Need to Know News - September 30th, 2026
In this week's Need to Know News edition:
🤖 Is Google officially BACK in the AI race with Gemini 4?... only a select group have the chance to find out right now.
🤖 OpenAI crammed more than 20 launches into one DevDay... and the biggest gives you an agent with its own computer that works while you're gone.
🤖 ElevenLabs' new voices obey script cues like [door slams]... and the fast one picks up a call in under a sixth of a second.
And a whole lot more!
At DevDay, OpenAI Turned ChatGPT Into an Office Where Agents Work For You
OpenAI packed more than 20 announcements into DevDay 2026 on Tuesday, and nearly every one pushes ChatGPT from a chat box toward a full-blown workplace.
The big idea is agents that take on ongoing jobs and keep working alongside you. OpenAI also threw ChatGPT open to outside developers, who can now build their own experiences for its 1.2 billion weekly users. These Five launches carry most of the weight:
- Dots, the agents that don't clock out: Dots are OpenAI's direct competitor to Grok Bot and Muse. Each one runs on GPT-6 Astra with its own cloud computer and plugs into more than 4,000 apps. The longer you work together, the more it learns your preferences, and sometimes it acts before you ask.
One early tester's dot noticed he'd forgotten to invoice a publication, prepared the invoice, and sent it once he approved. Dots are rolling out to Pro and Business Premium in eligible markets, and your first one comes at no extra cost.
- GPT-6.1 Sol, the new workhorse model: Powering that kind of work cheaply is GPT-6.1 Sol. It nearly matches the flagship GPT-6 Astra on coding, computer use, and professional tasks at a fifth of Astra's price.
On AutomationBench, a test of multi-step business workflows, it beat Anthropic's Opus 5.5 by 2.2 points at roughly a third of the cost. - Codex in the cloud, coding that doesn't need your laptop: Codex, OpenAI's coding agent, now works from the cloud on any device, phone included. Point it at your GitHub repositories once, and it builds a reusable workspace your whole team can share.
Each task then keeps running while your computer sleeps. It's available on paid plans from Plus on up. - ChatGPT Space, where people and agents share the work: All that output needs a home, and Space replaces ChatGPT's Library to be it. Teammates, ChatGPT, and your dot edit the same pages, a new document type that can update itself from Slack, email, or your calendar.
OpenAI's example is a to-do list that keeps its own due dates current. Collaborative slides that export to PowerPoint or Google Slides arrive in the coming weeks, and Space is live on Pro, Business, and Enterprise.
- Everything else worth knowing: Business and Enterprise teams can now tag @ChatGPT in Slack and Microsoft Teams, and colleagues can chime in without their own ChatGPT license. Heavy users also get Pro 500, a new plan with 25 times the Plus usage allowance.
🎯 Why You Need to Know: Chat is graduating from a tab you open to a coworker that keeps going after you leave. Dots, team tasks, and self-updating pages all run on schedules or triggers like a new email or Slack message. OpenAI's own example is the weekly project update, which a team task can now gather and build on its own.
⚡ Your Move: On Pro or Business Premium? Create your dot in the ChatGPT desktop app and connect the apps you live in. Then hand it one recurring chore, like your weekly campaign recap. Launch month comes with extended limits for deeper work, so now's when you get the most room to test it.
Gemini 4 Argon Arrives, but Only Cyber Defenders
Google's new frontier model, Gemini 4 Argon, landed today and it's built to hold reasoning together across long, multi-step jobs. Access starts only with trusted cyber defenders in Google's Fairwind Program.
📋 The Details: The headline upgrade is headroom. Argon can write up to 1 million tokens (AI's units of text) in a single answer, up from 64,000. It also set a new high of 77.9% on DeepSWE, a real-world coding test. Its introductory price, once it opens, is $2 per million tokens in and $10 out.
🎯 Why You Need to Know: Argon leads the Vals Index, which grades finance, legal, and tax work, and it tops Zapier's business-workflow test, AutomationBench, at 51.3%. So the dense research and busywork you'd hand an AI is where Google says it's strongest.
📡 Watch For: The rollout to paid API customers and Google AI Ultra subscribers, who're next. Google hasn't named a date, only "as soon as possible."
🧵 The Thread: Argon's intro price matches GPT-6.1 Sol and Claude Sonnet 5.5, launched the two days prior.
