In April 2026 Claude Cowork became generally available on paid plans, OpenAI launched workspace agents for ChatGPT Business and Enterprise, Google turned Vertex AI into the Gemini Enterprise Agent Platform, and Anthropic released Claude Opus 4.7 and Claude Design. The shared theme: agents built once and shared by a team, with admin controls to match. Several of these products have since changed.
What should an operator have done that month?
Three things, in order of how many teams they affected. Each still applies in September 2026, under the names the products carry now.
- Decide which team gets a desktop or workspace agent first, and set spend limits and group access before rolling it wider.
- Pick one recurring team task, such as a weekly report, and build it once as a shared agent instead of as a personal prompt.
- Review which agents and connectors hold access to your Google Workspace, Microsoft 365 or Claude organization, and remove what nobody owns.
Claude Cowork became generally available, with admin controls for rolling it out
What happened (April 9, 2026). Anthropic moved Claude Cowork, the desktop agent that works on local files and across apps, out of research preview to general availability on all paid plans for macOS and Windows. Enterprise admins got role-based access controls with SCIM groups, group spend limits, usage analytics and an Analytics API, expanded OpenTelemetry events, per-tool controls on MCP connectors, and a Zoom connector. Source: Claude blog.
Why it matters. Anthropic reported that most Cowork use came from operations, marketing, finance and legal, not engineering. Spend limits and per-team access made a staged rollout possible without a separate budget fight.
What to do. Turn it on for one team, cap that group's spend, and restrict write actions on connectors until you have seen what people do with it. See what Claude Cowork does.
Since then. On September 16, 2026 Anthropic began merging Cowork into the main Claude app, rolling out to Pro and Max first, with Team and Free to follow and at least 30 days' notice for Enterprise admins. Source: Claude blog.
OpenAI launched workspace agents in ChatGPT for teams
What happened (April 22, 2026). OpenAI released workspace agents in research preview for ChatGPT Business, Enterprise, Edu and Teachers plans. They are powered by Codex, run in the cloud so they keep working when nobody is at the keyboard, connect to outside apps, and are built once and shared across the organization, in ChatGPT or in Slack, within the permissions admins set. OpenAI described them as an evolution of custom GPTs. Source: OpenAI, as also reported by 9to5Mac.
Why it matters. A custom GPT answered one person at a time. A workspace agent does a job for the team on its own, which is a different thing to budget for and to supervise.
What to do. Choose one recurring, low-risk task and build it as a shared agent. Name an owner who checks its output weekly.
Since then. Workspace agents became generally available on May 22, 2026, and began drawing down workspace credits on July 6, 2026, after OpenAI extended the free period from May 6 (reported by Tech Times, citing OpenAI's release notes). On September 11, 2026 OpenAI gave notice that custom GPTs will be retired in ChatGPT Enterprise on December 11, 2026 in favor of plugins and Skills. Source: OpenAI Help Center.
OpenAI released GPT-5.5 across paid ChatGPT plans
What happened (April 23, 2026). GPT-5.5 rolled out to ChatGPT Plus, Pro, Business and Enterprise, with GPT-5.5 Pro for Pro, Business and Enterprise. OpenAI pitched it at work that spans tools: coding, research, data analysis, documents and spreadsheets, and operating software until a task is finished. Source: OpenAI, as also reported by TechCrunch.
Why it matters. Chat users got it automatically. Teams calling OpenAI through the API had a model change to test before switching.
What to do. Chat users need do nothing. If you run prompts through the API, re-test them on the new model before changing the model name in production.
Since then. OpenAI announced the GPT-5.6 family, Sol, Terra and Luna, on July 9, 2026. Source: OpenAI.
Anthropic released Claude Opus 4.7
What happened (April 16, 2026). Opus 4.7 became generally available in the Claude apps, the API, Amazon Bedrock, Google Cloud Vertex AI and Microsoft Foundry, at the same price as Opus 4.6: $5 per million input tokens and $25 per million output tokens. A new tokenizer can turn the same text into about 1.0 to 1.35 times as many tokens, and Anthropic advised re-tuning prompts because the model follows instructions more literally. Source: Anthropic.
Why it matters. An unchanged list price did not mean an unchanged bill. API users had to measure their own token counts to know.
What to do. Run a week of real API traffic on the new model, compare token usage and output, then switch.
Since then. Anthropic released Claude Opus 5.5 on September 22, 2026. Source: Anthropic.
Claude Design arrived in preview, and Claude for Word reached Pro and Max
What happened (April 17, 2026). Anthropic Labs released Claude Design in research preview on Pro, Max, Team and Enterprise plans, for prototypes, slides and one-pagers, with export to Canva, PDF, PPTX and HTML. It is off by default for Enterprise organizations until an admin enables it. The same day, Claude for Word, the add-in that makes edits as tracked changes, became available on Pro and Max. Source: Anthropic Claude on X: Claude for Word.
