On 20 October, Ajay Gopal and Prof Benjamin Lee invited me to speak to their ACCT437 class at SMU, AI Literacy for Accounting Professionals. The talk was called "Know what good looks like", and I was surprised that the AI that drew my slides even knows what a viral meme from GENER8ION - STORM II looks like.
The first half was about building with AI and I narrated the story in the two products we built: our FS compilation tool and Frank, the agent that "brands" our finalised financial statements for signing. The second half was about the elephant in the room. These students are about to join a profession where AI already beats juniors at the entry level. Not only that, PM Wong and President Tharman have both warned that the bottom rung of every knowledge work career ladder seems to be breaking as more firms turn to AI for entry level work.
A firm that lets AI do all the junior work and stops training its juniors is eating its seed corn. Seed corn is the grain you keep back to plant next year. Eat it, and there's no harvest.
It's a tempting meal, and hungry farmers had been guilty of it long before AI came along. AI just makes the modern-day version cheaper, at least, prima facie. The entry level work is exactly what AI does well, and juniors take time and money to train. But I hope I demonstrated to everyone how knowing what good looks like comes by doing that entry level work first, and getting it wrong in front of someone who knew better.
What if it proves too hard to find a company that won't eat them? My suggestion is to plant yourself.
Which raises a fair question: what about Echtual?
Here, we are all for AI to take the tedious parts, like documentation and data fill. Our people still do the work that builds our judgment and expertise: the vouching, which reporting standards to apply, and the conversations with the owner behind them. When we hire, we care more that you know your double entry than how well you prompt. However, if you've built something useful on your own, we'll be delighted to hear about it!
And whatever the AI produces, someone qualified reviews it before it goes anywhere. One student asked whether logging your disagreements with the AI applies to audit work too.
It does.
ACRA's Audit Practice Guidance No. 1 of 2026, issued in October, expects audit documentation to show how AI-assisted outputs were reviewed and, where appropriate, the rationale for accepting or rejecting them.
Because a tool is only as good as the person who can tell when it's wrong, and that knowledge has to come from somewhere.
The world at large may have already started eating the seed corn. We'd rather keep planting.