MAKING THINGS WORK
  1. 01Start
  2. 02The layers
  3. 03In use
  4. 04Context you can trace
  5. 05Other contexts
  6. 06When a source moves
  7. 07The multiplexer
  8. 08Different work, same

SECOND BRAINS

Every AI tool in the business answers from a different version of the facts, and the people who actually know are a third version. The rest is in a few heads and in old file shares. I build a thin, maintained context layer that connects your authoritative sources to the decisions, rules, chronology and methods around them, so that people and agents work from the same facts.

Start a conversation See how it works

What is actually inside it

Five kinds of thing, kept separate and kept current: the sources that are authoritative, the decisions already made, the rules that apply, the order events happened in, and the methods worth repeating.

A task wakes only what it needs

Preparing the project status report needs the current milestones, one decision record and the agreed report format. Nothing else is loaded, so the answer is narrower, cheaper and far easier to check.

Context you can trace

The context attached to the task is readable, and each part carries the source it came from and the person who owns it. An answer nobody can trace is an answer nobody will act on.

The same structure, a different job

A research effort, a DevOps environment and a case chronology each bring different parts of the same structure forward. The shape of a second brain does not change with the subject.

When a source moves

A source that has changed or gone missing marks everything depending on it for review, and names the outputs affected. Stale context is worse than missing context, because it still looks like an answer.

One route to the work

A project lead asks for the weekly report. The routing layer uses that project's permitted context and the organization's report format, gathers the updates, prepares a draft and stops at a review step with a person's name on it.

Different work, same rules

A case chronology links events to the evidence supplied and keeps disputed entries visible instead of resolving them quietly. A research task runs to a bounded objective, with permitted tools, a budget and the conditions under which it stops.

What it is

A second brain is a maintained context layer for a project, a research effort, a DevOps environment, a case, a program or a team. It connects the sources that are authoritative to the decisions already made, the rules that apply, the order events happened in, and the methods worth repeating. It is built as a wiki, it can live in SharePoint as the knowledge anchor Microsoft Copilot reads from, and it works with Microsoft Copilot or with Foundry in your own Azure tenant.

The multiplexer sits on top of it: one place to surface and route work according to the organization, the person's role, the project or case in front of them, the rules that apply, and the tools permitted for that task. It covers document generation, form completion, autonomous research, case management, project management and chronology.

What I still do inside it

  • Intranets and knowledge hubs

    For a hospital foundation, a US family philanthropy and a large Toronto private foundation I built the intranets and knowledge portals their information now lives in.

  • Fifteen years on one estate

    For a national law firm I built and rebuilt the intranet, the personalized application directory and the search centre between 2010 and 2025.

  • Enterprise search

    For a national law firm I replaced a third-party on-premises search product with PnP Modern Search and contributed V3 support and a V4 beta back to the open source project.

  • Knowledge systems across geographies

    For a global road-construction group I built a knowledge management system for multiple regions and languages, and a facility database that non-technical teams fill in through forms and browse on a map.

  • Information design that won awards

    Earlier, as director of engineering at a Toronto web consultancy, I led a global law firm intranet across its international offices with a team of seven developers, work recognized by the Nielsen Norman Group Intranet Design Annual and, for a custom search solution, a Step Two Designs Intranet Innovation Award at platinum.

What AI adds

  • One set of facts for every toolA second brain shared between Claude, Perplexity and Microsoft 365 Copilot, so every agent starts from what the business actually decided.
  • A wiki that improves as the work proceedsDelivery knowledge captured as reusable skills inside a documentation wiki that the work itself keeps current.
  • Copilot reads from itA second brain in SharePoint that Microsoft Copilot uses as its knowledge anchor, with the skills that shape how it answers.
  • The multiplexerOne route to the workflows a person is allowed to run, chosen by their role, the case or project in front of them, and the rules that apply to it.

What you receive

  • A source and context modelWhich sources are authoritative, what context hangs off them, and how the two stay connected.
  • Working context modulesThe context itself, built and in use, rather than a description of what it would contain.
  • Ownership and refresh rulesWho owns each part, how often it is refreshed, and what happens when a source underneath it changes.
  • Integrations and repeatable methodsThe selected connections into the systems the context comes from, and the skills or methods that can be run again.
  • Tested example tasksReal tasks run against the context, so you can see what it produces before anyone depends on it.
  • Chronology and decision recordsFor case and project work, events and decisions linked back to the source they came from.

Where I have done this

Described by industry.

  • A skilled-trades governance startup

    The engagement runs on a second brain built as a wiki and shared between Claude, Perplexity and Microsoft 365 Copilot, so that every agent's context is anchored to what was actually planned. Alongside it I built the system architecture and an agentic build scaffold with multi-agent, test-driven build loops. Knowledge is built into the wiki, documents are generated from it, and questions are answered by it.

  • A national law firm

    A modern enterprise search centre replaced a third-party on-premises product, with verticals for sites, how-to guides, people, clients and matters, and precedents. V3 support and a V4 beta were contributed to the PnP Modern Search open source project during that build. The relationship ran from 2010 to 2025 and included a personalized application directory where roughly 825 users chose roughly 21,000 icons.

  • A global road-construction group and its Canadian asphalt subsidiary

    A knowledge management system served multiple geographies and languages, with requirements gathered through wireframing and mind mapping and training delivered to the knowledge team. A global facility database collects records through forms, geocodes the addresses automatically and presents them on a map with analytics behind it.

How it starts

The first conversation is about the thing your team keeps re-explaining, to each other or to a tool. I look at where the authoritative answer actually lives today, and come back with the sources worth connecting, the first context modules to build, and one real task to run against them.

Scope, duration and price are set after that conversation, in writing, against what we agree the work has to do. Nothing on this site is a quote or a guarantee.

Tell me what needs to work better

Pick the question your team answers over and over. That is usually where the first context module belongs.

Start a conversation

Related: Move Data.