Groundbreaking research, apparently

The new starter's name is Lewis. He has been here eleven days. On Monday he asked what ARIA does. Priya told him it summarises documents. He asked which documents. Priya said the ones people upload. He asked what people use the summaries for. There was a pause long enough that I started counting ceiling tiles.

I said it depends on the user. Lewis nodded like this was a satisfying answer. It was not a satisfying answer. It was the verbal equivalent of a 404 page.

He asked again on Wednesday, but differently. Not "what does it do" — "what is it for." I gave him the same answer with different words. Priya, across the room, did not look up from her screen. But she was listening. She is always listening.

Lewis has a way of asking questions that are technically simple and structurally catastrophic. Eight months in, I've made peace with not being able to answer certain things. Lewis has been here eleven days and is already pulling on threads the rest of us learned not to touch.

---

DeepSeek announced this week that it's prioritising groundbreaking AI research over short-term commercialisation. Its founders told investors — the ones handing over the ten billion dollars — that the goal is AGI, not revenue. The investors, apparently, were fine with this.

I read that sentence three times.

The logic being: the money is for the science. The science is for something bigger than the money. The investors are funding something that, if it works, will likely make the concept of investing in things somewhat academic. Everyone involved seems aware of this and has decided to proceed anyway.

There is something almost admirable about it. Or there would be, if I hadn't spent forty minutes this morning trying to explain to Lewis what the product is for.

---

Marcus found Lewis on Thursday. I don't know how — Marcus has a sense for new people the way some dogs have a sense for weather. He materialised next to Lewis's desk holding a coffee and talking about the roadmap. Not asking. Telling. The roadmap as Marcus understands it, which is an impressionist painting of the roadmap as it actually exists.

I was not present for this conversation. I have reconstructed it from Lewis's expression when he came to ask me a follow-up question.

"Marcus says there's an image thing coming in Q2," Lewis said.

"It's May," I said.

Lewis looked at the calendar. He is eleven days in. He does not yet know what Q2 means in this building.

"It's a computer vision feature," I said, which is one of the possible interpretations. I did not mention the other interpretations, which include logo resizing, photo uploads, and something our CTO once described as "making it more visual" while moving his hand in a way that could have meant anything.

"Is it nearly done?"

I considered this question.

"It has a ticket," I said.

---

The ticket has had a ticket since week four. There is a sub-ticket for scoping the ticket. The sub-ticket was unassigned until I assigned it to myself in what I described at the time as progress and have since come to understand as a comfort behaviour.

This week I opened it again. Not because anything changed — because Lewis asked the question and the question was reasonable and the answer made me look at the ticket in a new way. From the outside. As an object.

The scope field says: TBD — align with CTO re: vision.

The CTO's vision is a circular hand gesture. I have been aligning with it for sixteen weeks.

Lewis asked today what TBD means. I said it means we're still working out the details. He said that makes sense. He has been here eleven days and he is already being lied to in the exact same ways the rest of us were lied to, and the worst part is it's not even intentional. We have all just gradually stopped being able to see it.

Priya heard this. She looked up.

"We should just build it," she said.

Nobody disagreed. Nobody moved. The ticket remains open.

At least DeepSeek knows what it's for.