The art of the second act
Hi everyone, I’m writing this from the lovely city of Bristol where I’m spending a few days. Before we jump into this week’s newsletter, I’d like to thank Box for supporting Hypertext as its very first sponsor. Feel free to share this email with a friend, it really helps 💌 Sign up here.
Why AI is pushing startups to launch second products
Many years ago, a startup founder and a good friend of mine told me that it’s much easier to launch a second product when you already have one successful product. You’re not starting from scratch as you already have an audience. The second product can also start as an add-on and branch out to become its own standalone product.
In 2026, AI has made second-product launches more tempting because incumbents already have workflows, users, trust, data, and distribution. Launching a second product seems inevitable for a variety of reasons:
- Successful Software-as-a-Service (SaaS) companies are suddenly under pressure. They are generating revenue but many of them are on the verge of accelerated obsolescence. They need… something else.
- Launching a second product is easier than ever due to agentic coding. If you have an existing code base that you can leverage without having to redefine data formats, user accounts and all those pesky things, you can speedrun your way to a good minimum viable product.
- And the most important part: distribution is becoming more important than ever. If several companies can build the same product at the same time, your ability to stand out from the crowd is extremely important.
On that last point, let’s take voice transcription apps as an example. Many of you probably know Wispr Flow. On your computer or your phone, you press a button, talk naturally, and text magically appears on your screen with an unprecedented level of accuracy. The company is reportedly raising at a $2 billion valuation.
But there’s nothing magical behind Wispr Flow as a product. There are many excellent speech-to-text AI models out there (OpenAI’s Whisper models, Nvidia’s Parakeet models, more recently Cohere Transcribe).
That’s why there are many Wispr Flow alternatives: Willow Voice (a Y Combinator copycat), Monologue (from Every), a bunch of bootstrapped companies (Aqua Voice, Superwhisper, VoiceInk…). Even OpenAI created its own dictation feature that you can use in any app. It is currently an optional feature in Codex that you can activate it in the app settings.
Wispr Flow is a reminder that the hard part is not always building the underlying technology. The hard part is packaging it, distributing it, and becoming the default. And they’ve excelled at that.
Coming back to second products, two European unicorns have turned to this diversification strategy. Deel, the company that helps you hire remote workers without opening a subsidiary, launched Akai.
It started as an internal tool to create AI agents that connect with legacy banking systems. Many employees at Deel currently spend countless hours submitting payments across multiple banking portals, reconciling payments and more.
Akai captures what Deel’s employees do in a web browser and builds universal connectors so that it can do the same tasks automatically the next time. And as Deel handles sensitive data and large sums of money, Akai has been designed with compliance and audit trails in mind.
In other words, Deel is not just automating its own back office. It is turning that automation layer into a product.
Revolut is also branching out beyond its banking roots with GlobalHire, its own take on remote hiring and payroll. Instead of working with Deel, Rippling or Remote.com, Revolut has recreated its own alternative to Deel for its global workforce.
In other words, Revolut is not just solving an internal HR problem. It is turning that operational need into another product surface.
When you’re an executive at a large tech company, you often ask yourself whether you should build or buy a product. Maybe the question now is: build, buy or build and sell.
Huge amounts of business content remain ungoverned. Box brings structure, security, and control to your unstructured data, enabling secure, enterprise-ready AI.
The case for European companies buying Chinese AI startups
A few weeks ago, Chinese authorities decided to block Meta’s acquisition of AI startup Manus. A lot of articles mention Manus like it’s a mainstream product but maybe it’s worth spending a minute talking about what Manus actually does, right?
Originally developed in China, Manus is an AI agent workspace that can handle complex, multi-step tasks and create presentations, design websites or build a research report. It’s a bit like OpenClaw, but they come from opposite philosophies.
OpenClaw is a local-first AI agent framework. It runs on your own Mac Mini and interacts with your local apps to do things that you simply can’t do with ChatGPT. Manus also lets you create things that are impossible to make with ChatGPT, but through a highly-polished cloud-first interface.
As Meta was about to acquire Manus, the Chinese company moved to Singapore. And it wasn’t just a headquarters change. Around 100 employees literally moved to Singapore. And yet, that wasn’t enough to stop China from cracking down on Meta’s acquisition. Singapore-washing didn’t work.
And it brings me to this important geopolitical conclusion: maybe it’s time for Europe to take advantage of the ice-cold relationships between the U.S. and China. It’s time to start some M&A action with some promising AI companies based in China. Would Beijing react differently if the buyer were European rather than American? Probably.
Sure, SAP is investing in n8n and acquiring Prior Labs, two extremely promising German AI startups. But European incumbents should also look at the other side of the Eurasian continent. If U.S. buyers are becoming politically toxic for Chinese AI companies, there might be a narrow window for European companies to invest in, partner with, or acquire the next Manus.
A few words on layoffs
Unfortunately, these past few days, Cloudflare, DeepL and GitLab all announced massive layoffs. These companies are in very different positions and aren’t cutting jobs for the same reason.
GitLab and DeepL are both streamlining their organizations, reducing costs, refocusing, etc. They are also both competing with products that are gaining steam with the current AI wave: GitHub (and Hugging Face) for GitLab, and ChatGPT (and Claude) for DeepL. Most people around me switched from DeepL to using ChatGPT to translate big walls of text, even though DeepL is clearly the superior product for that task.
As for Cloudflare, it is investing more in AI tokens (+600% in 3 months) and less in people (-1,100 people). It is one of the first companies to admit publicly that AI spend isn’t just software spend. It’s an operational expense, just like salaries. It’s hard to predict exactly if this will keep happening if AI labs raise token prices to stop burning cash.
So when an HR person next asks you what’s your expected salary range, maybe you should ask whether they want an answer in euros or tokens.
Have a good day ☀️
Romain