Why it matters. A small team without a designer could produce a first-draft deck or one-pager in-house, and redline a Word document without pasting it into a chat.
What to do. Try one real internal deck in Claude Design before paying for a template library. Enterprise admins decide whether to enable it and for whom.
Since then. On September 16, 2026 Anthropic brought Claude Design into the main Claude conversation and added Claude Docs and Claude Slides in beta. Source: Claude blog.
Google Workspace added no-code agent skills, an MCP server and agent governance
What happened (April 22, 2026). At Cloud Next, Google announced Workspace Intelligence, a shared context layer across Gmail, Docs, Drive and Chat for Gemini agents; skills built in Workspace Studio that can be invoked anywhere Gemini runs in Workspace; a Workspace MCP server in preview; and agent governance controls in the admin console, rolling out in the following weeks. Source: Google Workspace blog.
Why it matters. A written procedure can become a reusable skill without a developer, and the MCP server lets Claude, ChatGPT or a custom agent reach Workspace data through a supported route.
What to do. Turn one written procedure into a skill. Before connecting any outside agent through the MCP server, decide who approves it in the admin console.
Google renamed Vertex AI the Gemini Enterprise Agent Platform
What happened (April 22, 2026). Google launched the Gemini Enterprise Agent Platform as the evolution of Vertex AI, with future Vertex AI services delivered through it. It adds agent identity, an agent registry and an agent gateway with prompt-injection protections. The Gemini Enterprise app gained a no-code Agent Designer, an Inbox for monitoring long-running agents, and Projects that hold shared team context. Source: Google Cloud blog: Agent Platform Google Cloud blog: Gemini Enterprise.
Why it matters. Anything your developer or vendor built on Vertex AI now lives under a new name and roadmap. Firms on Gemini Enterprise got a place to see what their agents are doing.
What to do. Ask whoever built your Google Cloud AI work whether anything changes for you. Most small teams needed to do nothing else.
Anthropic launched Claude Managed Agents for developers
What happened (April 8, 2026). Claude Managed Agents opened in public beta on the Claude Platform: hosted infrastructure for running agents with state, memory, permissions and scheduled execution. It is billed at standard token rates plus $0.08 per session-hour of active runtime. Multi-agent coordination was in research preview on request. Source: Claude blog.
Why it matters. A custom agent for your business no longer needs its own servers to keep state or run on a schedule, which lowers what a custom build costs to host.
What to do. If you are scoping a custom agent, ask your developer whether a managed runtime fits before they build hosting. What is an AI agent? covers the basics.
VTS added AI lease abstraction with human review for asset managers
What happened (April 1, 2026). VTS launched Asset Intelligence inside its platform. It abstracts leases with AI, has lease administration experts verify each abstract, and lets teams query lease data conversationally through Asset Chat for compliance, renewals and reconciliation. Source: VTS.
Why it matters. Lease abstraction is one of the clearest AI uses in commercial real estate, and human verification is what makes an abstract usable in a renewal or a sale.
What to do. If you are on VTS, ask what verification turnaround and error handling look like before moving leases in. Our guide to AI rent roll and lease abstraction has the checks.
An item makes the digest only if it changed a decision a small or mid-size team would make that quarter, and only with a dated primary source. Model benchmarks, leaks and funding news without a product consequence are left out.
Common questions
What were the biggest AI changes for business owners in April 2026?
Claude Cowork became generally available on paid plans on April 9, OpenAI launched workspace agents for ChatGPT Business and Enterprise on April 22, Google renamed Vertex AI the Gemini Enterprise Agent Platform the same day, and Anthropic and OpenAI released Claude Opus 4.7 and GPT-5.5. The common thread was agents that a whole team shares, with admin controls.
Is Claude Cowork still a separate product?
It was a separate tab in the Claude desktop app when it became generally available in April 2026. On September 16, 2026 Anthropic began merging Cowork into the main Claude app, starting with Pro and Max. Work done in Cowork stays where it was.
What happened to ChatGPT workspace agents after the April preview?
They became generally available on May 22, 2026 for Business, Enterprise and Edu plans, and started using workspace credits on July 6, 2026. OpenAI is also retiring custom GPTs in ChatGPT Enterprise on December 11, 2026 in favor of plugins and Skills.
Should we still use Claude Opus 4.7 or GPT-5.5?
Both have been superseded. Anthropic released Claude Opus 5.5 on September 22, 2026, and OpenAI announced the GPT-5.6 family on July 9, 2026. In chat, pick the current model. API users should test it on their own prompts and change the model name once results hold.
Why was this April page written in September?
The monthly digest started in September 2026, and earlier months were written afterward so the archive is complete. Each item is dated to April 2026 from a primary source, and anything that changed later carries a dated note saying what changed.