Agents+5Jun 24
Google DeepMind adds computer-use capability to Gemini 3.5 Flash
Gemini 3.5 Flash now supports computer use — the ability to observe a screen and take actions on a computer — making this capability available in a fast, lower-cost Google model. Developers and businesses building automated workflows can now use Gemini 3.5 Flash to drive desktop or browser tasks, giving them a cost-efficient option alongside existing computer-use tools from Anthropic and OpenAI.
What still breaks: Computer-use agents still struggle with ambiguous UI states, multi-step tasks involving login flows, and applications they have not seen before.
Text+5Jun 22
OpenAI launches Daybreak security tools including GPT-5.5-Cyber and Codex Security
OpenAI released GPT-5.5-Cyber and Codex Security under its Daybreak initiative, tools designed to find, validate, and patch software vulnerabilities at scale for organizations and open-source maintainers. Security teams and open-source maintainers can now use an AI purpose-built for vulnerability scanning and remediation rather than adapting general coding models, potentially accelerating patch cycles on widely used software.
What still breaks: AI-assisted security tools can generate false positives, miss novel vulnerability classes, and require human review before patches are applied in production.
Other+2Jun 22
NVIDIA liquid cooling now supports 45C coolant temperature, reducing AI data center energy use
NVIDIA's newest AI servers can run liquid cooling at up to 45 degrees Celsius coolant temperature, higher than previous systems, which allows data centers to reject heat without as much additional refrigeration energy. Higher coolant temperature means cheaper, simpler cooling infrastructure and lower power-usage effectiveness for AI factories, making it less expensive to operate large GPU clusters.
What still breaks: Retrofitting existing data centers to take advantage of higher-temperature cooling requires infrastructure upgrades that many operators have not yet made.
Other+2Jun 18
FERC issues ruling on large-load grid interconnection, affecting AI data center power access
The Federal Energy Regulatory Commission issued a ruling on large-load interconnection that changes how high-power facilities like AI data centers can connect to the US electric grid. Getting grid connections approved faster and more predictably removes one of the key bottlenecks slowing AI data center construction in the United States, which affects how quickly new compute capacity can come online.
What still breaks: Regulatory rulings take time to filter into actual interconnection queues; near-term power availability for planned AI campuses is unlikely to change immediately.
Announced or reported, but nothing you can use or verify. Some of these will become real. Some won’t.
AgentsJun 21Samsung Electronics deploys ChatGPT Enterprise and Codex to employees worldwide
Waiting on: Not yet publicly usable
OtherJun 24OpenAI and Broadcom unveil Jalapeno, a custom LLM inference chip
Waiting on: Not yet live, signed, deployed, or otherwise concrete
OtherJun 26Vantage tops out second building at OpenAI Lighthouse campus in Wisconsin
Waiting on: Primary verification
RobotsJun 24Agility Robotics announces SPAC merger raising $620M to advance Digit v5 and fulfill orders
Waiting on: Primary verification
TextJun 26OpenAI previews GPT-5.6 series (Sol, Terra, Luna) in limited partner access
Waiting on: Not yet publicly usable
