🥀 Why an assistant without rhythm withers
New here? "Rot" here isn’t an empty metaphor: it’s what happens when the context (the files that describe your life, your business, and your preferences—created in Phases 1 and 2) stop being updated while the reality they describe keeps changing. The assistant doesn’t make mistakes because it’s poorly made. It makes mistakes because it’s responding based on a snapshot of a moment that has passed.
Think about what changes in any business or routine over six months: priorities shift, an important client becomes less important and another takes their place, a manual process becomes automated (perhaps thanks to a skill you built yourself in Phase 3), the team grows or shrinks. If the CLAUDE.md and context files still show a snapshot from day one, the assistant will suggest the wrong priorities, forget new people, and keep pushing a process you’ve already dropped.
💡 Main Concept
A digital assistant doesn’t “learn on its own” outside conversations—it only knows what’s written in the context files and what was decided before. Without a review routine, those files freeze, and the assistant works for an outdated version of you.
- •The business changes direction → outdated context delivers the wrong recommendations.
- •Team changes → the assistant mentions or involves people who have left, or ignores those who joined.
📅 Monthly rhythm: the quick review
The monthly review is intentionally short — 15 to 20 minutes, once a month, on a fixed day on the calendar (not “when I remember,” because that never happens). The goal isn’t to rewrite anything, but to ask three questions: what’s changed in the business since the last review, what’s changed on the team, and what’s changed in next week’s priorities. Update the context files in the folder you created in Phase 1 with your answers.
Caption: the monthly review is a short cycle — three questions go in, the context files come out updated, and the clock is already set for next month.
🔍 Inside
- What changed in the business: a new customer, a discontinued product, a changed goal.
- What changed on the team: who joined, who left, who changed roles.
- What changed in the priorities: what was urgent no longer is, or the reverse.
- Communication preferences: adjust the tone, level of detail, and when to let you know before acting—based on what you noticed in use over the past month.
🗓️ Quarterly rhythm: the deeper review
Every three months, the review shifts focus: it’s no longer “what changed?” but “what’s actually working?” Three questions guide this longer conversation (30 to 45 minutes): has the first skill you built in Phase 3 proved its value, or has it become a forgotten piece? Is there a new repetitive process that already warrants a second skill? And are any of the connected MCPs—the tools that give the agent hands (Track 2)—sitting idle, unused, and just taking up space in the configuration?
✓ Proper quarterly review
- ✓You measure whether the Phase 3 skill actually saved time, not just whether it "seems good"
- ✓Unused MCPs are removed or replaced, not left in place out of inertia
- ✓Add new skills only when they’re justified by a real repetitive process
✗ Quarterly review missing
- ✗The skill stays there “because it was hard to build,” even though no one uses it
- ✗MCPs accumulate without anyone knowing which ones are still useful
- ✗The assistant grows in complexity without growing in usefulness
💡 Practical Tip
Add the date of the next quarterly review to your calendar at the end of each review—don’t leave “in three months” floating; turn it into a specific date with a reminder.
📝 Decision log: why record the “why”
New here? One decision log
is a file (or folder of files) that records what was decided and why — not what was done, since the work history already shows that, but the reasoning behind it. Phase 1 already had you create the folder
registro-de-decisões/ within the context of
your assistant. Phase 4 is where this habit really pays off.
Without a record, you and the assistant repeat the same debate every few months: “Why don’t we use that cheaper vendor again?”—a question that was already discussed and dismissed six months ago, but no one remembers why. With one recorded line (“we ruled out vendor X in March because the delivery time made our client SLA impossible”), the assistant can cite the reason right away without reopening the discussion.
The decision in one sentence
What was decided, plainly—“let’s prioritize returning customers over new customers this quarter.”
The why, in two lines
The real reason, not the obvious one — what made you lean one way rather than the other.
The date
Decisions have an expiration date — what made sense in January may no longer make sense in September, and the date helps the assistant (and you) know whether it’s out of date.
🧠 Permanent memory: what sets it apart from a chatbot
This is where the monthly and quarterly rhythms add up to something bigger. Each interview reviewed monthly, each decision recorded, each new skill validated quarterly—none of it disappears after the conversation. It’s all written in the assistant’s context, and the next conversation starts from there, not from zero.
This is exactly what sets an assistant built with the WAT framework apart from a regular chatbot: the chatbot “forgets everything” with each new conversation (module 0.1 already covered the difference between a chatbot and an agent—here it comes up again on a larger scale). Your assistant, by contrast, accumulates knowledge. After a year of monthly and quarterly routines, it knows your business better than a newly hired employee—because everything that happened is recorded somewhere it can read.
📊 The buildup in practice
- Month 1: basic context files, created in Phase 1 — initial snapshot.
- Month 6: 6 monthly reviews + 2 quarterly reviews + a handful of recorded decisions — the assistant already understands recurring patterns.
- Month 12: a complete history of an entire annual cycle — seasonality, decisions revisited and confirmed or reversed, skills that proved their value.
This isn't "AI learning on its own"—it's you keeping the record up to date and the assistant reading that record every time it talks with you.
⚠️ Attention
Permanent memory doesn’t mean an infinite, disorganized file. If the context folder becomes a repository of everything that’s ever been said, it becomes hard to use — the quarterly cadence is also there to remove what’s no longer relevant, not just to add more.
🌀 The 4-phase spiral: why it isn't a circle
The capstone’s four phases—Home (7.1), Life (7.2), Hands (7.3), and Growth (this module)—don’t end here. They repeat, but never the same way twice: each turn starts with richer context than the last, because the rhythm of this phase has fed everything. That’s why the diagram below is a spiral, not a closed circle—the same path, but rising a level with each cycle.
Caption: follow the cyan arrow from the center outward—Casa creates the skeleton, Vida connects to the real world, Mãos delivers the first skill, Crescimento reviews everything and feeds the next, larger round, starting again in Vida with more context than before.
🎯 Main Concept
- •It’s not about “finishing the capstone and moving on”—it’s about entering a cycle that repeats, with the assistant getting sharper each time.
- •Modules 7.5 (backup and version control) and 7.6 (final delivery) wrap up Track 7 by protecting this cycle.
# Goal: ask Claude Code to conduct the monthly review — read the context, ask what changed, update the files
Você é meu assistente executivo. Faça a revisão mensal do meu contexto: 1. Leia todos os arquivos em `contexto/` e `registro-de-decisões/`. 2. Resuma em 5 linhas o que você entende hoje sobre meu negócio, meu time e minhas prioridades atuais. 3. Me pergunte, uma de cada vez: "o que mudou no negócio desde a última revisão?", "o que mudou no time?", "o que mudou nas prioridades para os próximos 30 dias?". 4. Com minhas respostas, proponha as edições nos arquivos de contexto (mostre o antes/depois antes de aplicar). 5. Se eu confirmar, aplique as edições e registre um resumo datado em `registro-de-decisões/<data>-revisao-mensal.md`. <troque "contexto/" e "registro-de-decisões/" pelos caminhos reais da sua pasta se forem diferentes>
How to verify: after running, open the context files and check whether the answers you gave actually became text in them — and whether the decision log got a new entry with today's date. If either is missing, ask the assistant to apply it again before closing the session.
Quick check (optional): What’s the difference between the monthly and quarterly cadence?