Image via Unknown
Wednesday, January 21, 2026
OpenAI's hardware device is finally coming
Hey builders, today brings us more AI news. OpenAI's doubling down on enterprise with GPT-5.2 powering ServiceNow workflows at scale (yikes, that's a lot of automation), while DeepSeek's Engram is quietly solving the memory problem that's been haunting LLMs forever. Oh, and OpenAI casually confirmed their first hardware device drops by end of 2026 (details mysteriously TBA). Meanwhile, humans& just raised $480M betting that AI without the human touch is basically useless, which tracks with today's wildest stat: 74% of companies still haven't unlocked real AI value. Turns out the problem was never the models, it was always the people. So here's the real question: Are you building AI that solves problems, or just automating stuff because you can?
Image via Unknown
Top Stories
OpenAI's models will power ServiceNow's enterprise automation platform, enabling AI-driven workflows across 80+ billion annual processes with native multimodal capabilities and end-to-end agentic automation for Fortune 500 companies.
DeepSeek AI
DeepSeek's Engram introduces conditional memory as a sparsity primitive for LLMs, complementing MoE with efficient O(1) lookup that unexpectedly boosts reasoning and long-context performance while reducing computational overhead.
OpenAI officially confirmed it will launch its first hardware device by end of 2026, though the specific product type remains a mystery as the company experiments with multiple prototype categories ranging from wearables to accessories.
A new $480M human-centric AI lab founded by veteran researchers from major AI labs aims to shift focus from pure capability building to AI systems that strengthen human connections and organizational collaboration.
Despite widespread AI investment, 74% of companies have yet to realize tangible value from AI, while leaders outperform peers significantly by prioritizing core business transformation, strategic focus, and people-centric implementation over technology and algorithms.
Industry Voices
Yejin Choi
Researcher at Together AI
She's building common sense reasoning into AI systems and consistently ships creative benchmarks that expose what models still can't do.
Federico Bianchi
Researcher at Stanford
He dissects how language models actually represent topics and concepts under the hood, not just how they perform on benchmarks.
Pip White
Head of UK, Ireland and Northern Europe at Anthropic
She translates AI safety concerns into practical governance frameworks that European regulators actually implement.
Enjoyed this issue?
Get daily AI intel delivered to your inbox. No fluff, just the stories that matter.