🚀 WATCH: How These AI Copy Bots Are Producing World-Class Sales Copy 50X Faster Than Even The "BEST" Copywriters On The Market…
(Plus… They Don't Get Sick, Miss Deadlines, Or Ask For Raises Either!)
Watch the full AI Copywriting Tell-All Video Here
HeyGen Survey: AI Avatars Got a Warmer Welcome Than Skeptics Predicted
Small business owners are sitting on a pile of videos nobody will ever see. HeyGen, which sells AI avatar video, surveyed more than 1,000 of them for its 2026 avatar report. It found 71.6% have recorded a business video and never posted it... usually because they didn't think they looked professional enough.
📋 The Details: You'd think an avatar would make that worse, and 50.7% of non-users worry it'll look fake. The owners already using one report a friendlier crowd, rating their audience's reaction 3.49 out of 5 where skeptics predicted 3.06. Baby Boomers are the eager ones, with 64.9% on board against 40.7% of Gen Z.
🎯 Why You Need to Know: Authenticity is the bar. Asked what earns their trust, the owners ranked looking and sounding real first at 64.6%, way ahead of disclosing AI at 39.9%. An avatar reciting lines you'd never say flunks the test that counts.
⚡ Your Move: Read any avatar script out loud first and cut every line that doesn't sound like you. Then answer the comments yourself. For 41.2%, pairing avatars with real human contact earns trust.
ElevenLabs Rolls Out Eleven v4 and a Turbo Version for Live Calls
ElevenLabs has a new generation of AI voices out, and the upgrade is emotion you can direct line by line. Eleven v4 is its most emotive model yet, and Eleven v4 Turbo carries that range into live calls. Both read a script like a voice actor, knowing who's talking and what just happened.
📋 The Details: You direct the performance right in the script with tags like [laughs] or [door slams], and v4 follows them more reliably than v3 did. Turbo, meanwhile, starts talking in about 150 milliseconds. Professional Voice Clones are back too... those near-perfect copies of a real person's voice went missing in v3. A free plan covers 10,000 credits a month, and paid plans start at $6.
🎯 Why You Need to Know: One cloned voice now works across both models. So the founder narrating your ad can also answer when customers call your AI agent.
⚡ Your Move: Creator+ plans get triple credits until October 12. Burn a few rerecording an old ad read in v4 with some direction tags, then play both for someone who can't tell which is which.
Manus 2.0 Rebuilds Its AI Agent and Spins Off a New App
Manus 2.0 landed Monday, built to carry an AI first draft all the way to work you'd actually ship. The groundwork is a new engine that pulls in specialized tools only when a project needs them.
The finishing happens in Manus Studio, the renamed desktop app, which now has professional editors for video and games. Manus also spun off Cue, a new app for personal agents on your phone.
📋 The Details: That engine, Cascade, ran 32% cheaper than the old one in a tested setup. It also powers Automations, which kick off a task when something happens in a connected app... a new email, say, or a swing in ad performance. Everything's live on web, desktop, and mobile.
Cue is free in early access with code MEETCUE.
🎯 Why You Need to Know: Your ad account just picked up a night-shift watcher. An Automation can act the second your numbers move, and Studio's video timeline keeps your own footage in the final cut.
⚡ Your Move: Tell Manus in one prompt to watch your ad performance and draft a new variation when it dips.
Meta's Muse Gets a Small-Business Mode That Reads Your Books
Meta's Muse agent is clocking in at small businesses. Muse for Small Business lets owners hand it a goal, like finding new customers, and it works across their existing tools. Owners told Meta they're short on hours, so Muse skips the getting-to-know-you phase and starts out knowing what they sell.
📋 The Details: The connectors cover the usual small-business stack, from QuickBooks and Shopify to Stripe and Canva. Hook up your Facebook Pages and Meta ad accounts, and Muse learns your brand voice. Meta keeps it on a short leash... nothing goes out or gets spent without your okay. And for most jobs, it's free.
🎯 Why You Need to Know: Meta's sample prompt is a planning meeting in one line. It asks Muse to review a year of sales, campaigns, and social, then write next year's growth plan. With your books and ad account plugged in, it pulls numbers that normally sit in separate apps.
⚡ Your Move: Connect only your Meta ad account first and ask Muse what's working in your ads. Kick the tires on its draft campaign before you let it near QuickBooks.
