Example: Pyotr, an accountant, completes the full Volume 1 workbook
Pyotr is a composite character and the figures are illustrative. This example shows the workbook filled in by someone from the profession people name first when they list jobs "under threat". It is the second full pass through the Volume 1 workbook, alongside Marina the marketer: her 🟢 share came out around a third, Pyotr's above half. The difference is telling; the conclusion is not.
Pyotr is 41 and the chief accountant at a manufacturer with about 120 employees. Source documents, payroll, month-end close, reporting, the bank and the auditor. He has tried AI: asking it to explain a letter from the tax authority, asking it to draft a reply. His concern is not abstract: "people write that my profession goes first, and I cannot honestly say that is nonsense."
Step 1 — Value lever (chapter 2)
- The main lever in my work: cost saving (lever ②) — and it is unpleasant to realise, because that is the lever getting cheap fastest.
- What is getting cheaper (output): posting source documents, reconciliations, preparing routine filings, drafting replies.
- What stays mine (outcome): the decision in an ambiguous case, and the signature under it. When a transaction is unusual, the data is incomplete and the treatment is arguable, I am accountable — not a program.
Why this is an honest answer: Pyotr did not write "my lever is accuracy". Accuracy is a requirement on the output, and the machine now delivers it too. The lever is what the company pays him specifically for.
Step 2 — The week in baskets (chapter 3)
He listed 16 real tasks from last week, taken from his calendar and inbox.
- 🟢 "AI on its own": about 55% of his time. More than half — and it is the most honest line he wrote.
- 🟡 "human + AI": drafting notes to the accounts, working through counterparty letters.
- The main 🔴 task: deciding on a disputed expense the tax authority may reject — and the conversation with the director about what that would mean.
- The uncomfortable finding: 4 of 16 tasks were busywork with no lever. Moving data by hand between systems that "we will integrate one day".
Why this is an honest answer: the temptation was to file reconciliations under 🔴 — "that takes experience". He filed them under 🟢 and noted separately: the experience is not needed for the reconciliation, it is needed for what to do about a discrepancy.
Step 3 — The ladder (chapter 4)
- Entry / middle / top: the top of the function — but the ladder underneath is crumbling. A junior used to spend a year posting documents and build a feel for the business that way. The system posts them now, and there is nothing left to build it on.
- One nearby foothold: decisions on provisions and estimates run through him and then into management reporting. That is a step outside bookkeeping.
Step 4 — Level of usefulness and the +1 move (chapter 5)
- Default level: 3 (orchestrator) during month-end close, 1–2 in everything else.
- Candidate task for +1: a monthly note on "where we lose money on tax and why". Nobody produces it today. Not "the filing is done" but "here are three decisions and what each risks".
Step 5 — Your personal system (chapter 6)
- My weakest layer: memory. Every time I explain our accounting policy to the model again, what our contracts look like, and where the sales team usually gets it wrong.
- What AI already does: draft replies, explaining other people's wording, searching regulation — always checked against the source.
What Pyotr did within a week: started one file, "how things work here", and now hands it over with every question. That is the first step towards a skill of his own.
Step 6 — One skill worth more (chapter 7)
- The skill I train deliberately: explanation. Not "I calculated it" but "I explained to the director what the number means and what we do now." Once a month, a ten-minute verbal read of the accounts, no slides.
Why that one: the machine takes over assembling the data; it does not take over explaining and handling the exceptions. That is the common thread in current accounts of the profession, and it matches what people already thank Pyotr for.
Step 7 — Working mode with AI (chapter 8)
- My mode: centaur, but at risk of sliding. The model's answers on regulation look convincing, and checking them feels like effort.
- The rule I introduced: any reference to a rule gets opened in the source. No exceptions. The model's wording can be used; its citation cannot.
Step 8 — Diagnosis (chapter 9)
Half the week is already reproducible. What is hard to replace rests on three things: the decision in a disputed case, the signature under it, and explaining the consequences to the person deciding about money.
Step 9 — The weak multiplier (chapter 11)
- Where the zero is: transferability. It all rests on Pyotr personally being at the desk. If he takes leave, what stops is not the bookkeeping but the understanding of why it is set up this way.
Step 10 — The 90-day plan
- Days 1–30: the "how things work here" file — accounting policy, standard contracts, frequent mistakes. Hand it to the model with every question.
- Days 31–60: the first "where we lose money on tax" note — three decisions with the risk on each. Show the director; ask which of it is useful.
- Days 61–90: hand over document posting and reconciliations entirely, keep the discrepancies. The check: how many discrepancies I explained, not how many lines I matched.
What is worth noticing
Pyotr's 🟢 share is higher than Marina's — and that is a map, not a sentence. The move he chose was not "get faster at 🟢" but shift the weight to where the price stays: exceptions, the decision, the explanation.
That shift — from assembling data to handling exceptions and explaining them — runs through current accounts of the profession. It is the same logic behind the "what gets cheaper / what gets more valuable" line in the three value levers.