Twenty years of never being wrong, and he is right not to trust this
I wrote to a bookkeeper yesterday whose profile leads with two things: twenty years in the work, and zero-error accuracy.
He is the hardest person I could possibly try to sell AI training to.
He is also right.
Before any of that, the practical bit, because it has a date on it. The final free week of live AI training runs 28 September to 2 October, registration closes Friday 25 September, and there is nothing after it.
I am not going to open with a story about how he came round. He has not replied, and he might not. What I want to write about is why his suspicion is correct, because almost everyone selling this pretends it is not, and that is the reason so many capable people have quietly decided this whole category is nonsense.
Start with the part nobody wants to say
These tools will confidently tell you something untrue.
Not occasionally. Not in edge cases. A language model will produce a number, a date, a citation or a flat statement about a rule with exactly the same fluent certainty whether it is right or invented. There is no tremor in its voice when it is wrong. That is the single most important thing to understand about it, and it is the thing most demos are carefully built to avoid showing you.
So when somebody whose whole working identity is built on not making mistakes looks at this and says no thank you, they are not being a dinosaur. They have identified the failure mode in about four seconds, which is faster than most of the people selling it.
Why the usual pitch fails on exactly the wrong people
The standard argument is that it saves time.
Think about who that lands on. It lands on the person for whom speed is the product. It does not land at all on the person whose product is being right, because to them "faster" is not an improvement, it is a risk. Everyone I know who has been burned was burned by something that went out quickly.
You cannot sell speed to somebody who sells accuracy. You have to make a different argument, and the different argument happens to be the true one.
The tools are bad at one job and very good at a completely different one
Here is the distinction that took me four months to say plainly.
These tools are unreliable at being correct. They are excellent at doing the typing.
Those are not the same job, and almost all of the confusion about AI right now comes from treating them as one thing.
Doing the typing means pulling a figure out of a document that was formatted to be hard to read. Turning a messy export into a table. Drafting the same email you have written four hundred times. Producing the first version of a proposal, a policy, a job description, a client update, a month-end narrative. Reading forty pages and finding the six paragraphs that actually matter.
Being correct means deciding whether the figure belongs there. Whether the claim holds up. Whether the explanation is plausible. Whether the answer that came back looks like something you have seen before, or something that should not exist.
The first list is mechanical and enormous and it is eating the evenings of almost everyone I speak to. The second list is judgment, and it is yours.
This is not really about bookkeeping
I should say plainly that I specialise in accounting and the training is not only for accountants.
The week is built for anyone whose work is part typing and part judgment, which is nearly everyone who runs or works in a business. Business owners. Chief executives. Operations people. Office managers. Whoever ends up writing the thing, chasing the thing, formatting the thing and checking the thing, usually all four.
The argument above does not change by role. The owner who knows instantly that a supplier quote is wrong. The ops manager who can tell from the tone of an email that a project is slipping. The long-serving staff member who spots that a number is out by a factor of ten because they have seen the real one a thousand times. Every one of those is the same skill the twenty-year bookkeeper has, and every one of them becomes more valuable in exactly the same way.
If you have been reading AI advice aimed at accountants and thinking it does not apply to you, the advice was narrow. The tool is not.
Which means the sceptic gets more valuable, not less
This is the part I would actually like the twenty-year bookkeeper to hear, and the owner, and the person on their team.
If you take the mechanical half off somebody's desk, the volume of work that can move through them goes up. And the moment volume goes up, the bottleneck moves. It stops being how fast the typing happens and becomes how good the checking is.
Checking is the scarce skill. It always was, but it was hidden underneath hours of keystrokes that disguised it as productivity. Anyone who has spent years building an instinct for what does not look right is holding the exact capability that becomes the constraint. The keystrokes were never the moat. The eye was.
AI does the typing. You make the judgment calls. I have been saying that for months and I mean it far more literally than it sounds.
And the uncomfortable half, because a one-sided argument is not worth reading
If your value genuinely is the typing, this is not good news and I am not going to dress it up.
Whoever is fast and accurate at producing documents and nothing else is in a harder position this year than last year. That is true, it is not nice, and pretending otherwise would insult everyone reading. The honest response is not reassurance. It is that the way out is upward, into the judgment, and that the door is open right now in a way it will not stay open forever.
So here is the invitation, and it is a strange one
This is the FINAL FREE WEEK. Not the last one for now, and not the last before a break. We are stopping the free weeks entirely and there is nothing after this.
September 28 to October 2. Registration closes Friday 25 September. One hour a day, live at 3 PM Central, every session recorded. Hands on, on your own real work, not a demo. No cost, no card, nothing renews. Complete it, live or on the replays with the work done, and you choose any five AI agents to keep.
Come and try to break it.
I mean that as a genuine invitation rather than a rhetorical flourish. Bring the file you think it will get wrong. Bring the job nobody else wants. Bring the thing you are certain a machine cannot handle, and find out precisely where the edge is.
You will get more out of that week than the people who arrive already convinced. The enthusiasts find out what it can do. The sceptics find out what it cannot, which is the more useful half, and the half almost nobody is teaching.
Years of refusing to be wrong is not an obstacle to this. It is the qualification.
If a whole week is genuinely not realistic, claim one free agent instead. Yours to keep either way.
And the community stays free and stays open after the final free week has gone, including for the people who turn up purely to argue.
Yvonne Razo. Founder and CEO, Legacy Small Business Consulting. CFO, Simple Academy. Co-founder, AI for Accounting.
No comments:
Post a Comment