An AI operating layer for your whole business.

An AI operating layer is a layer that sits underneath the business and runs over every touchpoint: calls, emails, deals, project state, documents. It is made of four parts: a skills library that codifies your playbooks, a meeting-capture layer that turns every conversation into structured action, a connector layer that plugs AI directly into the systems where the work lives, and an agent layer that watches the business and surfaces what needs attention. Skills are passive and run when invoked. Agents are always-on and event-triggered. Together they shift AI from a tool the team reaches for to a layer running underneath the work.

Skills are structured playbooks the model follows when invoked by a person or event: a discovery-prep skill, a status-report skill, a proposal-builder skill. Agents are always-on processes that monitor business events and propose actions to humans for approval. Automations are the connector-layer wiring that lets skills and agents read from and write to your CRM, project tracker, and document store. You need all three: skills do the thinking, automations give the model the data, agents decide when to act.

Invisible AI is the end state of an AI operating layer: the layer runs underneath the human work, surfaces only when it should, and reduces 'what should I be doing right now' to a Slack message you say yes or no to. The team interacts with the business through conversation. The operating model (common work automated, rare work human) is structurally different from a traditional consultancy or operations team.

No. Claude is what we run on internally and what most rollouts start with, but the same skills, connector, and agent pattern installs on OpenAI's API, Google Vertex, Microsoft Copilot Studio, or whichever AI platform your team already uses. The training, skills documentation, and agent playbooks are model-agnostic by design. They're Markdown files and JSON specs, not Anthropic-specific. If you already have an enterprise contract with one provider, we build on it. If you are choosing, we will recommend the right fit during the discovery call.

Licences give your team a chat interface. They do not give the business an operating model. Without skills, every team member improvises prompts and gets inconsistent output. Without connectors, the model has no access to your real CRM or project data, so its work is generic. Without agents, nothing happens until a human remembers to ask. The licences are the model. The operating layer is the system that makes the model do useful work consistently across the business.

The AI Transformation Roadmap is a four-week strategy engagement that produces a board-ready document. The AI Operating Layer is the implementation that follows the roadmap. Most clients run the roadmap first to identify which functions to install the layer against and in what sequence. A few skip the roadmap because they already know what they want. For those, the first phase of the operating-layer engagement absorbs the relevant strategy work.

The first function is typically live in six to ten weeks: a narrow vertical slice that touches every layer (skills, connector, agent, Slack approval). Full multi-function rollouts run over three to nine months depending on the number of business functions, the state of the underlying SOPs, and how much data integration is required. Each phase is independently testable. You do not commit to the full programme before the first phase is live.

Your team. The whole point of the layer is that it is owned by you, not rented. We install the skills library and train your team to extend it as part of doing the work. New team members learn the workflows by reading the skills. Improvements made by anyone on the team flow back into the skill itself, so the next person inherits the upgrade. The learning agent then closes the loop automatically: when a human edits an AI-generated draft, the agent updates the source skill and the evaluation suite so the same correction is never needed twice.

Rollouts are scoped on a discovery call. Pricing is a fixed fee per phase (Design & Build, Integration, Adoption), not hourly. Cost depends on the number of functions in scope, the state of the underlying SOPs, and how many systems need connectors. Most engagements run alongside an existing AI Transformation Roadmap or follow on from one.

A narrow vertical slice: one compliance rule, one trigger, one Slack channel. We pick the rule that closes the biggest documented operational gap (often something like 'won deals must have a complete handoff package before kickoff is scheduled'), build the agent locally, deploy it, and wire the trigger to your CRM or project tracker. The team sees the first real Slack approval inside a few weeks. The slice proves the pattern works on your business, with your data, before you commit to the full rollout.

Yes. We have deployed operating-layer engagements multiple times across different industries and operating models. The breadth and scale of each design and implementation is scoped to the company we are working with and the problems it is actually trying to solve. A two-sided SaaS marketplace looks very different from a professional-services firm, which looks different again from a compliance-heavy operator. The underlying discipline (map the lifecycle, codify the skills, wire the connectors, install the agents) holds across all of them.