Software is English now

For seventy years, building software meant writing code: a specialist language, compiled into something a machine could run. AI changes the input language. You describe what you want in plain words and a model builds it. The sentence is the program.

This isn’t just a turn of phrase. Andrej Karpathy called English “the hottest new programming language” back in 2023; analysts now expect natural language to write the majority of net-new software within a few years (IDC, via The New Stack).

MemMini takes that literally. Your files aren’t notes about you — they’re the source code of you. who.md says how you reason and what you stand for. how.md says how you work. The wiki/ says what’s true in your world. An AI reads them and behaves accordingly, the same way a runtime reads a binary.

Once it’s the source code of you, you do with yourself what an engineer does with code: read · own · shape · grow · carry · prove. Six moves, one you — you never start over.

Reason → see → build

Plain-English instructions are portable in a way code isn’t. The same words run on the tool in front of you today and the one that doesn’t exist yet, because each AI fits them to its own surface. When an AI reads your files it does what a good engineer does:

reason what’s needed → see what’s already there → build the rest

Write a workflow once; every AI that reads your repo can run it. That portability is why MemMini is files, not an app — and why it can outlive any single vendor.

What this buys you

• Continuity. Say it once; every AI knows it. No re-explaining across tools.
• Ownership. The “you” that tools are learning lives in your repo, not someone else’s product. Nothing phones home.
• Compounding. It drafts its own next version after each session and waits for your yes. It writes itself. You own yourself.

Where the engineering is

“Plain English” doesn’t mean “anything goes.” Because these files are prompts, how you write them changes how well they work — across model families, under a token budget, against the way attention actually behaves. That’s the rest of these docs:

• Why context rots — write for how models read
• Know it’s working — prove a rule works
• The contract-first graph — remember all, read what matters

The long-form rationale (with citations) is the design document.