Need to Know News - August 13th, 2026
In this week's Need to Know News edition:
🤖 Your AI agent can now carry a real corporate card... and it can run its own purchases end to end, receipts and all.
🤖 Grok Bot takes a whole project off your hands and runs it start to finish... then hands it back done while you're busy elsewhere.
🤖 AI Anchors Are Now Reading the News Live on X...
And a whole lot more!
Mercury Gives AI Agents Their Own Company Cards
Founders keep handing work to AI agents, so Mercury built them a card of their own. On Tuesday it launched Mercury Spend: budgets, self-enforcing policies, and a card an agent can swipe. A person sets the ceiling, the agent spends under it, no way around.
📋 The Details: The agent cards are trackable, auditable, cancellable, and only a human can create one. Mercury declines any over-budget charge at checkout and freezes a card the moment a required receipt goes missing. Receipts collect on their own from Gmail or a texted photo. It's live now for Mercury business customers.
🎯 Why You Need to Know: Handing an agent a credit line used to mean trusting it with a live card number and hoping. Now the limits ride on the card, so an agent buying software or ads stays in the fence you drew, and every charge traces to who set it loose.
⚡ Your Move: If an agent already touches money in your business, even a subscription it renews, give it a dedicated card with a hard cap and a merchant restriction before you scale up.
AI Anchors Are Now Reading the News Live on X
The anchor delivering your headlines may not have a face that exists. On August 12, Mirage, the firm once called Captions, put what it bills as the world's first AI news network on the air, streaming nonstop on X via @TryMirage, no human on camera.
📋 The Details: Others tried this and stalled: Channel 1 showed a reel in 2023, and state outlets in Kuwait, India, and China ran virtual presenters, each beside human broadcasts rather than a channel of its own. Mirage got here on about $175 million in funding, including a $75 million March round, and already makes AI actors for brands. It chose X to ride its live-video push.
🎯 Why You Need to Know: A synthetic anchor broadcasts in any language, nonstop, with no salary or studio behind it. Mirage already aims that pipeline at branded video, so the economics pulling it into news could reshape how cheap ad footage gets made.
📡 Watch For: Whether viewers stay once the novelty fades. It's the first big audience test of AI anchors, and retention numbers tell every brand whether a face nobody trusts yet can hold a crowd.
🚀 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
Grok Bot Is an AI Teammate That Signs Into Your Tools
xAI wants you to stop picturing Grok as a chatbot. Its new Grok Bot logs into your apps, works them like you would, and comes back when it needs a yes, running on its own computer around the clock, laptop open or not.
📋 The Details: Point several Bots at different jobs and watch them hand work between themselves instead of clearing every step. Show a Bot a workflow once and it saves the steps as a routine it reruns itself. It's on macOS and iOS, free to start, with plans from $120 to $200 a month.
🎯 Why You Need to Know: A tool that keeps its own logins and reruns a routine after one watch can take recurring web work off your plate, the report pulls and inbox triage you'd never script. It works inside your live accounts, so start on tasks you can afford to have go wrong.
⚡ Your Move: Take one weekly chore you run the same way each time, a standing report, say, and teach a Bot by running it once while it watches. Next week, see if it repeated on its own.
Grok 4.6 Ties GPT-5.6 on a Nine-Test Intelligence Score
xAI shipped more than a teammate this cycle. Grok 4.6, its new model, arrived level with GPT-5.6 Sol on the Artificial Analysis Intelligence Index, a nine-benchmark average, built for the long, multi-step agent work Grok Bot takes on.
📋 The Details: It's out now in Cursor and Grok Build, plus the API and partners like OpenRouter and Vercel, at $2 per million input tokens and $6 out, the fast variant double that. xAI is doubling included usage in both for the first week. It says the model is best at turning a rough idea into a working first version, and began self-checking during longer tasks.
🎯 Why You Need to Know: A model built to keep working through many steps is one you can hand a small project and walk away from, instead of steering every reply. The free week keeps the cost of trying it near zero.
⚡ Your Move: During the double-usage week, hand Grok 4.6 one real build, a small internal tool or landing page, and rate it on that first draft, which xAI says is where it gained most.
Claude Will Start Watermarking the Text It Writes for You
Anything you draft with Claude is about to carry a hidden signature. Anthropic confirmed on a support page it will watermark the text and files its models produce, meeting the EU's Transparency Code, effective August 2, which makes AI firms mark machine-made content so other systems can flag it.
