AI for the workyou cannot get wrong.
Ten years deploying AI where a wrong answer is a compliance event. Every job runs probation under a named reviewer, and every action lands on a record you can hand to a regulator.
- 10 years in regulated industries
- Patent authors
- University of Toronto lecturers
- $100k Lovable Shipped 2025 grand prize
What AI may touch, and who is watching it.
Every job gets an address on this grid before it runs: the row sets what it can reach, the column how deeply it’s built. Nothing moves up a row without earning it.
Five levels of trust
A capability starts at the lowest tier that does the job. Each step up reaches deeper into your systems and carries heavier supervision by default.
- T0
Public data
Public data in, insight out. No internal access
Supervision · Job spec + output log - T1
Reads internal
Reads internal data, writes nothing
Supervision · Probation review + access scoping - T2
Drafts for humans
Drafts things humans review and send
Supervision · 100% human approval until graduated - T3
Writes, quarantined
Writes into systems behind a review queue
Supervision · Confidence scoring + permanent quarantine + audit trail - T4
Operates systems
Drives software the way staff do (agentic)
Supervision · Pilot only. Longest probation. Never promised
Tier 4 is AI that operates software the way a person does. We do that work, and we refuse to sell it as finished: it runs as a pilot behind the longest probation we have.
Three ways to build the same job
The row is about risk. The column is about effort: days of configuration, weeks of assembly, or months of engineering for the proven, high-volume work.
Configured
Skill + template on certified equipment in your tenant
Days of work · Scoped to your volumesAssembled
Scheduled workflows, connected tools, dashboards
Weeks of work · Scoped to your volumesEngineered
Purpose-built system with its own review UI and integrations
Months of work · Scoped, with a run retainerWhere the library sits today
Every deployable job from the catalog, plotted on the grid above. Circles are sized by the scale of the work they take on — click one for what it does and how it is governed.
Circles are colored by the domain they serve. The higher one sits, the more of your systems it touches, and the more supervision it ships with by default.
- Insurance broking
- Legal practice
- Healthcare administration
- Distribution and supply chain
- Retail and ecommerce
- Industrial and engineering
Insurance, legal and healthcare, workflow by workflow.
Every row runs the same shape as anything else we build: a trigger, a read, a confidence check, and a named reviewer who signs off before anything writes back.
Predictable before it runs. Provable after.
A general agent improvises and hands you a transcript. You can’t put that in front of a regulated process. Everything we build is mapped: predictable before it runs, provable after.
Human in the loop, always
Every job works probation like a new starter: a named reviewer signs off on everything until measured accuracy on your real samples earns more autonomy. Review never fully ends.
Cited to your own rules
It never invents a figure. Every number cites your own versioned rules: rate tables, appetite, policy wordings. Change a rule and the citation moves with it.
A tamper-evident audit trail
Every action, human or AI, lands on a hash-chained ledger you can export. A record you can hand to a regulator, not a transcript you have to trust.
Book a discovery call.Leave with a plan you can take to compliance.
Start with the map of your organization, or with the one job that hurts. Measured in hours and money, and everything we build stays yours.