QUANTUM BEAT | June 8, 2026
SIRI FINALLY WAKES UP: APPLE BETS THE KEYNOTE ON A GOOGLE-POWERED SOUL TRANSPLANT FOR ITS LONG-SUFFERING VOICE ASSISTANT
Apple’s annual developer conference, WWDC 2026, kicked off today with what may be the most consequential keynote the company has held since the original iPhone. Tim Cook walked on stage this morning in Cupertino and did what Apple has been promising and not delivering for about two years: showed the world a version of Siri that can actually hold a conversation without forgetting what you said three seconds ago.
Here is what Apple actually announced. Siri is getting a complete redesign and its own standalone app. You can now pick your AI model the way you pick a coffee order: Google Gemini is the default, but ChatGPT and Anthropic’s Claude are also on the menu. This is not a small thing. Apple, a company that once threatened to remove apps from its store for making breathing a feature, is now handing users a choice of AI engines and saying go ahead, pick the one you like. That tells you everything about how intense the competitive pressure has gotten.
The new Siri is a proper chatbot with a dedicated chat interface, context awareness across apps, the ability to see what is on your screen and act on it, and a Dynamic Island integration that looks genuinely nice in the demos. iOS 27 and macOS 27 are the packaging. The operating system updates include AI photo editing tools, improved autocorrect that has apparently gotten less infuriating, and natural language shortcuts so you can create automations by just describing what you want in plain English.
Now here is the thing. Apple has been here before. Apple Intelligence launched in 2024 with a promotional campaign that would make a pharmaceutical ad blush, and then proceeded to disappoint pretty much everyone who actually turned it on. Siri got marginally better. Notification summaries were embarrassing enough that Apple had to pull them after they started mangling news headlines in genuinely dangerous ways. The gap between keynote Apple and real-world Apple is well documented at this point.
So the question is whether this time is different. And there are reasons to think it might be. Licensing Gemini’s 1.2-trillion-parameter model at reportedly a billion dollars a year is not a nothing gesture. That is real money for real capability. The move to let users choose their AI model suggests Apple has accepted that it cannot win the foundation model race and has decided to be the platform instead, which is actually a smart play. Platforms usually outlast individual models.
What Apple is betting on is that people will stay on iPhone not because Siri is the best AI, but because the entire experience around it, the privacy, the hardware, the ecosystem, is still worth paying for. The AI becomes the new camera. Not every iPhone user needs the world’s greatest model. They just need one that does not embarrass them in front of their colleagues. Whether Apple can actually deliver on that this time, rather than announce it and then apologize six months later, is the only question that matters.
Source: Bloomberg Technology
THE AI BOT THAT WRITES 90% OF ITS OWN COMPANY’S CODE JUST RAISED $1 BILLION AT A $26 BILLION VALUATION AND ITS CUSTOMERS INCLUDE NASA AND GOLDMAN SACHS
Let’s talk about the most recursive thing happening in tech right now. A company called Cognition makes an AI coding agent named Devin. Devin writes code. And 90% of Cognition’s own internal code is now written by Devin. The company that built the robot that replaces software engineers is running almost entirely on the robot. If you are a software engineer reading this, this is probably a good moment to take a long walk.
Cognition just closed a Series D at $1 billion raised and a $26 billion valuation, more than doubling its worth since September 2025 when it raised at $10.2 billion. The round was co-led by Lux Capital, General Catalyst, and 8VC, with Peter Thiel’s Founders Fund throwing in. The company has now raised more than $2.5 billion in total funding. Paying customers include Goldman Sachs, Mercedes-Benz, NASA, Santander, and several parts of the US government.
The revenue numbers are the jaw-dropper. In May 2025, Cognition was doing $37 million in annualised revenue. Today it is $492 million. That is a 13-fold increase in twelve months. The company says it aims to cross $1 billion in annual revenue before the end of 2026. Many enterprise software companies spend a decade trying to hit that number. Cognition is doing it with a product that has been commercially available for less than three years, from a company that went from a viral demo to a $26 billion business in the time it takes most startups to close a Series A.
Devin is not a code completion tool. It does not sit next to a developer and suggest the next line. Devin takes a description of a task and does the whole thing: plans it, writes it, debugs it, deploys it. CEO Scott Wu describes Cognition’s value as being in the agent layer, the intelligence that sits on top of foundation models and orchestrates their outputs into finished, working software. Cognition routes customers to whichever model produces the best results for their specific task, using a mix of its own proprietary models alongside those from OpenAI and Anthropic.
The competitive situation is genuinely interesting. OpenAI and Anthropic are Cognition’s model suppliers and its direct competitors simultaneously. Both companies are building their own coding agent products. Wu’s bet is that the agent layer captures durable value while the model layer commoditises over time. It is a defensible thesis. It is also a bet that two of the most aggressive and well-funded AI companies in existence will not eventually figure out how to build what Devin does.
