Stop Re-Explaining Yourself to AI

You have already worked out that AI gives better answers when you give it context first. The problem is that you supply that context from memory, by hand, and never quite in full - in every new chat, in every tool. A working context and memory system ends that.

Get the template
The Problem

Every new session, you start from scratch

The efficiency AI promised quietly disappears each time you rebuild your context from memory. Here is what that looks like in practice.

The blank chat window

Every new conversation starts from zero. You rebuild a partial, improvised picture of your work - slightly different each time, never quite complete.

Per-tool amnesia

What one tool knows, the other does not. Switching tools means starting again. Your context does not travel with you.

Repeating yourself

Having to re-explain is exactly what kills the efficiency the tool promised. You spend time briefing instead of working.

Built-in memory guesses

Native memory stores plenty but selects unpredictably. A relevant fact can exist without being retrieved when it matters most.

The fix is not a better prompt. It is deciding, deliberately, what the tool should know - and loading it on purpose.

The System

Three layers and one discipline

The system is built from three plain files and a short ritual at the start and end of every working session. It requires no platform, no subscription, and no technical setup - only the decision to maintain it.

1

Active Rules

A short, curated set of standing preferences and rules, edited in place so it always reads as current state. This file is read first, every session - it is what makes your sessions feel continuous rather than improvised.

2

The Log

An append-only, dated record of decisions, corrections and learnings. History is never rewritten; it is where provenance lives. The log is what lets the system learn rather than only react.

3

Durable Reference

The stable facts about your work, clients or organisation that any session should be able to load. This changes rarely but earns its place every time a session needs background that would otherwise take minutes to reconstruct.

The discipline is knowing what not to record. One-off outputs and routine activity are not logged - only things that change future behaviour, or that you will genuinely need to re-read. Load the relevant layers at the start of a session; at the end, decide what deserves recording, and append it. It works in any AI tool, because it is plain files, not a platform feature.

Point, do not copy

Most of your real work already lives outside these files - in documents, decks, spreadsheets and threads. Copying any of it in would turn a system you can read at a glance into a second copy of everything, one that immediately starts going out of date. So the system records a pointer instead: what the thing is called, where it lives, what it is for, and when a future session should bother opening it. That last part is what makes it work, because it lets the assistant decide whether to load something rather than loading everything. The result is that the system stays small while staying connected to everything, and nothing gets written twice.

A nervous system for your work with AI

The tools already supply the thinking - what they lack is a body that remembers. Active rules behave like reflexes: standing responses that fire without being reasoned out from scratch each time. The log is memory proper, a dated record of what happened and why. Durable reference is the body of knowledge everything else draws on. The session ritual is consolidation: at the end of a working session you decide what is worth keeping, and it becomes part of the system instead of being lost. And the pointers are the connections, which is why the system can stay small and still reach everything.

This is an analogy to aid understanding, not a technical claim.

It does not stop at notes

The same three layers hold the two things a working life actually runs on: outstanding actions and live relationships. Actions belong in a task manager rather than in the notes - the free tier of a common task tool is enough for a solo practice, with the caveat that free tiers limit how many people can share a project.

Relationships need nothing more than one row per organisation or contact, recording route in, stage, last contact and next action. That is a fraction of what a CRM offers - and the part people actually use.

Both main assistants can now connect directly to common task and note tools, so the AI can read what is due and record what changed. The system is updated as a by-product of the session rather than as a separate chore. That is what makes the discipline survivable.

The connectable task tools, correct at the time of writing and worth checking against your own assistant's directory since these lists change, include Asana, Todoist, Trello, ClickUp, monday.com, Linear and Jira - most have a free tier that covers one person. The method does not depend on which one you use, only on having somewhere outside the notes where outstanding actions live.

Honest Comparison

What about the memory built into the main AI assistants?

A fair question - and one that deserves a straight answer rather than a sales pitch. Here is the honest comparison.

Native Memory

Zero effort, broad reach, and improving fast

Selection is uncertain - the tool decides what is relevant

Tied to one account and one vendor

Works automatically without any ritual or maintenance

A Deliberate System

Narrower but curated - loaded on purpose

Current decisions reliably reach the model, every session

Portable across tools, accounts and vendors

Yours as plain files - costs maintenance discipline

What will and will not close

Memory inside any one tool is improving quickly, and the selection problem will keep getting better - so the case here does not rest on today's weaknesses. The gap that is not closing at the same speed is the boundary between tools. Each vendor's memory is a retention feature for its own product, not a portable record of how you work.

Where this is heading

The vendors have begun shipping memory export and import. Third-party memory layers now sit above several tools at once using open connection standards, and a vendor-neutral memory specification has been proposed. Expect portability to improve. Even if it arrives fully, the part that stays yours is the judgement about what was worth recording in the first place - and that is the part no vendor can do for you. Plain files already survive whatever the vendors decide.

The two are complements, not rivals. Native memory catches what you forgot to record; the deliberate system guarantees what you cannot afford to have missed.

Seen in Practice

This is already running - daily

This system is run daily across two organisations and multiple AI tools. The same session ritual has kept context continuous across months of work and tool switches - including picking up a client's thread weeks later without re-briefing the tool at all. The context was already loaded; the session started where the last one ended.