At unBoxed, Amazon Folds Its Ad Tools Into One AI Agent
Amazon Ads brought nine announcements to unBoxed 2026 in San Francisco, and most hand the grunt work of running campaigns to AI you talk to. The hub is Amazon Ads Agent, which merges the DSP (Amazon's programmatic ad-buying platform) and Ads Console into one place, so you type requests instead of clicking through settings.
📋 The Details: Full-Funnel Campaigns is the standout if you've only run sponsored ads. Set a goal, and AI builds the campaign across video, display, and audio. Amazon says beta testers saw 67% higher long-term return on ad spend and a 29% lower cost to land new-to-brand customers.
🎯 Why You Need to Know: Branded Conversations hands brands the mic inside Alexa for Shopping. You feed it your own explanations, like which cleanser suits oily skin and which suits a beginner. When a shopper asks for help building a routine, your advice rides along in the answer.
📡 Watch For: Full-Funnel numbers from brands outside the beta. That 67% came from Amazon's own testers, so see it hold for everyday advertisers before betting a budget.
Microsoft's New Copilot Is Built Around Agents That Work Without You
Microsoft rolled out a rebuilt Copilot on Friday, and it's designed to finish whole jobs, like an RFP response or a launch kit, on its own. You hand those jobs off from Home, which puts quick chat, delegated tasks, and the Office apps on one screen.
When the job needs a tool you don't have, Code builds one from a plain description. And when work stretches over days, Autopilot, formerly Scout, picks it back up without a prompt.
📋 The Details: Home and Code start reaching Frontier, Microsoft's early-access program, over the coming weeks, while Autopilot hits private preview at month's end. Read the pricing twice, though. Everyday chat stays on a flat per-user license, but Cowork, Code, and Autopilot all bill by usage.
🎯 Why You Need to Know: Delegation now comes with a meter running. The long jobs you hand off bill separately from your flat seat, so admins will want spending policies ready before the team starts offloading.
📡 Watch For: Code's preview for Microsoft 365 Premium and Pro subscribers, promised for later this year. That's when marketers outside big-company rollouts get their hands on it.
Adobe Arrives in Gemini and Adds Acrobat to Its Claude Plugin
Adobe is chasing customers into the chat window, one chatbot at a time. On Thursday it landed in Google Gemini, joining ChatGPT, Claude, Slack, and Copilot.
That means Photoshop, Lightroom, Express, and Firefly now take orders from a Gemini conversation. Claude's Adobe plugin grew the same day, adding Acrobat and hands-on editing.
📋 The Details: The Gemini version is rolling out globally on every plan. To switch it on, head to Personal Intelligence and pick Adobe from your connected apps. Over in Claude, the plugin packs 80-plus tools and a new layer editor, so you can change a font or swap an image in an Express design without regenerating the whole thing.
🎯 Why You Need to Know: Adobe's examples read like a small shop's to-do list. One fixes the lighting and cropping across a batch of product photos, and another spins a single design into versions for every social channel. And Gemini reaches more than a billion people a month.
⚡ Your Move: Connect Adobe in Gemini and upload five product photos with one note about the look you want. Line the results up against your usual edits.
Claude Sonnet 5.5 Nearly Matches Opus on Office Work at Half the Price
Anthropic's everyday Claude model, Sonnet, hit version 5.5 on Monday, running 30%+ faster than Sonnet 5 and breathing down Opus's neck. On GDPval-AA, a test of real work across 44 occupations, Sonnet scored 1844 to Opus 5.5's 1846. That's two points apart... at half the price per token.
📋 The Details: The sticker price didn't move from Sonnet 5, still $2 per million tokens in and $10 out. But Anthropic says the new model burns so many fewer tokens that a task costs up to 30% less. It's live in the Claude apps and on every major cloud. For sprawling, open-ended work, Anthropic still sends you to Opus.
🎯 Why You Need to Know: Decks are where you'll feel it. Anthropic fed Sonnet 5.5 a public company's earnings materials and a slide template, then asked for a 10-slide operating review. Two experts read the first draft and called it ready to send as is.
📡 Watch For: Claude Haiku 5.5, the budget model for high-volume work, due in the coming weeks. It'll tell you whether these savings reach the cheapest tier.
Thanks for reading.
Until next time!
The AI Marketers
P.S. Help shape the future of this newsletter – take a short 2-minute survey so we can deliver even better AI marketing insights, prompts, and tools.
[Take Survey Here]
