Method · Patterns · Operating Model

How we work.

Method, patterns, and operating model, codified from production systems and re-applied with discipline.

01 The method

One engagement, twelve months minimum.

There is one engagement, and it begins before there is an invoice. Scope, architecture, and the business case are worked out in conversation. That is business development, not a billable deliverable. We do not sell assessments, because an assessment is not the thing that changes an operation, and a client who has bought only a roadmap has bought the one part of this work that cannot act on its own.

The work starts the day the contract is signed, with analysis and a benefit impact assessment: what the operation needs most, weighed against what can realistically be built and in what order. That is what sets the build priority. The sequence follows the benefit rather than a generic maturity model. There is no separate onboarding stage to buy; that part is administration.

From there we build against the highest-value bottleneck first: the full Business Operating System where that is warranted, a single bespoke system where it is not. Then we run it, on a monthly retainer, with a quarterly review of how the system is performing. The term is twelve months at minimum.

A roadmap changes nothing on its own. We build the system, and then we run it.
02 The model layer

The models are substitutable.
The system around them is not.

The Business Operating System is model-plural. A router sends each task to whichever model suits it, weighing reasoning depth, latency, and cost per workflow instead of settling it once in a procurement meeting. Anthropic, OpenAI, and open-weight models sit behind one interface, and the selection is revisited as the field moves. None of that is remarkable; it is basic competence at the model layer, and we treat it as such.

What is not substitutable is the operating system: the governed tool surface, the orchestration, the composition surfaces where people review and commit, and the audit trail beneath all of it. That end-to-end platform is what an engagement delivers, and it is where the value stays as models come and go. We build in Claude Code every day. That is a fact about how we deliver, not a dependency in your architecture.

So what a client buys is a working system and the accountability for it, not fluency in someone else's tooling.

01
Model routing

Each task goes to the model that suits it, weighing reasoning depth against latency and cost per workflow instead of defaulting. Basic competence at this layer, instrumented rather than assumed.

02
Provider independence

Anthropic, OpenAI, and open-weight models behind one interface. Changing model is a configuration change, not a rebuild, so the architecture never depends on one vendor's roadmap.

03
Configuration as code

Prompts, skills, and subagent definitions are versioned, reviewed, and shipped with the application around them. Nothing is pasted into a console and called deployed.

04
Model Context Protocol

Enterprise systems exposed through governed MCP servers, an open protocol and the single seam where scope, identity, and audit are enforced.

05
Cost and quality instrumentation

Every call carries its cost and its measured outcome. Routing decisions are reviewed against real spend and observed quality, not vendor benchmarks.

06
Evaluation & governance

Behaviour measured against written evals before deployment; safety, refusal, and review surfaces designed in, not appended.

Model vendors build the models and the platforms. We build the operating system, the governance, and the accountability around them, and we never imply otherwise.

03 The pattern library

Five patterns. Each one battle-tested before it ships.

The library is the firm's leverage. Each pattern is documented, instrumented, and reused. New engagements pay for novelty only where novelty is warranted.

01
MCP Gateway

A governed seam between agent systems and the existing enterprise stack.

Production · Regulated FS · Sustainability comms
Stack · Cloudflare Workers · MCP · OAuth 2.1 · audit log
Phase · Build and run

The Gateway sits between agents and the systems-of-record. It enforces scope at the tool level, attaches caller identity, logs every invocation, and rate-limits per-agent and per-tenant. Without it, agents accumulate trust the organisation cannot reason about.

In regulated contexts, the Gateway is the audit surface. Compliance reviews look at it, not at the agent's prompt.

// MCP tool definition — registered on the gateway
{
  name: 'list_invoices',
  description: 'List unpaid invoices for a tenant.',
  scopes: ['finance.read'],
  rate_limit: '60/min',
  audit: 'full',
  handler: async (ctx, args) => { /* … */ }
}
02
Portal Shell

The composition surface where humans review, edit, and approve agent output.

Production · Education · Regulated FS
Stack · Nuxt 3 · MCP client · Workers
Phase · Build and run

The Portal Shell is where humans stay in the loop. Agents do the drafting; humans do the review and the irreversible commit. The Shell makes that boundary structural rather than a checkbox.

Every panel renders agent output adjacent to its inputs and audit trail. Approvals carry the reviewer's identity into the system-of-record.

03
Upstream MCP Template

A repeatable scaffold for exposing third-party services as MCP tools.

Production · Education · UN programmes
Stack · TypeScript · MCP SDK · Workers
Phase · Build

Most clients have ten to twenty services that an agent benefits from reading or writing: CRM, ERP, document store, calendar, billing, ticketing. The template gives each one a consistent MCP surface in under a day of work each.

Naming, scoping, error handling, and audit all share one shape. Agents learn one pattern, not twenty.

04
Agent Configuration as Code

How model, prompt, and tool surfaces are versioned and deployed.

Production · All sectors
Stack · Model APIs · Skills · Subagents
Phase · Build and run

Prompts, skills, and subagent definitions are code, versioned, reviewed, and shipped with the same rigour as the application around them. We never paste prompts into a console and call it deployed.

Configuration travels with the engagement: a client receives the full kit, not a black box.

05
Portal Chat

In-context conversation that grounds in the portal's state, not a generic transcript.

Production · Education · Regulated FS
Stack · Portal Shell · MCP · streaming
Phase · Build and run

Chat is the wrong default for most enterprise workflows, but the right default for ad-hoc questions, drafting, and explanation. Portal Chat lives inside the Shell, sees the same state the user sees, and routes to the same governed tool surface.

It is not a separate product. It is the conversational facet of the Portal.

04 Reference architecture

The Business Operating System, in five layers.

Our reference architecture is informed by publicly-discussed AI-maturity frameworks, DAIN among them, but the five-layer model below is our own. We name what we built and credit what we built on; we don't claim either as the other.

L1
Foundation Models
Anthropic, OpenAI, open-weight. Routed per task; selection is engagement-specific.
External
L2
Tool Surface
MCP servers expose enterprise systems as governed tools. Single seam.
Ours
L3
Agent Orchestration
Subagents, skills, and routing. Compensation, retry, and audit baked in.
Ours
L4
Composition Surface
Portal Shell + Portal Chat. Humans review and commit.
Ours
L5
Governance
Identity, audit, scopes, rate limits, reviewer attribution.
Ours

The Business Operating System, our five-layer reference architecture. The name describes our implementation, not a category we claim to own. Not every engagement uses every layer: a bespoke build may need only the tool surface, the composition surface, and the governance beneath them.

The reference architecture deploys as CLTF, the firm's codified platform, currently in production in our own operations and on selected client engagements. Client-integrated deployments adapt the platform into the client's own infrastructure; hosted engagements run on instances we operate. The platform is the firm's leverage; the engagement is the work.

05 Operating model

Agent-native, by construction.

The principle is short: control planes get MCPs, not UIs. Anywhere a tool exists primarily to coordinate work, we expose it as an MCP server and let agents drive it. Anywhere a tool exists for composition or review, the human surface stays.

Internally, the firm runs the same shape: a TobDog scheduler, an EA agent, a business-development agent, a delivery agent, and a CFO agent. They run the practice between client engagements. That is what makes a one-architect consultancy possible without theatre.

Begin

Start a conversation.

Scope, architecture, and the business case are worked out before anything is signed. Send a brief with the context of your situation.