Working Memory

How many open loops are draining your attention right now?

Your cognitive workspace. The small, volatile buffer where active thinking happens — roughly 4-7 chunks at once. Every context switch flushes the buffer. It takes 23 minutes to return to deep focus after an interruption. Context switching consumes up to 40% of productive time.

Without Memory Management	With Memory Management
Constant mental juggling	Externalised, trusted system
Attention residue between tasks	Clean transitions
Shallow work by default	Deep work by design
"Where was I?"	"Here's where I left off"

Your brain is for having ideas, not storing them.

Memory Architecture

The same patterns appear in human cognition and AI systems:

Type	Human	AI Equivalent	How to Manage
Working	Active thought buffer (4-7 chunks)	Context window	Externalise — write it down
Episodic	Specific experiences and events	Conversation history, logs	Journal — capture while fresh
Procedural	Learned skills and habits	Fine-tuned behaviours, routines	Practice — repetition builds muscle memory
Semantic	General knowledge and facts	Training data, RAG retrieval	Reference systems — don't memorise what you can look up

Memory sophistication is the differentiator. For both humans and agents, better memory management means better reasoning.

Externalisation Protocol

Get everything out of your head into a system you trust:

What to Externalise	Where	Why
Tasks and commitments	Task manager (one system, not many)	Open loops drain working memory in the background
Meeting notes and decisions	Shared docs, immediately after	Memory degrades within hours
Ideas and insights	Capture tool (always accessible)	Inspiration decays fast
Project context	"Where I left off" notes	Eliminates 23-minute recovery cost
Reference material	Searchable knowledge base	Don't memorise what you can look up

The test: do you trust your system enough to forget? If not, your brain is still holding the load.

Chunking Strategy

Working memory holds 4-7 chunks. Make each chunk bigger:

Raw Information	Chunked
0-4-1-5-9-2-6-5	04-15-92-65 (four chunks, not eight)
12 unrelated tasks	3 projects with 4 tasks each
Complex argument with 8 steps	3 phases: setup, core, conclusion
New codebase with 50 files	5 modules, each with a purpose

Experts and novices have the same working memory size. Experts just have bigger chunks.

Context Protection

Threat	Cost	Defence
Notifications	23 minutes per interruption	Silent mode during deep work
Task switching	40% productivity loss	Batch by type, theme your day
Open browser tabs	Continuous partial attention	Close everything not needed now
Meetings without breaks	Residue from previous context	5-minute buffer between meetings
Unclear next action	Decision fatigue, procrastination	End each session by writing the next step

Daily Protocol

1. Morning — Review your system. Load today's context. Identify the one thing.
2. Deep work — Peak energy hours. Single task. No inputs.
3. Process — Clear inbox, respond, update system. Batch reactive work.
4. Close — Write where you left off. Empty your capture tool. Close all loops you can.
5. Rest — Trust your system. Let your brain recover.

The Shadow

Hoarding information. Refusing to externalise because "I should remember this." Over-optimising systems instead of doing the work. Mistaking organisation for output. Building the perfect system instead of shipping the imperfect thing.

By Archetype

Archetype	Memory Style
Engineer	Holds systems in mind — builds from complexity with external scaffolding
Philosopher	Synthesises across domains — needs deep, uninterrupted context
Coach	Remembers people — context about individuals and relationships

Context

• Focus — Memory management sustains deep work
• Planning — Working memory enables execution
• Deep Work — Protecting the cognitive workspace
• AI Data Pipelines — How AI extends memory