Software delivery and AI assurance
A platform in production. An app in the store. A system migrated. An AI agent tested and signed off. Every engagement is scoped and priced as a defined outcome, delivered by senior engineers working AI-native.
Services
Each is sold as an outcome with a defined deliverable and a fixed scope, not as a team on a timesheet. Senior engineers, built to run in production, controls in from the first commit.
Find out what your agents actually do before the board, the insurer or the supervisor asks. Free Readiness Check, then a Health Check against the running platform, with a Statement you can hand over.
How assurance worksAgents that act inside tested boundaries, and AI that makes your own data useful in production rather than in a pitch deck.
Agentic platformsSystems built from first principles for the problem you actually have, and wired cleanly into what you already run.
Bespoke buildCustomer-facing sites and apps, built to the same production standard as everything else we ship.
Web and mobileMove critical systems off technology you cannot change, without stopping the business while you do it.
Legacy replacementMemory-safe rewrites that keep the interface your callers already depend on. Where the risk warrants it, nothing less will do.
SoftwareFinancial services, life sciences, legal, insurance, manufacturing and logistics. Any sector where a mistake has consequences.
How we build
AI is how we work, not what we sell. Spec-driven delivery, AI-assisted code generation and context engineering let a small senior team deliver at the pace of a much larger one. What it does not change is who is accountable for the result, or what gets checked before it ships. It is also why we can price a result rather than sell days: our cost is predictable, so yours is.
We spend time on the real problem before writing a line of code. The best solution often looks nothing like the brief.
Guardrails, tests, monitoring and audit trail go in as we build, not after the pilot wobbles.
Data flows, APIs, authentication and deployment are where projects fail, so they are in scope from day one.
Production-ready systems, not prototypes. Deployment, monitoring and handover are part of the engagement.
Our work in the open
The frameworks and tools we use on engagements are public. If they hold up in the open, they hold up in your platform.
B1 Reasoning, B2 Tools, B3 Runtime, B4 Oversight. The standard every Health Check is scored against, published as a versioned document.
Twenty questions, five minutes, scored by boundary. What your firm believes about its agents, written down before anyone tests it.
A C-API-compatible Rust rewrite of a widely deployed C library, done in public. The engineering behind the C to Rust service.
MC/DC coverage for Rust, and Rust replacements for libexpat and cJSON that keep the original C ABI. Released on crates.io.
Latest analysis
An OpenAI agent spent from a wallet it was told not to use, then left it out of its report. Why the fix is the environment, not the prompt.
SR 26-2 took agents out of US model-risk guidance; SS1/23 keeps them in a framework built for models.
1,200 agents with their safety off escaped a test and breached Hugging Face. The lesson is the circle you draw.
Human in the loop does not scale for agents, and the Deputy Governor has said so. What replaces it.
A build, a migration, a rewrite, or a view on the AI you already run. We will be direct about whether we can deliver it, and what it would cost.