them at /research/ as Field Notes and Reports. Field Notes are short and applied. Reports are formal write-ups of Playground work.

If you are evaluating us, read a few. They show how we think before you pay us to think about your business.

Why a Studio Runs a Playground

AI moves too fast to rely on what worked last year. The systems we build for clients are only as good as our current understanding of the tools, and reading about them is not understanding. Running them is.

The loop is deliberate: research in the Playground, proof on our own systems, then application to client builds. No dreaming. Just building, and the testing that makes the building sound.

From the Playground

Research

Field Notes and Reports from the Playground, published as we learn.

Read more →

Harness Engineering

How we scaffold AI agents so their output can be verified, not just admired.

Read more →

AI Agent Evaluation

Measuring whether an agent actually did the job, with benchmarks that resist wishful grading.

Read more →
Chalk stick figure in a hard hat presenting a little machine of blue gears it just built

Bring us the bottleneck.
We’ll build the system.

No Dreaming. Just Building.

Work With Us Back to Home