The Signal

OpenAI announced the general availability of its Decisions API, built around GPT-6 Luna and designed for high-frequency enterprise decisions such as classification, routing and workflow action selection. The API supports text, images and structured inputs, with inference residency options in the United States and European Union. Published pricing starts at US$0.10 per million input tokens, with cached input priced at US$0.01 per million tokens. OpenAI claims the API can be up to 10 times faster than accessing GPT-6 Luna through its Responses API. Named customer evaluations provide more specific evidence. Sierra reported p95 latency below 280 milliseconds and approximately 90% lower cost per call. Box reported approximately 5 times faster routing in a 30-case evaluation. Ramp reported 4 times faster AI cost attribution at 30% lower cost, while Datadog reported a 40-fold improvement in p95 latency against its comparison models. These results are company-reported and reflect different workloads and baselines. They should not be interpreted as general enterprise performance guarantees.

IBM announced expanded AI Asset Discovery capabilities within watsonx.governance, increasing visibility into AI assets distributed across enterprise cloud platforms and agent environments. The enhancement introduces discovery support for Google Cloud Gemini Enterprise Agent Platform and AWS Bedrock AgentCore, alongside existing support for environments including Azure AI Foundry, AWS Bedrock and IBM watsonx.orchestrate. The update also extends visibility into knowledge bases associated with agents, helping organisations identify not only the AI systems they operate but also the information sources those systems can access. IBM cited research suggesting that only 18% of surveyed organisations maintain an up-to-date AI inventory, highlighting an important governance challenge as enterprises expand agent deployment. A key limitation is that newly introduced automated handling capabilities initially apply to watsonx Orchestrate and IBM Bob, rather than every supported platform. The announcement establishes expanded product functionality but does not provide measured reductions in governance incidents or audit costs.

OpenAI announced an agreement to distribute its frontier AI models and Codex workflows through Google Cloud Model Garden, creating another commercial route for enterprises to access OpenAI technology. Under the arrangement, eligible customers will be able to purchase access through Google Cloud Marketplace and apply existing Google Cloud spending commitments. The agreement is intended to combine OpenAI's models with Google Cloud infrastructure, enterprise access controls and governance capabilities. For organisations already operating substantial workloads on Google Cloud, the arrangement could reduce procurement friction and provide greater flexibility in allocating existing cloud commitments. However, general availability was planned for the weeks following the announcement. The partnership should therefore not be described as a generally available integrated service during 19–25 October. Microsoft remains OpenAI's primary cloud partner.

OpenAI introduced several updates to ChatGPT Enterprise and Edu administration, strengthening usage visibility, workspace management and user-level security controls. On 23 October, workspace analytics gained additional adoption and usage reporting capabilities, including workplace-health reporting and by-user usage exports. OpenAI also introduced a seven-day grace period for affected users required to migrate personal workspaces into enterprise environments associated with verified domains. On 20 October, the company introduced locked chats, allowing users to protect selected conversations through a four-digit PIN or supported device authentication. Locked chats were introduced through a gradual rollout and remain subject to applicable organisational retention and compliance requirements. These features improve administrative oversight but do not independently demonstrate higher employee productivity or better financial returns.

Uber announced the company-wide rollout of ChatGPT Work, expanding from an initial 1,000-employee pilot to planned availability for more than 30,000 employees across its global workforce. The platform supports multi-step knowledge-work activities, including research, meeting preparation, presentations, data analysis and cross-application workflow execution. ChatGPT Work connects with enterprise applications such as Google Drive, Gmail, Google Calendar and Slack, enabling employees to work across existing business systems. The deployment targets functions including operations, finance, marketing and product development. Uber is supporting adoption through an internal AI skills library, approved integrations and dedicated employee training. The expansion represents more than 30 times the original pilot population, although that figure describes intended access rather than active adoption. Neither Uber nor OpenAI disclosed independently verified productivity improvements, cost savings or realised financial returns. The companies indicated that additional adoption and outcome information would be shared as the rollout progresses.

Accenture Federal Services and Google Public Sector announced Gemini launchpads intended to accelerate the development and deployment of purpose-built AI agents across US federal agencies. The initiative combines Google's Gemini technology with forward-deployed engineers working alongside government personnel to integrate AI agents with mission data and operational systems of record. Initial applications focus on enterprise functions such as human resources, finance and supply chain management. The initiative forms part of Accenture Federal Services' previously announced commitment to establish a 1,000-person forward-deployed engineering workforce. That commitment predates the reporting week and should not be counted as a newly announced investment. The launchpads are designed to address implementation challenges, including systems integration, workflow configuration and operational deployment. However, the announcement does not disclose named agency deployments, contract values, completed implementations or independently measured productivity gains.

Germany-based ARIS announced Process Context for Enterprise AI, a capability designed to help AI agents understand enterprise workflows, organisational responsibilities, systems, rules, risks and operational controls. The platform combines Object-Centric Process Mining (OCPM), Process Lens and Model Context Protocol (MCP) connectivity to provide structured operational context to enterprise AI applications. ARIS also published research conducted with The Hackett Group involving 200 respondents from large and multinational organisations. The findings indicate that 82% of respondents believe process context is necessary for deploying AI agents, while only 21% report comprehensive, real-time visibility into enterprise workflows. ARIS additionally reported that organisations using process-context tools were more than five times as likely to report measurable AI ROI. This is an observed survey association, not proof that the tools caused the reported returns. The research sample included organisations from North America (45%), Europe (34%) and other regions (21%). Approximately 80% represented organisations with revenue exceeding US$1 billion, while 60% came from organisations employing more than 10,000 people. Although the announcement occurred on 20 October, product availability was scheduled for 26 October, immediately after the reporting window.

Deliverance AI and HPE announced a validated joint solution combining Deliverance AI Agentic OS with HPE Private Cloud AI, targeting enterprises operating sensitive and regulated workflows. The architecture supports private-cloud, on-premises and air-gapped deployment environments, addressing requirements involving data sovereignty, controlled information access, auditability and human oversight. Its capabilities include multi-agent orchestration, model routing, quality evaluation, approval controls and audit mechanisms. The solution identifies four principal workflow categories: Respond, Produce, Review and Resolve. Deliverance AI also described a live engineering demonstration involving an unnamed UK mobile network operator. During the demonstration, agents reportedly answered a network investigation question in under five minutes, compared with the operator's estimate of up to three hours for equivalent manual work. This is a company-reported demonstration, not an independently validated productivity study. The announcement does not establish the exact date of the underlying demonstration or provide evidence of sustained production performance.