Relationship tracking without CRM software

The same three layers now carry the relationship tracking for the whole practice - holding every organisation, how the thread started, what stage it is at, when it was last touched and what happens next, alongside a task tool that holds only what is genuinely outstanding. It does the job people buy CRM software for, without CRM software.

A conventional CRM is only ever as good as the discipline of updating it, and updating it is a separate chore done after the work - which is why so many sit half filled and quietly stop being trusted. Here there is no separate chore. The record is updated as a by-product of the working session itself, because the assistant you are already talking to about the client is the thing that writes down what changed, what was agreed and what happens next. The system stays current because keeping it current is not a task.

Origin and grounding

This grew out of the same discipline behind Grant Resource Studio's funding work. AI is only as good as the knowledge beneath it. There, the knowledge is a Funding Knowledge Base; here, it is your own working context.

More Than One Person

It works across a team, even when everyone uses a different AI

Because this is plain files in a shared workspace rather than a feature inside one assistant, several people can run the same system together while each keeps their own AI account and their own preferred tool. One person works in one assistant and a colleague in another, and both read the same active rules, append to the same log and load the same durable reference. Nobody has to standardise on a vendor, nobody pays for seats they do not want, and no one's individual chat history is shared - because what is shared is the deliberate record rather than the conversations.

Two disciplines make it hold: the log is append-only, so nobody overwrites anybody, and entries are dated so it stays clear who decided what and when.

Where the files actually live

On your own machine, as a folder of plain files. This is the simplest option, and it already works across more than one AI tool - any tool with access to that folder reads and writes the same files, and a tool without folder access can still be given them by hand. Put the folder in a synced drive and it follows you between machines. Choose this if the system is yours alone.

In a shared workspace - a team wiki or a shared document space. This takes a little more setup, and it is the option to choose if anyone else will ever use the system with you. A folder on one person's machine cannot be a shared source of truth: a shared workspace is a requirement for working with other people, not a preference. A common arrangement is to keep the shared workspace as the master and a local copy as a backup.

This is the version that survives a team, because it belongs to the work rather than to one person's account.

Get the Template

Build your own

A free skeleton template set as plain files - ready to fill in, not to admire. This includes the relationship tracking shown earlier, so you can build the same thing yourself. Nothing is held back: this is the whole method, and anyone who takes these files and does the work will end up with a working system. No email required.

Active Rules file

Your standing preferences and rules, ready to edit in place and load at the start of every session.

Log file

An append-only template for dated decisions, corrections and learnings. History lives here.

Durable Reference file

The stable facts about your work, clients or organisation - the background any session can load.

Relationship tracker skeleton

One row per organisation or contact, with the columns that actually earn their place.

One-page session guide

Covering the session ritual, how to point the tracker and a task tool at your assistant, and the what-not-to-record test.

Three steps to get started

01

Fill the reference file in one sitting

Roughly and honestly. Done beats perfect - you will refine it as you go.

02

Run one week of sessions loading it at the start

Notice what you reach for that is not there yet. Add it to the right layer.

03

Start appending to the log

Only when something genuinely changes how you would work next time. That is the whole test.

Honest Limits

This takes discipline - and that is worth saying plainly

Value builds over weeks

The payoff is real, but it is not instant. The system earns its keep through repetition, not through a single impressive session.

An unmaintained system quietly rots

If you stop appending and stop loading, the system drifts out of date. The ritual is not optional - it is the mechanism.

Chaos in, chaos out

If your notes are chaotic, the system organises your chaos rather than fixing it. Clear thinking still has to come from you.

Who it repays most

The people it repays most are those juggling several clients, projects or organisations at once - where the cost of lost context is highest and the benefit of reliable continuity is most felt.

Workshops

Run this as a workshop

For networks, membership bodies and infrastructure organisations: a hands-on session where participants design their own system in the room. Scoped at 45 to 90 minutes, suited to training budgets and member-value programmes.

Participants leave with a working skeleton of their own system - not a slide deck to file away, but a set of files they have already started filling in. The session is structured around the three layers and the session ritual, applied live to each person's actual work context.

Format

45 - 90 minutes, hands-on in the room

Suited to

Networks, membership bodies, infrastructure organisations

Outcome

Each participant leaves with a working system skeleton

Everything on this page is yours to take and build yourself. If you would rather not spend the next month working out what belongs where, some people have it built with them instead, shaped around how their own practice actually works. If that is you, book a call.

The Funding Bridge

The bigger version of this problem

If the bigger version of this problem in your organisation is grant applications, the same thinking applied to funding knowledge is the core of what Grant Resource Studio does. The method here - deliberate context, curated layers, disciplined recording - scales directly into the funding world, where the knowledge beneath the application is everything.

Grant Resource Studio

The home of the funding knowledge approach. grantresourcestudio.com

Free Grant Readiness Starter Kit

A free starting point for organisations thinking seriously about funding. starterkit.grantresourcestudio.com

Early Years Cocoon Case Study

See the approach applied in practice, in a real organisation with a real funding knowledge base, and what changed as a result. casestudy.grantresourcestudio.com