A pricier Business seat, a page-as-tools toggle, and an agent loop in Java

Three doors into the same workday

The Register reports that OpenAI has introduced ChatGPT Business Premium seats at 125 dollars a month, or 100 dollars a month on an annual plan, against 25 dollars and 20 dollars for a standard Business seat. Premium seats carry 5 times the usage capacity and are exempt from the daily limit of 5 hours on advanced features. Brandon Vigliarolo writes that OpenAI says customers asked for higher limits for deeper work on the Business tier: GPT-5.5 Instant is described as virtually unlimited on Business seats, while GPT-5.5 Thinking is capped at 3,000 requests a week and Pro at 15 a month, with Premium seats receiving 5 times those allocations. OpenAI is also running a waitlist promotion that gives up to 5 seats 100 dollars in workspace service credits each for the first 10,000 customers, though advanced features still require ChatGPT credits. Separately, InfoQ reports that Cloudflare has opened a developer preview in which a site can switch on WebMCP support from a single dashboard toggle so that AI agents call named functions on the page instead of scraping its HTML. The preview sits under Agent Readiness. On the GitHub Blog, Edward Burns introduces the GitHub Copilot SDK for Java, a client library that lets a server-side Java application create Copilot agent sessions, register tools, send prompts and read structured responses. It is a preview at version 1.0.7-preview.1 and is distributed as a Maven dependency.^{[1], [2], [3]}

Access, exposure, and the in-process loop

The three releases organise different entry points into ordinary software work. The Premium seat is an access and quota product: it multiplies capacity, removes a daily advanced-feature clock, and keeps deeper work behind a higher monthly price and remaining credit rules. Cloudflare's toggle is about tool exposure rather than seat count. Sergio De Simone writes that Cloudflare injects a bridge.js module at the edge with HTMLRewriter, composes the packs named in data-packs into one tool list, and registers each with registerTool. Two packs are available so far: Content Credentials, which lets an agent read C2PA metadata from images, and Site MCP server, which proxies a site's existing MCP server tools. WebMCP itself is still a preview specification and works only in Chrome 145 and later, so the reach of the toggle depends on a browser feature that is not yet settled. The Java SDK embeds the agent loop inside an application the team already ships. The post shows three ways to define a tool — the CopilotTool annotation, a lambda-style ToolDefinition.from call, or JSON Schema — and describes sendAndWait as the agentic loop in one line.^{[1], [2], [3]}

What the week jointly reports

Taken together, the week maps a path from paid capacity to page-level tools to in-process agent sessions. The library is framework-agnostic and supports bring-your-own-key configuration against direct model providers such as OpenAI, Azure, Anthropic, or OpenAI-compatible endpoints. Running it needs JDK 17 or 25, Maven 3.9 or later, an active GitHub Copilot subscription and Copilot CLI 1.0.71 or later. What the three items jointly report is concrete: deeper ChatGPT Business work is sold as a higher seat with 5 times the usage and no 5-hour advanced-feature day; a Cloudflare site can, in preview, hand agents a named tool list instead of raw HTML; and a Java service can open a Copilot session, register tools, and drive the loop from inside its own process.^{[1], [2], [3]}

Three doors into the same workday

Access, exposure, and the in-process loop

What the week jointly reports

References

More analysis from this desk