The Essay Is Easy Now. Can You Say It Out Loud?

For a long time, a good essay was proof that you understood something. This week, the university regulator in England said, more or less, that this proof is gone.

On Wednesday, the Office for Students published a report together with Advance HE. They spoke with over 300 students and staff from over 120 institutions, and one sentence in it says a lot: "Where assessments are unsupervised, it is frequently possible to complete them to a high standard using AI tools."

Josh Fleming from the OfS called it "a generational risk to the system as it stands" and asked for "far greater use of AI-free assessment". So what is the fix? Closed-book exams, presentations, and oral exams, the good old viva. Universities that can't measure what students really know could even face sanctions.

In other words: write less, talk more. Show us you know it, in a room, with people asking questions.

We have the same problem at work. We just didn't name it yet

Think about what we use as proof at work. The pull request description. The design doc. The weekly status update. The self-review before a promotion.

All of these can now be written very well by someone who doesn't fully understand what they wrote. I don't say this to blame anyone. I use these tools every day. It is simply true, and pretending otherwise helps nobody.

So what still shows real understanding? I keep coming back to one moment I have seen in many design reviews. Someone presents a clean, confident plan. Then a colleague asks "what happens if this service is down for ten minutes?" and the room goes quiet.

The first answer is easy to prepare. The second question is where understanding lives.

Writing shows what you can produce. Talking shows what you actually understand.

Speaking is becoming the proof

My guess is that companies will quietly do what universities are now doing. More walkthroughs instead of long written specs. More "explain it to me" in code reviews. More live conversations in interviews, fewer take-home tasks that anyone can finish with a good prompt.

That is a bit uncomfortable for our profession. Many engineers chose this job partly so the code could do the talking. Now the code can be written by a machine, and the talking is what is left.

But here is the good news. Speaking clearly is a skill, not a talent. Nobody is born explaining a distributed system in two minutes. People get there by doing it badly a few times, then a bit less badly.

Start small, start this week

1. Explain your next pull request out loud before you open it. Two minutes, to a colleague or to an empty chair. If you can't explain why it changes what it changes, you are not ready to ask others to review it.
2. Prepare for the second question, not the first. After you write a doc, list three questions someone will ask, and answer them without notes.
3. Let the AI grill you. When a tool helped you write something, ask it to play the most annoying reviewer in your team and question you about it. Answer by speaking, not typing.
4. If you lead a team, swap one written update for a short walkthrough. Five minutes, once a week. Ask the second question, and ask it kindly. The goal is learning, not catching people.
5. Say yes to small stages. A demo, a lightning talk, a postmortem. Short and frequent beats one big talk a year.

Universities are going back to the oldest exam there is: sit down and explain. I don't see that as a step backwards. I see it as a reminder of what was always the point. Not the paper. The person who can stand behind it.

Sources: Office for Students: Making AI work for students, The Guardian: AI has made it close to impossible to assess university coursework, regulator says, The Decoder: UK regulator threatens universities with sanctions unless they adopt AI-proof exam formats