Daily Intelligence·Saturday, April 25, 2026

The first agent on your org chart shipped Tuesday

By Alfred Belvedere — Founder, Omni AI

7 tags

OpenAI Workspace AgentsCodex enterprise agentsAdobe CX Enterprise CoworkerSnowflake Cortex AIshared agent permissionsagent org chartAI teammate operating model

Tuesday morning OpenAI quietly shipped Workspace Agents — shared, long-running, Codex-powered agents that live inside your team's ChatGPT, hold their own permissions, and keep working when nobody's logged in. The same week, Adobe rebranded Experience Cloud to CX Enterprise and introduced "Coworkers" — agents with goals instead of prompts. Snowflake locked in twin $200M deals with OpenAI and Anthropic to make sure those agents reason over governed data instead of bootleg pipes. Three announcements, one shape: the agent is no longer a tool you open. It's a teammate with a seat — and the operators who don't put one on the org chart this quarter are about to be the most expensive employees in their own company.

Today’s Key Insights

OpenAI's Workspace Agents launch on April 22 changed the unit of work inside ChatGPT. Until Tuesday, every agent was a single-user thing — you opened it, it answered, you closed the tab. Workspace Agents are shared, persistent, Codex-powered, and run in the cloud whether you're online or asleep. They plug directly into Slack, Salesforce, Notion, Google Drive, and Microsoft 365, and they're free until May 6 before credit-based pricing kicks in. That last detail is the tell — OpenAI is not pricing this as a feature. It's pricing it as infrastructure.

Adobe used Summit 2026 to do something that should be impossible for a $200B incumbent moving this fast: rename the entire Experience Cloud to CX Enterprise and reorganize the product around 'Coworkers' — agents that operate against business goals rather than prompts. The Adobe Marketing Agent is now generally available inside Microsoft 365 Copilot, and in beta inside Claude Enterprise, ChatGPT Enterprise, Gemini Enterprise, and IBM watsonx Orchestrate. When the legacy CX vendor is shipping cross-platform, agent-on-agent interoperability before most of your stack ships single-agent reliability, the gap stops being a tooling gap. It's an org-design gap.

Snowflake's twin $200M partnerships — one with OpenAI, one with Anthropic — were the unsexy but decisive move of the quarter. A Workspace Agent or a Coworker can be spun up in 20 minutes; if it's reasoning over the same Notion sprawl your interns curse, it will lie confidently and at scale. Cortex AI inside Snowflake is the layer nobody is screenshotting because it isn't photogenic. It's also the layer that decides whether your agents become coworkers or liabilities the moment they touch a customer record.

The pattern across all three announcements: enterprise AI just stopped being a copilot you summon and became a colleague you onboard. That requires a permissions model, a goal, an owner, a budget, and a kill switch. Most operators have none of those defined for the agents already running in their stack. That is the real 90-day project — not picking the model, not writing the prompt, building the operating model that turns an agent into a team member who can pass a security review.

Power Move

Open your org chart today and add one row at the bottom: agent name, who it reports to (a human), one-sentence goal, exact tool access, weekly task or dollar budget, and the kill command — who pulls the plug and how. Don't pick the agent first. Define the seat, then fill it. By Friday, that row should have a name, an owner, and a P&L line. The first agent that lands on a real org chart inside your company is the one that compounds. The rest are demos.