Allbirds goes full AI pivot, Snap lays off 1,000, and Adobe rewrites creative editing

Something quietly broke in the "AI pivot" playbook this week - and now it's impossible to ignore. A company that once sold wool sneakers at a $4 billion valuation just rebranded as a GPU cloud provider and watched its stock jump nearly 600% in a single day. Meanwhile, Snap is laying off a thousand people and pointing directly at AI as the reason. These two stories aren't unrelated - they're the same story told from opposite ends.

DeepL launched its voice translation feature this week - but how many languages does DeepL currently support for text translation?

The answer is hiding near the bottom of today's newsletter... keep scrolling. 👇

Allbirds - yes, the wool sneaker people - officially announced it is abandoning footwear entirely and rebranding as NewBird AI, a GPU-as-a-Service company targeting AI compute demand. The company secured a $50 million convertible financing facility and its stock shot up 582% in a single trading session. The shoe business itself was sold off separately for $39 million to American Exchange.

This is either a genius use of a public shell listing or the most transparent example of AI hype-chasing you'll see this year. Allbirds had never turned a profit since its 2021 IPO, and sales had dropped nearly 50% between 2022 and 2025. The AI infrastructure demand is real - but whether a failed sneaker brand can compete with established GPU cloud providers is a very different question. Still, the market loved it.

Read the full story →

Snapchat's parent company announced it's cutting 16% of its workforce - roughly 1,000 people - citing "rapid advancements in artificial intelligence" as a core driver of the decision. The cuts were announced in an internal memo this week and come amid a declining stock price and pressure from an activist investor.

This is a pattern worth watching closely. When companies blame AI for layoffs, it's doing two things at once: justifying headcount reductions to investors while signaling confidence in automation to the market. LinkedIn data, interestingly, shows hiring is down 20% since 2022 - but attributes most of that to higher interest rates rather than AI replacement. The real picture is messier. If you're tracking the future of work closely, Snap's move is a signal, not an anomaly.

Read the full story →

Adobe just announced a new Firefly AI Assistant that works across its entire Creative Cloud suite - Photoshop, Premiere, Lightroom, Illustrator, and more. Instead of clicking through menus, you describe what you want in plain language and the assistant executes across apps. Adobe is calling it a "fundamental shift" in how creative work gets done.

The practical implication here is significant. You won't need to know the difference between a curves adjustment and a levels adjustment - you just say "make the shadows warmer" and it handles it. For non-technical creators, this genuinely lowers the barrier to professional-quality editing. For experienced editors, it's a speed layer on top of tools they already know. Adobe is betting the future of creative tools looks less like a toolbar and more like a conversation.

Speaking of building things fast - if you want to spin up a website without touching a single line of code, 60sec.site lets you build and launch an AI-powered site in under a minute. Worth a bookmark.

Read the full story →

DeepL - best known for its highly accurate text translation - announced it's moving into real-time voice translation. The company says the technology is designed to integrate with meeting tools like Zoom and Microsoft Teams, handling spoken language in real time. This is a direct move into territory dominated by tools like Google Translate's live feature and Microsoft's real-time captions.

DeepL has built a strong reputation for outperforming Google Translate on nuance and accuracy for written text - especially in European languages. Bringing that quality to real-time voice is a harder problem, but if they can pull it off, there's a real enterprise market here. Multinational teams running calls across language barriers would pay well for something that actually sounds natural rather than robotic. The voice AI space is getting very crowded, very fast.

Read the full story →

The Federal Court of Australia issued new guidance this week warning the legal profession about the dangers of using generative AI in court proceedings. The guidance "embraces" the use of technology but flags that lawyers who "mislead the court" with AI-generated errors could face penalties. It's one of the more concrete institutional responses to the hallucination problem in a high-stakes professional context.

We've seen this play out badly before - lawyers citing non-existent case law generated by ChatGPT, resulting in sanctions and embarrassment. Australia's court system is essentially telling its legal profession: use AI if you want, but you own the output. That's a reasonable standard, but it raises real questions about how much due diligence is realistic when AI tools are producing confident-sounding wrong answers. The legal technology space is navigating this tension in real time.

Read the full story →

Researchers at UCSD and Together AI introduced Parcae, a new looped language model architecture that reportedly achieves the quality of a transformer twice its size. The key idea: instead of scaling by adding more parameters, Parcae routes information through the same layers multiple times - essentially getting more "thinking" out of less compute. The architecture is also described as stable, which has historically been a challenge for looped model designs.

This matters more than the headline suggests. As AI inference costs remain a major constraint for deployment - especially at the edge and on mobile devices - architectures that deliver GPT-4-class quality at half the parameter count are genuinely valuable. If Parcae's efficiency gains hold up under scrutiny, it's the kind of AI research that quietly reshapes what's possible outside of hyperscaler data centers. Check our token calculator if you're estimating inference costs for your own projects.

Read the full story →

The answer is 74 languages! DeepL currently supports over 30 writing systems and 74 languages for text translation - a long way from its European-language origins. Its voice translation expansion will need to cover a meaningful slice of that library to compete seriously with the big players.

The Allbirds story is either brilliant opportunism or peak AI bubble behavior - and I genuinely can't decide. What do you think: is the "any company can pivot to AI" moment a sign of a healthy ecosystem or a bubble about to pop? Hit reply and tell me your take - I read every single response and will share the best ones next week.

That's all for today - catch more daily AI coverage at Daily Inference. See you tomorrow!