Learn · Articles
AI Stack S3 (layer 3, scripts/CLI tools)
Layer 3 of the AI stack is the one most people skip: small tools your team AND your agents can both run.
01 · The stack
Where layer 3 sits
(Quick recap of the stack I drew a few weeks back: 1 — files and folders in plain markdown, 2 — written process guides and logs, 3 — scripts and command-line tools, 4 — the AI agents themselves, 5 — email as the front door. Git underneath all of it.)
Layers 1 and 2 are words — structure and instructions. Layer 3 is where the words turn into something that actually runs.
02 · The idea
A small tool, not a prompt
The idea is simple: anything that's a fixed, repeatable calculation should be a small tool, not a prompt. We run a biowaste recycling business with insects, so ours are things like a command that pulls the live sensor readings off a specific container, a calculator that runs the economics of a BSF case, a script that recalculates the budget from the latest numbers. Boring, single-purpose, deterministic.
03 · Three rules
Three rules for agent-ready tools
Three rules make those tools double as agent tools, not just human ones:
Keep them command-line first. Text in, text out. A tool a person runs in a terminal is a tool an agent can run too, with zero extra work — no buttons, no separate "AI version."
Let the code do the maths, not the language model. The AI should decide WHEN to recalculate the budget; it should never be the thing adding up the numbers. Deterministic code is right every time. A model is usually right — and for anything financial or operational, "usually" isn't good enough.
Build them so the same tool serves the person on Tuesday and the agent on Wednesday. One source of truth for the logic, used by both kinds of worker — exactly like the files and guides in layers 1 and 2.
04 · The thread
The thread through the series
That's the thread running through this whole series: your automation level is set by your documentation and your tools, not by which AI you happen to buy. A small command-line tool that does one thing correctly is worth more to an agent than the cleverest model with nothing solid to stand on.
05 · Over to you
The first repeatable calculation
What's the first repeatable calculation in your business that should be a small tool instead of a person with a spreadsheet?
Keep exploring
Related topics and pages
- Articles library — the written guides this post joins
- BSF business course — where documented assumptions shape a case
- BSF start-up course — the validation steps a small team can run with shared records
Read next
- AI Stack S2: Document your processes like you onboard a new hire — the layer below this one, on written process guides and logs
- AI Stack S1: Reimagine your files and folders — the bottom layer, on files, folders and version control
- Our AI stack, demystified — the layered overview this series extends
- AI Stack S4: Agent software and connectors — the layer above this one, on agent software and connectors
Next step
Make the repeatable part a tool
Manna can help identify the fixed calculations in a BSF case and build the small, deterministic tools that keep them correct for both the team and its agents.