Ahead of the Technology

Got something to say to this podcast? Send a voicemail, and you might hear yourself on the next episode!

Every time an AI assistant “remembers” something or finds the right answer in a sea of documents, a quiet piece of technology is doing the heavy lifting: the vector database.

In this episode, I unpack what’s actually going on under the hood. How words, images and ideas become numbers, how a machine measures “closeness” in meaning, and how it finds the right match among millions of options in milliseconds without checking every single one. No PhD required, just simple stories and intuition.

I’ve led teams that built vector search engines and AI products that lean on them heavily, so this is the version I wish someone had told me at the start: what it is, the maths that makes it work, and why it matters for every AI product you’ll use next.

🎧 Read the interactive version on https://vithushansylvester.com/writing/vector-search-under-the-hood Thinking ahead of the technology.

Got something to say to this podcast? Send a voicemail, and you might hear yourself on the next episode!

On Thursday 5 November 2026 the Nasdaq fell 5.5% and chip stocks about 11%, the worst day since the tariff crash of April 2025. Roughly $3 trillion left US stocks in a single session.

I recorded this the same evening, from my seat as someone who builds AI products. I walk through why local votes against data centres set off the sell-off, why the market was already fragile after a month of negative cash flow and expensive credit, what the demand for AI still looks like, and the best, middle and worst cases from here with my odds. I finish with the four decisions I'm taking into Monday.

Numbers are as of Thursday's close, and a few were still being confirmed when I recorded.

Full article with charts and every source: https://vithushansylvester.com/writing/the-3-trillion-thursday

Vithushan Sylvester · Co-founder and CTO, IdeaBoxAI · Thinking ahead of the technology

Got something to say to this podcast? Send a voicemail, and you might hear yourself on the next episode!

On 1 October 2026, arXiv capped its authors at two papers a month after a record 40,363 submissions in September. It looks like proof that AI has sent research output through the roof. The data says something narrower.

In this episode I go through ten years of publishing data by year, field and country:

• Why world journal output grew faster before ChatGPT than after it
• How China and India added more papers between 2021 and 2024 than the whole world did
• Why mathematics, not computer science, shows the sharpest AI break
• How the famous “50% more papers” estimate shrank to 7–25% under stricter methods
• What heavy AI writing is doing to peer review, and why light use carries no visible penalty

My take: AI made papers cheap. It did not make discoveries cheap. The bottleneck in research has moved from writing to verifying.

Read the full article, with charts and all 24 sources: https://vithushansylvester.com/writing/did-ai-speed-up-science

Got something to say to this podcast? Send a voicemail, and you might hear yourself on the next episode!

US tech employment has been falling for almost four years while the rest of the economy kept adding jobs. The easy explanation is AI, but the timing doesn’t fit: demand for software developers peaked in February 2022, nine months before ChatGPT launched.

In this episode, I go through seven years of data from the US, Europe, India and China. Interest rates started the decline, AI is why the recovery never came, and the cost is landing on the youngest workers in every market we can measure. I also cover who is still getting hired, why 71% of the rebound in US developer postings went to senior roles, and what this means for companies, engineers and graduates.

Every chart and source is in the full article: https://vithushansylvester.com/writing/the-real-impact-of-ai-on-tech-jobs

Got something to say to this podcast? Send a voicemail, and you might hear yourself on the next episode!

AI now beats the best human teams at programming contests, yet employers are still setting the same puzzles. LeetCode-style questions appear in about 30% of technical interview reports, the same as 18 months ago.

In this episode, I look at what the data says about the coding interview, what companies like Google, Canva, Meta and Shopify are changing, and why the new rounds all test judgment. I also explain why that judgment pays most close to revenue, and why being a product manager isn’t enough unless you can also build.

Every chart and source is in the full article: https://vithushansylvester.com/writing/is-the-coding-interview-dead

The link will only work once the article is live at that address, so publish it first or change the link.