AI News, July 9: OpenAI Launches ChatGPT Work, and Every Lab Now Sells an Agent for Your Job
For two years, the pitch behind every frontier lab’s roadmap has been the same sentence: an AI that doesn’t just answer your questions, but does your work across your apps. Anthropic built its whole Claude strategy around it. Today, OpenAI shipped its most direct version of that pitch yet — and Google spent the last week making sure its own entry, Gemini Spark, was sitting on your Mac before OpenAI’s livestream started.
Here’s what mattered on July 9, 2026.
The Big Story: ChatGPT Work Turns ChatGPT Into an Agent That Ships Finished Work
OpenAI launched ChatGPT Work today — an agent inside ChatGPT that takes an outcome instead of a prompt, gathers context across your connected apps and files, breaks the job into steps, and works independently for hours. The output isn’t a chat reply; it’s finished spreadsheets, slides, documents, and interactive web apps. It runs on GPT-5.6, which OpenAI released the same day — the model is GPT-5.6, ChatGPT Work is the agent product built on top of it.
The integrations story is the part worth reading twice. ChatGPT Work launches with a directory of over 1,400 connectable apps that you can @-mention to pull context from, and early testers wired it into HubSpot, Gong, Slack, email, and project management tools. Zapier says an automated lead-triage workflow “helped us identify and hand off seven figures in pipeline every month,” and Virgin Atlantic claims a competitor-analysis cycle that took weeks now takes hours. Those are vendor-supplied numbers from launch-day testers, so apply the usual discount — but the direction is clear.
Two details matter for anyone budgeting for this. First, ChatGPT Work is usage-metered: tasks consume a variable slice of your plan’s allowance depending on complexity, the same structure Codex uses, and there’s no published per-task price. Enterprise admins get spend controls; everyone else gets to find out how fast a long-running agent burns a monthly allowance. Second, autonomy is gated — a Plan mode shows you the steps before it starts, and check-ins and action approvals keep a human in the loop. Rollout started today for Pro, Enterprise, and Edu on web and mobile (and all plans on the new macOS desktop app), with Plus and Business following over the next few days.
Today’s Top Stories
Gemini Spark, Google’s 24/7 Agent, Is Now on Your Mac
Google didn’t wait for OpenAI’s event. On July 1, Gemini Spark landed on macOS in beta, folding Google’s always-on agent into the Gemini desktop app. Spark — introduced at I/O in May as a 24/7 agentic assistant with Gmail integration — can now work with local files on your computer, sort and organize them, and turn them into Workspace docs. It also picked up real-time topic tracking (sports scores, stock moves, breaking news) and new connections to Google Tasks and Keep.
The integration list is telling: alongside Google’s own apps, Spark connects to Canva, Dropbox, Instacart, OpenTable, and Zillow Rentals, and supports the Model Context Protocol for wiring in anything else. The catch is access — the Mac beta is US-only and restricted to Google AI Ultra subscribers, Google’s most expensive tier. Google has the deepest default integration moat of anyone (your Gmail, your Calendar, your Drive), and it’s rationing it to the top of its pricing ladder.
One Desktop App: Codex Merges Into ChatGPT, Atlas Winds Down
The quieter half of today’s OpenAI announcement is a consolidation. The standalone Codex app is merging into a single ChatGPT desktop app that puts Chat, Work, and Codex side by side on every plan, including Free — and OpenAI is beginning to sunset Atlas, its standalone browser. The strategy reads plainly: one surface, three modes of the same agent, and no more side projects competing with it. If you built habits (or workflows) around Atlas, the clock is now running.
GPT-5.6 and a New Voice Mode Round Out the Launch
Underneath ChatGPT Work sits GPT-5.6, OpenAI’s new flagship, released today with the usual claims of better long-horizon task reliability — the specific capability an hours-long agent lives or dies on. ChatGPT Voice also got rebuilt on a new model line, GPT-Live-1 for paid users and GPT-Live-1 mini for free users. Voice matters more than it looks here: an agent that works for hours needs a low-friction way to take instructions and give status updates, and talking beats typing for both.
Quick Hits
- Pricing structures diverge: ChatGPT Work is usage-metered against your plan allowance; Gemini Spark requires Google’s top AI Ultra tier. The “agent that does your work” category is arriving with very different bills attached.
- Security posture: OpenAI says its auto-review layer blocked 100% of prompt-injection extraction attempts in red-teaming — a controlled-test number, not a production guarantee, and worth watching once 1,400 connectors meet real inboxes.
- Cross-device agents are next: Google says Spark will soon take assignments across devices — ask on your phone, have it work through files on your Mac.
What This Means
Put today next to last week and the shape of the second half of 2026 is set. Anthropic made agents cheaper to run with Sonnet 5, Microsoft made Copilot a permanent line item, and now OpenAI and Google have both shipped their versions of the same product: an agent that connects to your work apps and delivers finished output. The differences that remain are the ones that will decide who wins — how it’s priced (metered allowances vs. flat tiers), how it’s gated (approval flows vs. full autonomy), and whose integrations actually hold up when the task runs for hours unattended.
That last one is the real test. Every player can now claim a connector directory; far fewer can claim an agent that reliably finishes a job across those connectors without babysitting. Products like Carly, which have been doing scheduled, always-on work across 200+ integrations since before “agent mode” was a launch-event slide, get judged on exactly that bar — and starting today, so does everyone else.
Ready to automate your busywork?
Carly schedules, researches, and briefs you—so you can focus on what matters.
See what people say
"Before Carly, I relied on a Calendly link, but the whole process felt impersonal and not very professional. Carly changed that by handling all the back-and-forth, so I'm no longer stuck in endless email threads trying to line up schedules.
Now Carly reaches out to candidates, shares my real-time availability, lets them pick a slot, then sends a Zoom link and drops it straight into my calendar. She sends reminders to both of us before each call, which has significantly reduced no-shows and last-minute confusion.
On top of scheduling, Carly acts like a full executive assistant, sending me my schedule the night before so I can prepare for each call. It reminds me of the old x.ai assistant, but Carly is noticeably smarter, faster, and better suited to my healthcare recruitment business."
