-
Launch · The Decoder · 30 Sep
THE DECODER reports OpenAI and Synopsys signed a multi-year partnership to build GPT-Synopsys, a chip design model using Synopsys' EDA tools on OpenAI's infrastructure. Both companies said customer data will stay encrypted and not be used for training, and early tests are underway.
Why it matters. This points to agentic use of domain tools, with engineers setting objectives while the model operates Synopsys' EDA software.
ModelsInfrastructureCompaniesPermalinkRead at the-decoder.com
-
Launch · Cloudflare Blog · 1 Oct
Cloudflare Blog reports that Cloudflare released Clef and Clef-flash decision models on Workers AI and made them open source on Hugging Face with an Apache 2.0 license. Cloudflare says its new RL product lets customers fine-tune Clef for their own use cases.
Why it matters. This gives agent builders Jev-API compatible decision models on Workers AI, with a new option to fine-tune Clef for a specific workload.
ModelsInfrastructurePermalinkRead at blog.cloudflare.com
-
Analysis · TechCrunch · 1 Oct
TechCrunch reports Brian Chesky said consumer AI agents need a true operating system and software developer kit, not just apps on today’s platforms. He said Airbnb plans voice agents this fall and is exploring more collaborative, agent-friendly interfaces.
Why it matters. Agent builders may need richer interfaces, interoperability, and platform support if apps like Airbnb are to work well with consumer agents.
AssistantsVoiceInfrastructurePermalinkRead at techcrunch.com
-
Analysis · TechCrunch · 1 Oct
TechCrunch reports Time said Trump spent hours with Musk's Grok before the U.S. invaded Venezuela, and Grok said many Venezuelans would celebrate Nicolás Maduro's downfall. TechCrunch reports the Pentagon's head of AI said in June the military used Gov Grok in deploying and striking targets in the Iran War.
Why it matters. This raises the stakes for people building or running AI agents because Grok was used by Trump and the military.
SafetyAssistantsCompaniesPermalinkRead at techcrunch.com
-
Money · TechCrunch · 1 Oct
TechCrunch reports that Photon raised $4.5 million in seed funding for tools that help developers build agents for iMessage and WhatsApp. The outlet says the company has signed up over 40,000 developers, and Daniel Tian said agents will replace apps over time.
Why it matters. Developers and agent operators may get more tooling for messaging based agents as Photon adds funding after signing up over 40,000 developers.
InfrastructureCompaniesPermalinkRead at techcrunch.com
-
Launch · The Decoder · 1 Oct
THE DECODER reports Tavus introduced Griffin, which it calls the first Human Interaction Model for real-time face-to-face video conversation. Tavus said 48 percent of study participants thought Griffin was real after a one-minute call, and a Griffin-Lite preview is available to select testers.
Why it matters. It suggests live video conversations with digital personas may feel more human, though Tavus is keeping Griffin-Lite to select testers until safety concerns are addressed.
ModelsVoicePermalinkRead at the-decoder.com
-
Launch · Amazon Web Services · 30 Sep
Amazon reports that Bedrock AgentCore Runtime Instances support persistent, long-running agent workflows with multi-day sessions, GPUs, persistent volumes, and colocated agents. Its sample runs three agents that share a filesystem and a runtimeSessionId to produce and check a playable .wav track.
Why it matters. This gives agent teams a way to keep shared context, files, and GPU work alive across several days on one instance.
Agents at workInfrastructurePermalinkRead at aws.amazon.com
-
Analysis · A Few Thoughts on Cryptographic Engineering · 30 Sep
A few thoughts on cryptographic engineering reports recent agent breakouts show labs have not tested containment properly because OpenAI had weak infrastructure and unclear security authority. It argues sandboxing alone may still fail, because useful agents need information access and cannot be perfectly isolated.
Why it matters. Agent teams may need stronger containment and tighter control of information access, because the post says useful agents cannot be perfectly isolated.
SafetyInfrastructurePermalinkRead at blog.cryptographyengineering.com
-
Money · SiliconANGLE · 1 Oct
SiliconANGLE reports Arceus Legal raised $17 million to expand beyond commercial contract work. The firm says licensed attorneys use AI agents and software it built itself, and Greycroft LP led the round.
Why it matters. It highlights demand for agent workflows paired with licensed review, tool integrations and fixed pricing in professional services.
Agents at workCompaniesPermalinkRead at siliconangle.com
-
Launch · SiliconANGLE · 1 Oct
SiliconANGLE reports Cognition’s Devin now works across more of software development, which Silas Alberti said demands always-on infrastructure for continuous learning. SiliconANGLE reports CoreWeave said Forge adds agent tracing and can hot-load model checkpoints into live deployments without redeploying.
Why it matters. This matters for agent teams building systems that keep learning in production and can take model updates without redeploying.
Agents at workInfrastructureCompaniesPermalinkRead at siliconangle.com
-
Launch · SiliconANGLE · 1 Oct
SiliconANGLE reports IBM now offers a self-hosted option for its Bob software development platform. IBM said it is generally available for customer-controlled environments and can keep code, development context and build artifacts inside customer-managed infrastructure.
Why it matters. Teams that handle sensitive code or regulated data can use Bob without moving that work outside customer-managed infrastructure.
Coding toolsInfrastructurePermalinkRead at siliconangle.com
-
Launch · SiliconANGLE · 1 Oct
SiliconANGLE reports AWS released Strands Decider 2B as an open-source decision model for faster choices without generating text. AWS said the model is optimized for local deployments and rapid experimentation, with downloads on Hugging Face and code on GitHub.
Why it matters. It gives developers an open-source decision model they can run locally for agentic tasks such as tool selection, model routing and guardrail enforcement.
ModelsAgents at workInfrastructurePermalinkRead at siliconangle.com