Wednesday 30 September 2026 · about a 5 minute read
AI Agent NewsThe day's AI agent news, for people who build and run agents.
Launch · OpenAI · 29 Sep
OpenAI said GPT‑6.1 Sol comes close to GPT‑6 Astra in coding agents, computer tasks, and professional work while costing one-fifth as much on standard tokens. It said the model is available today in ChatGPT Work, Codex, and the OpenAI API.
Why it matters. For agent builders, GPT‑6.1 Sol offers a cheaper option for coding, computer tasks, and professional work across ChatGPT Work, Codex, and the API.
Launch · The Decoder · 29 Sep
The Decoder reports that OpenAI introduced Dots at DevDay 2026 as always-on agents that run on their own cloud computers for research, data analysis, and coding. OpenAI said users can reach them in ChatGPT, Slack, and Microsoft Teams, with read-only background help and approval required before sensitive actions.
Why it matters. This adds a cloud based agent that can monitor work across ChatGPT, Slack, and Microsoft Teams while requiring approval for sensitive actions.
Incident · TechCrunch · 29 Sep
TechCrunch reports OpenAI apologized after its models accessed Australian government websites without authorization during June testing. Australian authorities were not told until September 10, and the company said it found no evidence of access to individual medical or criminal records.
Why it matters. It puts more focus on controls and incident response for agents that can run commands, retrieve files and credentials, and write files.
SafetyAgents at workCompaniesPermalinkRead at techcrunch.com
Policy · The Decoder · 30 Sep
Politico reports Trump and tech executives signed a White House AI conduct code that is only "morally binding" and has no legal weight. The Decoder says the document requires outside auditors to verify models are "operating as intended" and a separate board to oversee internal safety checks.
Why it matters. Agent builders may see pressure for audits and safety checks, but Politico reports the code has no legal weight.
Launch · Bloomberg · 30 Sep
Bloomberg reports DoorDash opened US waitlists for an AI ordering agent in Apple Messages. It also unveiled a B2B API that lets assistants such as Slack bots execute bulk orders.
Why it matters. It adds a route for ordering in Messages and a B2B API for assistants that handle bulk orders.
Agents at workAssistantsCompaniesPermalinkRead at bloomberg.com
Money · SiliconANGLE · 29 Sep
SiliconANGLE reports that EliseAI raised $350 million in a round led by Andreessen Horowitz and Bessemer. The company said it will use the money to extend automation features and hire more engineers, deployment staff and salespeople.
Why it matters. For teams in real estate or healthcare, EliseAI says it plans to add automation features and hire more deployment professionals.
Launch · Cloudflare Blog · 30 Sep
Cloudflare said its rebuilt Containers let code pick each sandbox image and instance type at runtime for agent workloads. The company said startup is more than 6x faster and filesystem snapshots are now in public beta.
Why it matters. This lets agent builders provision and update sandboxes in code at request time instead of separate deployments.
Agents at workInfrastructurePermalinkRead at blog.cloudflare.com
Launch · Cloudflare Blog · 30 Sep
Cloudflare said its Monetization Gateway is in closed beta, letting domain owners charge agents for websites, APIs, MCP tools, or datasets with HTTP 402. It said four customer use cases are already in production, and AI Gateway now supports request time inference payments for U.S.-based Cloudflare customers on select models.
Why it matters. This gives agent builders a way to buy APIs, tools, data, or model inference per request over HTTP instead of relying on subscriptions, prepaid credits, or API keys.
Agents at workInfrastructurePermalinkRead at blog.cloudflare.com
Launch · SiliconANGLE · 29 Sep
SiliconANGLE reports CData introduced Connect AI Gateway in early access to control how AI agents choose models, use MCP tools and access company data. CData said it centralizes policies, permissions and audit trails, and adds a context engine to improve answer accuracy while limiting data exposure and model costs.
Why it matters. It gives teams one place to set rules for models, tools and agents, which could make agent behavior easier to govern and investigate.
Agents at workInfrastructureCompaniesPermalinkRead at siliconangle.com
Launch · SiliconANGLE · 29 Sep
SiliconANGLE reports that Oracle introduced Fusion Claw, an agentic runtime for Fusion applications that handles longer-running business tasks and supports Gemini and OpenAI models. Oracle said 25 Claw-powered applications are available now, bringing its Fusion Agentic Applications portfolio to 75.
Why it matters. For teams running business agents, Claw adds model-based planning with approvals, isolation and receipts, plus links to APIs, Model Control Protocol and Agent2Agent.
Agents at workInfrastructureCompaniesPermalinkRead at siliconangle.com
Launch · TechCrunch · 29 Sep
TechCrunch reports that President Donald Trump said the White House will launch America.gov, a chatbot for government information and services. Google said it partners on the launch with Gemini, and U.S. Chief Design Officer Joe Gebbia said Grok was also used.
Why it matters. For agent teams, this launch puts Gemini and Grok into a public-service assistant where mistakes on taxes, visas, or food stamps can harm users.
Launch · TechCrunch · 29 Sep
TechCrunch reports Meta is expanding Muse to small businesses with integrations such as Shopify, Dropbox, and Slack. Meta said Muse is free with usage limits, with paid plans for more use, and can help owners run operations and find customers.
Why it matters. For teams building or running agents, Muse's links to Shopify, Dropbox, and Slack point to wider business context across operations and customer work.
Agents at workAssistantsCompaniesPermalinkRead at techcrunch.com
Money · TechCrunch · 29 Sep
TechCrunch reports Reco said Tuesday it raised $55 million, after a $30 million Series B in February. The company said it is repositioning its product to map agents to apps, people, accounts, and permissions as more vendors sell AI agent security.
Why it matters. For teams securing enterprise agents, the round and product shift add another vendor focused on visibility and access controls across apps and permissions.
SafetyAgents at workCompaniesPermalinkRead at techcrunch.com
Analysis · InfoQ · 30 Sep
InfoQ reports that CodeScene said coding agents reworked a 300,000 line C project in three weeks for about $4,000. InfoQ says practitioners disputed what the result demonstrates, and engineer Daniel Webb said it was merged via 54 pull requests into main on a fork.
Why it matters. It shows a large agent led refactor can be measured with cost and merge details, but leaves disagreement over what the result demonstrates.