Exploration
Ideas get followed because they are worth understanding.
Research is the center of this company. The exploration is open-ended on purpose: our research practice investigates what agentic systems can be, close to real problems, unhurried about the fundamentals, and honest about what is actually true. The engineering is where the ideas prove out, and what the lab learns flows into the consulting practice, never the other way around.
The active program
Boundary Engineering
One development program leads the work: Boundary Engineering, the capability-level method for deciding where model judgment belongs and what authority it has earned. The program builds the decision science, the reference controls, the evidence systems, and the working artifacts that make bounded autonomy operational.
The technical base
Seven research areas keep the method honest.
Boundary Engineering draws on a wide technical base: how agentic systems are built, extended, coordinated, given state, measured, and used to write software. We maintain an informed view across all of it, and concentrate development where the questions of authority and evidence are decided.
Harness Engineering
An agent is a model plus a harness, and harness choices can materially change the outcome. We study how scaffolding changes capability, evidence, authority, stop conditions, and recovery.
Agent Extensibility & Tooling
Agents are only as capable as the skills, tools, and delegation structures around them, and published capabilities often weaken under measurement. We study how extensions change what a capability can access, what authority it receives, how it should be evaluated, and which limits the runtime must enforce.
Orchestration & Durable Execution
An agentic workflow that cannot recover from a crash, deploy, or interrupted approval is not ready for consequential authority. We study durable state, safe side effects, resumability, compensation, and containment as reference control and recovery patterns.
Context, Memory & Knowledge
Agent behavior depends on the state it receives: context can bloat, memories can go stale, and retrieval can return the plausible instead of the true. We study how provenance, retention, retrieval, and context limits affect a capability’s evidence and permitted authority.
Evals, Observability & AgentOps
The hardest question in agentic AI is not “can we build it” but “did it work.” Outcome verification is one of four active Boundary Engineering workstreams: representative evaluation populations, evaluator validity, adversarial cases, operational signals, and rollback thresholds.
Agentic-Driven Development
AI can write code; the open question is what evidence justifies trusting the result. We study verification-first development loops, review capacity, and the authority coding agents should receive in consequential repositories.
Frontier Research
Every agentic system embeds a bet about where its logic should live: in prompts the model interprets, or in code the runtime enforces. We investigate that boundary because it directly affects suitability, authority, deterministic controls, and verification.
These are supporting research areas inside Boundary Engineering, not separate programs, products, or service lines. Explorations here become public claims only with independent evidence and an explicit graduation decision.
Evidence, in the open
We are building our own operations on the same class of systems we engineer, beginning with explicit Boundary Profiles and Capability Ledgers for consequential capabilities. As that work produces evidence, we will publish how the systems are designed, evaluated, and limited, because the strongest claim we can make is one that can be inspected.
Field notes
Published explorations will appear here with their experiment identifiers, what we tried, what we found, and whether it is worth pursuing. Nothing is published yet. We would rather show you an empty page than a staged one.