The reply nobody read
How we work with AI at Twipe | Twipe Insights with Joris Gielen

Published: 12 November 2026 | Photo / Video: FGEE
A few weeks ago, a pull request appeared in my review queue. For readers outside the engineering team: a pull request is when a colleague asks you to check their code before it goes live. This one was a minor change – just one extra validation step in our content processing workflow. The code looked fine.
I left one comment: “Have you checked what happens when a customer sends us an article without an author? I want to make sure we don’t scrap the whole issue just because of one missing byline.”
The reply arrived within two minutes and ran to four hundred words. It began by thanking me for the thoughtful question. It summarised my comment for me. It listed three points to consider, each under its own heading. It explained how validation works in general. It concluded by noting that both approaches have their merits and offering to elaborate further. It did not say whether anyone had tried submitting an article without an author.
I have received this reply, or one very much like it, more times this year than I can count, and I suspect most people reading this have too. An LLM wrote it – that much was obvious – and I have no objection to that. My problem was that I couldn’t tell whether anyone had read it before it reached me. It took me thirty seconds to type my question. Answering it properly would have taken the author five minutes. Instead, I spent those minutes reading four hundred words only to find out that my question was still unanswered.
Two problems hidden in one reply
The first problem is trust. Did the author give my question any thought, and did anyone check the answer? When I can’t tell, I stop taking replies at face value and start double-checking things I used to accept. That’s time-consuming, and amongst colleagues it’s damaging.
The second is readability. The reply was long and didn’t commit to anything. Reading it took me longer than it took the sender to write it. Even a perfectly reasoned answer falls flat when it comes across like this.
These are two separate issues that require different solutions. Disclosure sounds like the obvious answer, but it does not solve either problem.
There is no clear line to draw. “Pasted unread” and “iterated ten times, restructured, keeping only what I agree with” both count as using AI. Calling both cases “AI-assisted” does not tell the reader much. A label doesn’t change any decision either: I should be checking the claim regardless of whether it is labelled or not. And it cannot be enforced, because people are poor at spotting text written by machines. A 2023 study published in PNAS puts human accuracy at roughly the same level as a coin toss.
There is one exception. When nobody has read a message before it is sent, there is no author, and the reader deserves to know that. We deal with this through identity rather than etiquette: bots post as bots, from their own accounts. Anything posted under a person’s name belongs to that person.
The principles we keep coming back to
Agentic tools are now part of our day-to-day work at Twipe. This made it urgent to agree on what we expect from one another when using them. These are some of the principles we keep coming back to.
1. A human being is accountable. Always.
Whoever signs the code, the comment or the decision is responsible for it, regardless of the tools used. A named engineer is responsible for every ticket, however much of the work was carried out by an agent. A named individual is accountable for everything we release and every answer we provide.
Ownership of the work does not mean carrying out every single step of it. As agents take on more responsibility, what the human owns is the objective, the boundaries, the evidence that must be in place before something is released, and the moments when a person steps in. Where the work stands at any given moment is a decision, and someone is responsible for that decision too.
2. Do not do to others what you would not want done to you.
It’s the oldest rule in the book, and AI makes it easy to break it by accident. The most common scenario is a reply that nobody would want to sit through, including the person who sent it. So address the main point first and only add context where it helps. Read what you are about to send from the perspective of the person who will receive it. If you cannot put it in your own words, it is not ready.
3. Don’t trust the output.
AI can be wrong, but it is always confidently wrong. Check it as you would check work from a colleague you haven’t met yet. How thoroughly we check depends on the work. A change that has been tested, is visible in production and is easy to roll back can be released after a cursory glance, whoever wrote it. Where the judgement is difficult or a mistake is costly, a person needs to be closely involved. And when a reply feels insubstantial, do not try to guess from the writing style. Ask something specific about the content. If they have engaged with it, a specific follow-up question should be easy to answer.
4. Don’t do yourself what an AI can do 10 times faster.
Judging and deciding remain the preserve of people. Most of the rest – typing first and, increasingly, the steps in between – does not. The time this saves is what makes room for the first three principles.
Where that leaves us
Back to that pull request. The reply I wanted was just two lines.
Yes. A missing author results in an empty byline and a warning in the log. Test added in the same commit.
It takes ten seconds to read, and my question has been closed. The author had probably done exactly that work. The verbose reply obscured it.
Look at what the short reply contains: a test, in the commit, which I can go and read. My question was about one of those instances where a person had to step in. In this case, the step that was missing was simply checking the edge case.
That is what ‘AI-first’ means to us. People used to carry out every step. The task now is to decide what can safely be handed over, what evidence is sufficient to accept it, and where a person still needs to be present. Do that well and speed will follow. Do it badly and you end up with four hundred words that answer nothing.
The principles set out above are not ours to keep. Take them, use them, and if you think any of them are wrong, let us know.
Read the Original text
Joris Gielen is a Software Solutions Architect at Twipe
