AI Went Always-On This Week. Your Job Just Changed.
OpenAI and Google shipped always-on agent platforms in the same week. Non-technical teams can now set up agents by describing workflows in plain English. The skill gap is no longer how to prompt — it's how to scope, delegate, and verify autonomous work.
By Forge Team
The relevant new skill is not how to write a better prompt. It is how to hand off work to a system running in the background — one that has access to your team's email, project tracker, and CRM — and know in advance what it will and will not do on your behalf. That question became urgent this week.
What happened in four days
In four days, two of the largest AI companies shipped three agent products that operate continuously, not on demand. OpenAI launched Workspace Agents on April 22, replacing custom GPTs with always-on agents that connect to Slack, Salesforce, Gmail, and 60+ enterprise tools — agents that run when you are not there (OpenAI blog, Apr 22). The same day, Google announced Workspace Intelligence, making Gemini the continuous operating layer across Gmail, Drive, Calendar, and Docs with persistent awareness of your files and email history (Google Workspace blog, Apr 22). On April 24, OpenAI released GPT-5.5, benchmarked against 44 real occupations and scoring 84.9% on GDPval — a model explicitly framed as one that finishes tasks rather than answers questions (TLDR AI, The Neuron, Apr 24).
Ethan Mollick observed the same day that "organizational design for agents is hard, benchmarking agents working in concert is hard" — and that multi-agent coordination is "the next critical frontier" (X, Apr 24). Nobody, including the people who built these systems, has a tested playbook yet.
Three questions before you delegate anything
The technical bar for setting up an always-on agent dropped this week. The professional bar — deciding what to give it, how to scope it, and where to stand in the loop — did not. Answer these three questions before you hand any task to an agent:
Is this actually an agent use case? Agents do well on high-volume, defined-input tasks where an error is low-cost and detectable before it causes damage. They do badly on tasks where context shifts unpredictably, where output goes directly to a client without review, or where the person on the receiving end cares whether a human made the call.
What exactly does the agent own — and where does it stop? An agent connected to your email needs a specific brief. "Monitor unread messages and draft replies" is not a brief. "Flag any message containing a direct question from a customer, draft a response using our FAQ document, and hold it for my review before sending" is a brief. The scope defines the risk.
Where do you check the work? A checkpoint is not a safety net for when things go wrong. It is a designed stop where you review output before it moves forward. For any agent handling external-facing work, build at least one checkpoint per day — or per deliverable batch — before anything leaves the system.
Work through the decision: does this task actually belong to an agent, or something simpler?
What happens with a loose scope
A marketing coordinator at a 65-person B2B SaaS company sets up an always-on agent to monitor competitor announcements and draft a weekly competitive brief. She describes the workflow in plain English, connects it to her company's Notion workspace, and the agent runs every Tuesday morning.
Four weeks in, she notices the brief is getting longer. The agent has started including product updates from peripheral tools her company does not compete with — the brief is technically complete, but two-thirds of it is noise. The fix is a scope revision: she adds a list of five direct competitors and a rule that the agent includes only product, pricing, or feature announcements. The brief becomes usable again.
The skill she needed was not better prompting. It was knowing how to tighten a scope definition after seeing what the agent did with a loose one.
Write a scope definition that tells the agent exactly what it owns — and what it doesn't.
When smaller scope is the right call
A proposals manager at a 120-person professional services firm considers using an always-on agent to monitor incoming RFP documents and draft initial response outlines before she reviews them. The case is clear: she receives 15 to 20 RFPs a month, the initial scan is time-consuming, and the outline structure is mostly repeatable.
She starts narrower. The agent reads incoming RFPs and creates a structured intake document — title, client, deadline, key requirements, which service lines are relevant — and nothing else. No prose drafts. The intake document takes her from thirty-minute initial reviews to ten-minute ones. The checkpoint is the intake document itself: she reviews every one before outline work starts.
The agent does less than it could. That is the point. The smaller scope builds trust before the stakes rise.
Design the checkpoint that sits between your agent's output and the moment it matters.
What the shift actually requires
The professionals who do well with always-on agents in the next twelve months will not be the ones who set them up fastest. They will be the ones who defined what the agent owned before they turned it on.
Like this post?
Get the next one in your inbox. Practical AI skills, no filler.