A first-year audit junior spends most of their first year doing something very specific. They take last year’s file, update it, tie the trial balance to the schedules, request documentation, complete the lead schedules, run the tests the programme calls for, and send it to a senior. The senior reviews and writes notes. The junior reworks. The manager reviews. The partner signs. This process is the production line of the modern firm — it generates most of the revenue, absorbs most of the labour, and trains the future partners. It is also, though the firm has never described it this way, a very clear example of a specs-and-tests system.
The thing the firm has been doing all along
In software, a specs-and-tests system is one where a human writes the specification and the tests that define success, and something else generates the code that passes them. The reviewer’s job is not to write the code — it is to write good specs and tests and verify the result. The audit programme is a spec. The materiality assessment is a spec. The substantive procedures are tests. The review notes are quality gates. The junior is the generator. The firm has been running a specs-and-tests factory for as long as the modern audit has existed. The only thing changing is the cost and capability of the generator.
What the factory actually looks like
A serious AI-assisted working-papers process does not replace the senior, manager or partner — it replaces the file-preparation step. A human writes the spec (the programme, materiality, risk assessment), most of which already exists in the engagement planning. The system prepares the file: pulls the trial balance, runs the lead schedules, ties the numbers, generates the substantive tests, writes first-pass findings and flags exceptions. A reviewer then looks at the output — not line by line as they would a junior’s file, but the way they’d review a trusted contractor’s work: looking for the specific kinds of mistakes the system makes and applying judgement to the marginal calls. This is not science fiction; variants run today in firms doing the work to build it.
The career pyramid, inverted
Here it gets uncomfortable. The current pyramid assumes you become a senior by preparing faster, a manager by reviewing better, and a partner by handling relationships. The factory inverts the bottom of it. Preparation is no longer the entry-level work — it’s automated. The new senior spends two years reviewing AI-prepared files, learning to spot the errors a generative system makes. The new manager graduates by getting better at writing specs — designing programmes, setting materiality, identifying risk. The partner’s role is roughly unchanged. The pyramid isn’t flatter; it’s reshaped. There is no soft version of this: firms that take it seriously will end up with fewer juniors, more senior-grade reviewers, and different economics.
Drowning in manual, repetitive work? Tell us the task and we’ll show you what to automate.
The training problem, partially answered
If juniors aren’t preparing files, what are they learning by? A junior who spends a year reviewing AI-prepared files is learning to recognise rather than to prepare — and seeing twenty or thirty times the volume a traditional junior would. It’s not the same skill, and it may be a worse route to becoming a preparer — but preparing is no longer the job. The syllabuses and training-contract structures will need to catch up; the bodies are moving slowly, and firms ahead of the regulators will need to be deliberate about filling the gap.
The economic shock
Compliance work — the bulk of mid-tier revenue — will get significantly cheaper to produce. Early movers briefly enjoy expanded margins; later movers compete on price against firms whose cost base is half theirs. Compliance pricing and margins compress. This is arithmetic, not prediction. The compensating move isn’t cheaper compliance — it’s using the freed-up partner and manager capacity for more advisory, strategic, relationship-led work. Compliance becomes the entry point; advisory becomes the margin. That is the real strategic implication of the working-papers factory.
Part 5 of The Self-Improving Firm. Daniel Lawrence is the CEO and co-founder of Bots For That.
