Every company now has the same models. Almost none have the context.
We help businesses in Australia and the US put AI to work. We find where it would actually pay, write the strategy and build the systems, starting with the context layer that lets a model see your business at all.
Three engagements delivered so far: two in business transformation, one on the data layer.
- AI Review
- 2–3 wks
- Strategy
- 4–6 wks
- Build
- Staged
“Which jobs are at risk of slipping this month, and what is it worth?”
- 01Asked byOperations Manager, Delivery
- 03ReadERP schedule · contract CT-2291 · AR ageing
- 05Fact3 jobs behind schedule · $412k combined · 1 carries damages
- 08ConflictJob 4488: ERP $94k, Finance $88k · lower used, flagged
- 09Withheld3 records outside this person's access · never read
A frontier model is a brilliant graduate on their first day. Every single day.
Your pricing logic lives in a spreadsheet one person maintains. Your install standards live in a senior tech's head. The reason you walked away from a job in 2023 lives in an email thread nobody has opened since.
A frontier model sees none of it. It can write, reason and summarise better than almost anyone you employ, and it still starts from zero on every task, because nothing about your business was ever put in front of it.
That is not a model problem. It is a context problem, and it is the only problem worth solving first.
Most of that context has never been written down, which is why getting it out is a method rather than a workshop. Here is how we surface it.
- ERP
- CRM
- Inboxes
- Drive / PDFs
- Spreadsheets
- Chat threads
- Standards & policy
- Four people's heads
Brilliant. Knows nothing about you.
Come in for a review, a plan, or a build.
AI Review
A readiness, exposure and architecture review. We find where AI would actually pay in your business, what is already leaking through shadow AI, and what your context is worth.
- A scored context-readiness rubric
- A systems and data map
- A shadow-AI exposure assessment
- A ranked shortlist of candidate workflows
AI Transformation Strategy
The sequenced plan. What to build, in what order, with what it costs and what it returns, so a board can approve it without a leap of faith.
- A staged 12-month roadmap
- A governance and permissions model
- A baselined measurement plan
AI Implementation
We build it. The context layer first, then agents on top, shipped into your environment rather than handed over as a slide deck. The first increment can put a working automation live in 2–4 weeks.
- Connectors into your live systems
- The context layer and governed assistants
- Persona agents and the runbook to operate them
Org Brain. The context layer your business does not have yet.
Org Brain is our reference architecture and component set for a permissioned context layer. We build it inside your environment, you own what is deployed, and we keep the right to reuse the pattern.
It is not another dashboard. It is the layer underneath every dashboard, answer and alert you already have — the thing that resolves what is actually true before any of them render a number.
It resolves identity and permissions before any read happens, not after. It is built so that a person asking a question through Org Brain only sees what they could already see in the system of record. The layer is designed not to create a new door into your data; it reads through the doors you already have.
The model sitting on top of it is a swappable part — today's frontier model, next year's cheaper one, whatever wins the quarter. The layer underneath is the durable asset. It is what keeps paying down after the model you started with is obsolete.
- 01Identity resolutionResolves who is asking before anything is read, not after.
- 02Permission-aware readsAnswers only from what that person could already see in the source system.
- 03Entity resolutionOne customer, one job, one asset — not six half-matching records across four systems.
- 04Precedent and policyYour pricing logic, your standards, your past decisions, held as structure, not folklore.
- 05Freshness controlLive where a stale answer is dangerous, cached only where it is safe to be a day old.
“Which jobs are at risk of slipping this month, and what is it worth?”
- 1ERPjob 4471 · scheduleread live · 40s ago
- 2ContractsCT-2291 · clause 14indexed · 6d ago
- 3FinanceAR ageingread live · 2m ago
Three records matched this question but sit outside this person's access (subcontractor rates, two payroll records). They were never read, and the answer does not depend on them.
An answer you cannot check is a rumour with better formatting.
Every response separates fact from recommendation, names the records it read and how fresh each one was, flags any conflict between systems, and states what it was not allowed to read.
That last line is the one nobody else shows you. It is also the one that decides whether this can go in front of your whole team rather than three trusted people.
The format does not change with the business. A builder, a care provider, a manufacturer and an accounting practice get the same ten lines; only the records differ.
How the layer worksAn honest self-assessment.
Most businesses that think they are at Stage 1 are actually at Stage 0: a licence bought, nothing sanctioned underneath it. Find your stage before you plan the next one.
- Stage 0
AI Zero
Nothing sanctioned. Staff are using consumer AI anyway, and nobody can see what went in.
The testYou cannot answer what was pasted into a chatbot last week.
- Stage 1
AI Enabled
Licences bought, a governed assistant in place, a handful of automations running. Useful, but the model still starts cold on every task.
The testAI saves minutes, not decisions.
- Stage 2
Context Connected
The context layer is live. Systems are joined, permissions resolve per request, and answers cite where they came from.
The testYou can ask your business a question and get an answer with a source.
- Stage 3
AI Native
Agents hold standing jobs on top of the context layer. Each task starts at L2 Propose, moves to L3 within written bounds only on its own shadow-mode record, and money, safety and people stay at L4. The record is complete.
The testWork finishes while nobody is watching, and you can audit exactly how.
Autonomy tiers: L1 Assist · L2 Propose (the default) · L3 Act within written bounds · L4 Approval required always. What each tier means
A working endpoint you can point a client at.
An MCP server is how an AI assistant reads a business as structured data rather than scraped pages. This site runs one, so whoever checks suppliers for you can test ours in a minute.
/mcpget_company_overviewwho we are, and the thesislist_solutionsthe three engagements, and what we buildlist_industriesworked sector exampleslist_insightsevery field note, filterable by pillarget_contact_infohow to start
Point any MCP client at it and this whole site answers as structured data. An MCP server is one of the things an implementation can ship; this one is ours, so you can check the work before you call us.
- 01We baseline before we buildWe measure hours, quote turnaround, rework rate and cash days first, so the change is arguable later, not just felt.
- 02We read live, we do not hoardCopying your data into a second store expands your audit scope and goes stale the day after we leave. We read from source instead.
- 03Permissions resolve per requestNobody sees anything through the context layer they could not already see in the underlying system.
- 04We size the model to the jobMost work does not need the largest model available. Costs should trend down over the life of an engagement, not up.
If a stage will not pay back, we say so — a staged plan only holds together if every stage in it is honest.
We have no logos to show you. Here is our thinking instead.
Start with the argument, not a contract.
Read the case for context first, or skip straight to a fit call. Either way, nothing here asks for a signature before you understand what you would be buying.