AI for Bookkeeping: What It Speeds Up and the Totals It Should Never Touch
AI for bookkeeping, done safely: the tasks it speeds up, the totals it should never calculate, and a prompt that traces every figure back to its source.
Meelie Gagne
How to use AI at work
Every guide shows the prompt and why it works, so you can adapt it to your job. New to AI? Pick the track for your role. Already prompting? Skip ahead to agents and security.
SYSTEM:
You classify customer support tickets. Return only JSON matching:
{"label": "billing" | "technical" | "account", "confidence": number 0-1, "reasoning": string}
Labels:
billing: charges, invoices, payments
technical: product not working as expected
account: access, password, profile, subscription state
Choose the label the ticket primarily concerns. When ambiguous, prefer the
category that describes what the user wants to happen, not what caused the
problem. Include a one-sentence reason.
USER:
{ticket_text}Read the playbook: Build a Classification Prompt That Beats Fine-TuningAI for bookkeeping, done safely: the tasks it speeds up, the totals it should never calculate, and a prompt that traces every figure back to its source.
Meelie Gagne
Five subjects, from writing a better prompt to keeping an agent from being tricked. If AI is new to you, begin with Start here.
Chain-of-thought, few-shot, structured output, meta-prompting.
Multi-step chains, tool use, orchestration, feedback loops.
Injection defense, red/blue team, threat analysis.
Code gen, debugging, tests, CI: prompts that ship code.
Healthcare, finance, legal, marketing by role.
There's no newsletter yet. Until there is, here's how to hear about new guides and ask questions.
Every new guide lands in your feed reader, with nothing to sign up for.
Subscribe with RSSAsk a question, or share a prompt that worked for you.
Join the DiscordStuck on a prompt, or want a guide on something? Write in. A person reads every message.
[email protected]One job, start to finish: what you need first, each prompt in order, and what you end up with.
A production-ready classification prompt for your own labels, with self-consistency voting, structured output, and an eval harness, with 90%+ accuracy as the target and no fine-tuning.
A repeatable test suite measuring faithfulness, answer relevance, and context precision across 50-200 Q&A pairs. Ship gates on it; your RAG quality stops being vibes-based and becomes observable.
A regression-runnable suite of 40+ injection probes that catches the classes of attacks your agent will actually face. Run nightly, track pass rate per prompt/model version, catch regressions before users do.
Cut agent costs by routing cheap sub-tasks to Haiku-class models and reserving the frontier for planning. Includes a full Python reference implementation, cost accounting, and the fallback logic.
A pre-flight cost calculator that tells you, before you call, how many tokens a given prompt + expected response will burn, across multiple models, so your app can choose the cheapest model that fits the window.
Plain definitions from the glossary, each with a prompt to try and when not to use it, plus the topics we write about most.
That's a fair call. LTFI, the agentic business operating system from Kief Studio, can set up the work these guides describe and run it with you. No website yet? Kief Studio will build you one for free.
AI for administrative tasks, step by step: turn a full inbox, a meeting transcript or a long document into a first draft, and what to strip out first.
Meelie Gagne
How to use AI at work in your first week: five tasks to hand your team's AI assistant, three to keep for people, and the prompt for each one.
Meelie Gagne

A catalog, a task packet, and a fail-closed ledger, not a project-manager persona
Brian Gagne

Cached input bills at a fraction of the normal rate, but only for a byte-identical prefix. How to order tools, system prompts and agent history so the cache hits on every call.
Brian Gagne

The orchestration patterns behind Cursor's parallel agents and Claude Code's background tasks
Brian Gagne

The delimiter strategy that Claude, GPT, and Gemini all respond to better than plain English sections
Brian Gagne

Model Context Protocol is the new API layer for AI agents. Here's how to design tools that don't confuse them
Brian Gagne

Stop bolting retrieval onto bad prompts. Design prompts that actually use retrieved context correctly
Brian Gagne