How It Works in AI Engineering

Modern AI coding assistants like Claude Code have a simple but powerful feature: a memory file.

In Claude Code, it's called CLAUDE.md. It sits in your project root and contains everything the assistant should know about your codebase: conventions, architecture decisions, common pitfalls, how to run tests.

The file persists between sessions. Every conversation starts with that context loaded. The assistant isn't starting from zero; it's starting from everything you've taught it.

Some teams take this further with explicit "compounding" workflows. Dan Shipper and Keiran Klassen published a great post in December, and the Claude Plugin, Compound (I've added both to the notes).

Once this is set-up after solving a tricky problem, you tag it:

/compound

The assistant extracts the key insight and writes it to the memory file. Next time anyone on the team hits a similar problem, the assistant already knows the answer.

This creates a new dynamic. The AI isn't just a tool you use—it's a tool that gets better the more you use it. Knowledge compounds instead of evaporating.

From Codebase to Organisation

The same pattern applies beyond engineering. Every organisation has knowledge that's:

• Trapped in people's heads
• Buried in old Slack threads
• Scattered across wikis that nobody updates
• Lost when employees leave

Traditional solutions require deliberate documentation. Write it down. Update the wiki.

But documentation is a second job nobody signed up for. It requires context-switching from doing the work to describing the work. So it doesn't happen.

What if capture happened in the flow of work instead?

Tag, Extract, Persist

Imagine an organisational assistant that works like the engineering tools.

When someone discovers something worth remembering, they tag it:

"@assistant — a few things about client X: their procurement team restructured in March. Key decision maker is now Y, previous contact Z has moved on..."

Or consider compliance. A team member reads through new privacy legislation:

"@assistant — the updated Data Protection Act requires explicit consent for any automated profiling. 90-day implementation window from March 1st. Affects our customer scoring workflow."

Now everyone who asks about customer data handling gets that context automatically—not buried in a policy document nobody reads.

Or internal process discovery:

"@assistant — expenses over $500 need approval, but if you tag it as 'client reimbursable' it only needs manager sign-off. Learned this the hard way."

The knowledge compounds. Each insight makes the assistant more useful. Six months in, new hires benefit from hundreds of lessons without anyone writing a single wiki page.

The Technical Pattern

Under the hood, this is straightforward:

Memory files: Structured storage for captured knowledge, organised by category or domain

Extraction: When tagged, the assistant pulls out the key insight from conversational context

Relevance matching: In future conversations, surface memories that match the current topic

Deduplication: Don't store the same thing twice; update existing entries with new information

The engineering tools have proven this works. The question is applying it more broadly.

What Gets Captured

Not everything needs remembering. The system works best for:

Tribal knowledge: Undocumented quirks that experienced people just know. The staging environment that needs a VPN. The vendor contact who actually responds.

Decisions and context: Why Postgres over MySQL? Why is this timeout set to 30 seconds? Decisions without recorded context become mysterious constraints.

Real processes: How deployment actually works, not the outdated runbook.

Lessons from incidents: Outages and bugs contain valuable signal. Without capture, the same mistakes repeat.

The Compounding Flywheel

Traditional knowledge bases have a scaling problem. More content means harder search, more contradictions, more staleness. Nobody knows what's current.

A learning assistant inverts this. More knowledge makes it more useful because relevance surfaces automatically in context. You don't search—the right knowledge appears when needed.

This creates a flywheel:

1. Team captures knowledge
2. Assistant becomes more helpful
3. People use it more
4. More capture opportunities arise
5. Assistant becomes even more helpful

Engineers have seen this with their coding assistants. The more you teach them about your codebase, the more useful they become. The same dynamic can work for entire organisations.

From Tool to Team Member

The shift is subtle but significant.

An assistant that learns isn't just a tool—it's closer to a team member. It has institutional knowledge. It remembers what worked and what didn't. It gets better at its job over time.

This changes the ROI calculation. A stateless assistant provides the same value on day 300 as day 1. A learning assistant provides compounding returns. The value accumulates.

Getting Started

The technical implementation exists—engineering teams are using it today.

Extending it to broader knowledge management requires:

1. Making capture effortless: A simple tag, not a form
2. Making payoff visible: Show when captured knowledge helps someone
3. Starting focused: One team, a few knowledge categories, expand from there
4. Building the habit: Culture matters more than technology

To use: Select and copy everything above (from "How It Works in AI Engineering" down to "Culture matters more than technology"), then paste directly into your Webflow rich text field. The headings, bold text, italics, and lists should transfer over.

‍