Cognition also acquired the remaining assets of Windsurf last year after Google paid $2.4 billion for Windsurf’s top engineering talent. Cognition picked up the technology, customers, and employees who did not want to join Google. The fact that a company valued at $26 billion could be described as picking up the scraps of a deal Google did tells you something important about the scale of money flowing through AI coding right now. The software engineers who are not worried about Cognition are not paying close enough attention.
Source: TechCrunch
MILLIONS ARE FLEEING GOOGLE SEARCH AND DUCKDUCKGO IS HAVING THE BEST MONTH IN ITS ELEVEN-YEAR HISTORY
Google spent years making Search better. Then it decided to make it AI. And a lot of people apparently did not ask for that.
DuckDuckGo just reported its biggest install surge in company history. US app installs went up 18.1% week-over-week on average during the last week of May, peaking at 30.5% on a single day. On iOS specifically, growth hit 33% week-over-week on average and peaked at 69.9% in one 24-hour period. Visits to DuckDuckGo’s dedicated “No AI” search page tripled after Google’s I/O 2026 announcements and kept climbing from there.
Here is what is happening. Google I/O 2026 was essentially Google announcing that Search, the product billions of people use to find information, is now going to be rebuilt around AI. AI-generated summaries appear at the top of every query. An AI interprets what you are looking for and tells you the answer. The links are still there, but they are no longer the point. Google’s position is that this is better. The position of a significant and apparently growing number of users is that they would like to see the links and make their own determination about what the links mean.
DuckDuckGo positioned itself for exactly this moment. The company launched a “No AI search” page, which is as straightforward as it sounds: you type a query, you get a list of links, no AI summary in between. Traffic to that page more than tripled in a week. The main app surged alongside it. DuckDuckGo CEO Gabriel Weinberg said users were being “force-fed” AI with no way to opt out, which is the kind of framing that lands differently when the install numbers back it up.
Now, some perspective is warranted. DuckDuckGo has roughly 2% of the US search market. Google has something north of 85%. A 30% spike in DuckDuckGo installs is meaningful but does not represent an existential threat to a company doing 8.5 billion searches a day. The math does not work like that yet.
But the direction matters more than the absolute number. Every major platform shift in tech history started with a small group of users who wanted something different. The personal computer started with hobbyists. The web browser started with academics. Mobile disrupted desktop not because everyone switched at once but because a critical mass of users found it more useful for enough things. DuckDuckGo does not need to replace Google in 2026. It needs to keep capturing users who genuinely prefer a different approach and keep existing while the AI search debate plays out.
Google’s bet is that people will adapt to AI search the way they adapted to algorithmic feeds and personalised results. They probably will, for the most part. But the history of consumer tech is full of companies that were right about most users and still lost meaningful share to products built for the minority that wanted something different. Ask Myspace. Ask MapQuest.
Source: TechCrunch
ELEVENLABS DROPS AN AI MUSIC MODEL THAT CAN SWITCH FROM OPERA TO HEAVY METAL MID-SONG AND EVERY NOTE IS LEGALLY CLEAN
AI music is one of those topics that makes musicians furious and makes everyone else quietly fascinated. ElevenLabs, the company best known for building AI voices that sound disturbingly real, has now dropped Music v2, and it does something genuinely new: it can switch genres in the middle of a track without the whole thing collapsing into noise.
That sounds like a party trick until you think about what it actually enables. You are working on a film score. The opening is quiet and orchestral. It builds, shifts, becomes industrial and heavy, then resolves back into strings. Traditionally, that requires either a composer working for several weeks or stitching together separately generated pieces and hoping the transitions hold. Music v2 handles that in a single generation pass.
The model can go from opera to heavy metal and back, deliver fast rap without losing lyrical coherence, and layer non-musical sound effects directly into a track. Artists can rebuild specific sections without touching the rest of the song: generate the intro, the verse, the chorus separately and stitch them together, or flag one part that is not working and regenerate just that segment. It is a workflow tool as much as a generation tool.
Now here is the genuinely significant part. ElevenLabs trained Music v2 exclusively on licensed music. Not scraped music. Not music pulled from the internet and justified through creative interpretations of fair use. They signed actual deals with Merlin, which represents a massive portion of the world’s independent record labels, and Kobalt Music Group, the largest independent music publisher on the planet. The output is cleared for commercial use. You can put it in a video, a podcast, an advertisement, a game, a film, and no copyright claim arrives six months later demanding you take it down.
For context, this is a meaningfully different approach than what Suno and Udio took when they launched. Those companies trained on essentially everything and are currently in litigation with major labels over exactly that decision. ElevenLabs looked at that situation and built the licensed version instead. It costs more upfront in licensing fees. It creates a product that businesses can actually use without legal exposure.
The music industry is going to keep fighting AI in court. Some of those fights it will win. The EU AI Act’s new Code of Practice, finalized last week, requires transparency about training data for AI systems, which is going to make the “we trained on everything and hoped for the best” approach increasingly uncomfortable legally. ElevenLabs is building for the version of the market where licensing is table stakes and quality is the differentiator. The musicians who signed deals with Merlin and Kobalt get paid when their music is used. Music v2 is available now in ElevenMusic on iOS and ElevenCreative, with API access coming soon.
Source: TechCrunch