Executive AI OSNo. 04
Your new operating system depends on context
How to give AI the instructions, information and access it needs to help with everyday work.
Gonzalo GilgesHead of Practical AI
- Reading time
- 5 min
A person can be highly capable and still fail to do the work you need. Someone who has just joined the team does not yet know the clients, priorities or ways of working. They lack context.
The same is true of AI. To prepare a useful proposal, it needs to know what the client requested, what was discussed and what the company can offer. Without that, it may produce polished writing that is of little practical use.
Context is the information and criteria that allow a task to be completed in the way we need.
Working on context means recording that knowledge, keeping it current and giving AI access to it. A better prompt is not enough if the necessary information remains scattered or only exists in people’s heads.
We can organise it into three boxes: Instructions, Company information and Tools.
Fig. 1Three foundations for working with context
- 01 AI
- 02 Instructions
How to work.
- 03 Information
about projects, clients, suppliers, etc.
- 04 Tools
Access to sources. Consults Data
01
Instructions: document how work should be done
Consider the commercial report that a sales director prepares for the CEO: results, gaps against target and pending decisions.
That procedure can be saved as a skill: reusable instructions containing steps, templates and examples so AI knows how to prepare the work.
Share an approved report, explain how you obtain the figures and refine the result with AI. Then ask it to turn that procedure into a skill. Save it and enable it in your AI tool. Test it and update it whenever the way of working changes.
A financial analyst can do the same with the indicators and structure used in reports for an investment committee. The company being analysed changes, but the procedure does not need to be explained from scratch.
02
Information: capture what happens each day
The AI now has instructions on how to prepare the work. It also needs to know what happened: progress, decisions and changes in priority.
Tools such as Granola, Fathom or Piper AI can preserve meetings as notes and transcripts. Connect the tool to your AI through a compatible integration so it can consult those records when you ask.
You can share a link or ask it to find the note by name: ‘Find Monday’s project review. Identify decisions, tasks, owners and dates. Flag anything that was left unresolved.’
Then ask it to compare the updates with the project tracker and prepare the changes. If it has editing access, it can update the tracker when instructed. Discussions in face-to-face meetings and telephone calls also need to be recorded if they are to be used.
Fig. 2What happens each day becomes useful context
- 01 Notes and transcripts
Virtual meetings · Face-to-face meetings · Calls.
- 02 Email · Slack / Teams
Messages and emails in connected tools.
- 03 AI organises
Decisions, progress and open actions.
- 04 Documentation
Shared and updated by project or client. Validated corrections
- 05 Reports · Proposals
And work tracking.
AI proposes changes. The team reviews what requires judgement. The documentation is ready for future requests.
03
Tools: connect AI to the sources
One of this year’s major changes has been the expansion of connectors based on MCP, the Model Context Protocol. It is a standard that allows AI agents to connect with work tools in a common way. Compatible integrations can provide authorised access to information and, depending on permissions, allow actions to be carried out within those platforms.
Previously, much of that context was outside AI’s reach: people had to find and copy it manually. Now, when a tool offers a compatible connection, you can link it from your AI application and decide which access to enable. If you use a team or enterprise licence, the account administrator configures the available connections; each person uses them according to their permissions.
Google Drive or OneDrive can hold documents about projects, clients and suppliers. An analyst can ask AI to compare current financial statements with projections and previous reports.
Slack or Teams contain agreements, progress and blockers. It helps to organise conversations by project. If a decision only exists in WhatsApp and AI cannot access it, record it in the shared tracker.
You can also connect CRMs such as HubSpot to consult opportunities, values and next actions, or connect an ERP to review orders, invoices and collections through an appropriate integration.
If no connection exists, place an up-to-date export in Google Drive, OneDrive or an accessible folder on your computer. A file of unpaid invoices can be used to prepare a collections summary. Refresh it so that AI does not work from old figures.
Products such as Cowork and Claude Code from Anthropic, or ChatGPT Work and Codex from OpenAI, provide different ways to consult files and tools. Each offers its own functions and connections.
04
AI at the centre of the information flow
The three boxes complement one another: instructions explain how to work; company information describes what is happening; and tools make it possible to consult those sources or act on them. The team defines permissions, reviews results and keeps the information current.
If we still need to search for files, copy conversations and carry every answer back into another application, we are still moving information by hand. When AI can consult the sources and follow our procedures, it moves to the centre of the information flow rather than sitting at the side.
Start with one space per project: tracker, sources and skills. Every recorded decision updates the information; every correction to a procedure improves the instructions. That is the compounding effect: today’s work leaves a more useful base for tomorrow.
It requires patience and maintenance. Choose one recurring task and begin with its three boxes. Record what happens, document how the work is done and enable the necessary sources. With each review, you will discover which information is still missing for AI to get closer to the result you need.
You are welcome to write to me if you have a specific question about good practice for bringing AI into your day-to-day work.
Thank you for reading. See you next time.