Two of the biggest AI labs shipped products in the same three week window that quietly answer the same question: what happens to your AI assistant after you close the laptop. For most of this year, the honest answer was nothing. It waited for you to come back. As of this month, for a growing set of tools, it keeps working.

What happened

On July 7, 2026, Anthropic expanded Claude Cowork, its task automating agent, from a desktop only tool into a beta on web and mobile, with Max plan subscribers getting first access. The change is not just a new app icon. Anthropic built cloud based background processing so scheduled tasks, data pulls, and report building keep running even when your device is completely offline. A task can start on a laptop, keep going after the lid closes, and get reviewed later from a phone.

Fifteen days later, on July 22, OpenAI launched Presence, an enterprise platform for deploying AI voice and chat agents at companies. Presence ships with built in guardrails and policy controls, and a defined path for escalating to a human when the agent hits something it should not handle alone. Notably, OpenAI is not selling it as a self serve product. General availability is limited, and initial deployments are led by OpenAI's own Forward Deployed Engineers and a short list of systems integrators, the same delivery model enterprise software vendors have used for decades to make sure large rollouts do not fall apart in production.

Neither launch happened in isolation. The same stretch of July saw Microsoft push dedicated Sales Agent and Service Agent products, and Slack and Salesforce pair up on agent workflows inside tools sales and support teams already use every day. The pattern across all of it is the same: AI agents are moving out of the chat window and into the infrastructure layer, where they run continuously, get managed like software, and get sold with the support contracts that come with it.

Before and after, side by side

Flip the toggle below to see what actually changed between how AI assistants worked at the start of 2026 and how the newest agent products work now.

The AI agent's job description
Where it runs
Only inside the browser tab or desktop app you had open.
When it works
Only while you were actively prompting it, one step at a time.
What happens when you close your laptop
The task stops. Reopen it later and start from where you left off, if you remember to.
Who manages the guardrails
You, prompt by prompt, catching mistakes as they happen.
How it's sold
A per seat chat subscription you configure yourself.
Best fit today
Drafting, one off research, personal task lists.
Before, early 2026 vs. after, this month
DimensionBeforeAfter
Where it runsOnly the open tab or appAnthropic's and OpenAI's cloud, independent of any device
When it worksOnly while you prompt itContinuously, on schedules and multi step tasks
Laptop closedTask stopsTask keeps running, check results later on your phone
GuardrailsYou, prompt by promptBuilt in policy controls and human escalation, set once
How it's soldSelf serve chat subscriptionManaged platform deployment with delivery engineers
Best fitDrafting, one off researchRecurring workflows: support, sales follow up, reporting

Why it matters for companies

This shifts what buying an AI agent actually means. Six months ago, adding AI to a workflow was mostly a chat subscription decision, pick a tool, give it a prompt, review what it wrote. An agent that keeps running after hours, checks its own queue, and takes action without a person watching every step is a different kind of purchase. It touches procurement, security review, and whoever owns the process it is now allowed to act inside of. The question is no longer just "is the output good." It is "what happens when this runs unattended at 2am and gets something wrong."

The upside is real. Work that genuinely needs to happen outside business hours, after hours support triage, monitoring for issues overnight, chasing down a report before a 7am meeting, was previously either unstaffed or required someone on a night shift. An agent that keeps working while your team sleeps makes that coverage possible without hiring for it. That is the actual business case behind "always on," not novelty.

What to actually do about it

Do not buy on the always on pitch alone. Before deploying anything like this, get clear, specific answers to four questions: what does the agent do when it fails partway through a task, is there an audit log of every action it took, what is the escalation path when it hits something it is not confident about, and what does the background compute actually cost at your volume, since always on billing rarely works like a flat per seat fee. Start with one narrow, low risk workflow, an after hours support queue or a recurring internal report is a good first pick, rather than handing a brand new autonomous agent anything customer facing or financial on day one. Keep a human review step in place until the agent has a track record inside your business specifically, not just in a vendor's demo.

Not sure if an always on agent is the right first step for your business?

Take the AI Assessment. We will help you figure out what to automate first, and what still needs a person watching.

Take the AI Assessment →

Keep reading: Top 10 GitHub Repos This Week: July 23, 2026 →