Applied AI Lab for Expert Work Building Infrastructure for
AI-Native Professional Services
Plausible answers are not a professional standard. We are building specialised Expert AI and the auditable agentic infrastructure to perform multi-step expert tasks end to end: aligned with professional standards, verifiable by design, and able to escalate when human judgement is required.
For practice areas
- Legal & compliance
- Risk & regulatory
- Audit & assurance
- Accounting & tax
- Financial advisory
Why a lab
AI is advancing fast, and it can now produce answers that sound right. In professional services, sounding right is not the same as exercising professional diligence.
Expert work is held to professional and ethical standards, and those standards apply rigorously. Some of it requires human judgement that cannot be delegated. All of it has to stand up when someone asks how the conclusion was reached.
A system that is right most of the time, and cannot say when it isn’t, is not ready for that work. We are on a mission to close this gap: delegating to AI the work that can be delegated, and escalating the work that requires human judgement and supervision.
Our mission
To make expert work executable by AI — without compromising the standards that make it professional.
Where this is going
What changes when professional services are AI-native
Most firms have added AI to the way they already work. We deliver the larger change by executing the work using AI, and delegating to humans only the essential steps.
- The first pass
- TodayA professional, with an AI assistant open in another windowAI-nativeThe system, working to the firm’s own standard
- Checking the work
- TodaySomeone rereads everything the tool producedAI-nativeVerified against the standard, with the working shown
- Expert time
- TodaySpent on process, drafting and reviewAI-nativeSpent on the judgement only an expert can exercise
- What the client receives
- TodayA deliverableAI-nativeA deliverable, and the record of how it was reached
Research & development
Three pillars of our research
Each answers part of the same question: what would it take for AI to do expert work that a professional would put their name to?
01
Expert AI
Models that know the standard, not only the subject.
We train models capable of performing work that requires specialised expert knowledge and reasoning, and develop new methods for controlling the behaviour of LLMs in the delivery of professional services, so that what they produce complies with the standards the work is held to.
02
Agentic Delivery
Controlled environment for specialised agents to execute professional work end to end.
We develop systems that coordinate specialised agents across complete professional workflows while preserving context, permissions and accountability from instruction to final work product.
03
Verification
Human–AI collaboration built for regulated environments.
We work on novel methods of human–AI collaboration in regulated environments: a transparent escalation path for every decision a system should not take alone, and a complete record of who decided what, on what basis.
Research directions
What we are working on
The problems our work across all three pillars is focused on.
- 01
Context integrity and long-horizon memory
Architectures that preserve, retrieve and correctly apply material information across complex workflows — from initial instruction to final work product.
- 02
Interpretable and verifiable reasoning
Methods that make the basis of AI-generated conclusions traceable to evidence, rules and professional standards — exposing unsupported inferences, reasoning shortcuts and specification gaming.
- 03
Governed human–agent systems
Protocols for coordinating specialised agents and professionals within permissioned, observable environments, with explicit delegation, escalation and accountability at every stage.
- 04
Evaluation science for expert work
Domain-grounded benchmarks that measure completeness, reasoning validity, procedural compliance, calibration and outcome quality — not linguistic plausibility alone.
Design partners
Deliver more of your best work, to your own standard.
We’re looking for firms in regulated, knowledge-intensive practice areas to work with on real workflows. Design partners shape what the lab builds, and how it is evaluated.
We especially want to hear from teams in
- Legal & compliance
- Risk & regulatory
- Audit & assurance
- Accounting & tax
- Financial advisory
- 01
Scale without adding headcount
Start with one workflow your experts repeat. We make it deliverable by AI to your own standard, so capacity grows while your people focus on the judgement only they can exercise.
- 02
Evaluation built for your domain
We build evaluations specific to your practice area, measuring work against your standards and defining where judgement has to stay human.
- 03
Early access to research previews
Partners get access to new capabilities first, and shape what gets built next.
Already using the research preview? Sign in
Questions
What people ask us first
A research and development lab working towards professional services that are designed around AI from the start. Our focus is expert work specifically: work that requires specialised knowledge, is held to professional and ethical standards, and sometimes needs human judgement. “Applied” means the research is built and tested against real professional workflows, not benchmarks alone.