📋 The Details: Every model shipped after August 2 watermarks text and files automatically, files using the open C2PA standard, older models to follow. The mark lives inside the text, so it survives a copy-paste and some editing. It spans the API, Claude, Claude Code, Cowork, and the Slack tool.
🎯 Why You Need to Know: If you write with Claude and pass the words off as your own, AI-writing detectors now have a clean signal to read. Anthropic applies it at the model level, so it shows on every Claude surface.
📡 Watch For: How much editing scrubs the mark off. Anthropic hasn't said, so treat anything you paste from Claude as traceable. Black Forest Labs, Google, Meta, Microsoft, OpenAI, and Synthesia signed the same EU code, so this won't stay Claude-only.
Describe the Report You Want, and Google Ads Builds It
Pulling a clean report out of Google Ads has meant knowing which dimensions to click. Google is handing that to a prompt box: describe what you want, and its new Dashboards turn your data into charts, each topped with a note on why the numbers moved.
📋 The Details: The prompt-built Dashboards are live in Google Ads now, coming to Analytics, all on Gemini. Google also put AI Overviews on the Analytics homepage: a rolling summary of what changed since you last signed in, pushed to your phone or email. A benchmarking view stacks your results against anonymized averages from similar businesses. It's in beta for English-language accounts.
🎯 Why You Need to Know: When a plain-English prompt returns the chart and the explanation, the reporting know-how you used to hire or learn stops gating who can read performance. The benchmarking view is sharper still, since it says whether a soft month is you or the whole category.
⚡ Your Move: Next time you'd open the report builder, use the prompt box: ask for your last 30 days by campaign, then pull benchmarking to see if you trail peers.
Amazon Says AI-Made Video Lifted WD-40's Daily Sales 50%
WD-40 had a new degreaser to launch in France with no sales history to lean on. By Amazon's account, the brand and its agency Bizon made the video ads with Amazon's own AI generator, ran them across search and display, and saw average daily sales climb over 50%, units tripling.
📋 The Details: Video Generator builds a video ad from the product images and details already on an Amazon listing, so the team skipped a shoot. They ran Sponsored Brands video in search plus online video for reach, then measured before and after launch to isolate the lift. Amazon says 88% of buyers were new to the brand, and the ad drew a 45% higher click-through rate than the category norm.
🎯 Why You Need to Know: These are Amazon's numbers for a tool Amazon sells, so weigh them accordingly... the move underneath holds: a brand with zero footage made ad-ready video from its own product page and shipped a campaign fast.
⚡ Your Move: If you sell on Amazon, run one product listing through Video Generator and put the clip behind a small Sponsored Brands test before your next launch.
Stagehand v4 Makes Browser Agents Faster and Harder to Hijack
The AI agents that fill forms and pull dashboards for you break more than demos let on. Browserbase blames distance: the agent keeps its own copy of the page, so when the page changes, it acts on a stale one. Stagehand v4 moves that into the browser to read the live page.
📋 The Details: With the logic next to the page, a batch of actions runs about 1.6x faster in Browserbase's test, and it ships TypeScript, Python, and Go kits together. Security got stricter too: an agent's reachable sites are enforced in the browser, so a page trying to make it leak data is blocked before the request leaves.
🎯 Why You Need to Know: You probably won't run Stagehand, but the agents sold to you run on frameworks like it, and one reading the live page, not a stale copy, fails less on what you'd delegate. The security fix matters more as those agents read pages a stranger could rig.
📡 Watch For: Watch your automation tools for signs they've moved this logic in-browser. When your vendor stops blaming flaky runs on the remote browser, that's the shift underneath.
Adobe Finds 'Taste' Is What Sets Small Brands Apart Now
When rivals can all wield the same AI tools, one edge still separates the winners, and Adobe says it's taste. In its survey of 850 US small business owners with Morning Consult, more than half said that when everyone holds the same software, the edge goes to whoever has the sharper eye.
📋 The Details: Brand voice weighs on these owners. 79% said a clear point of view is what makes people follow them, and nearly one in three would rather post nothing than sound off-brand. They use AI anyway, 66% already for marketing, yet only 6% published their last AI draft untouched; 48% reworked it heavily.
🎯 Why You Need to Know: Your customers are wading through look-alike AI content, so posts that sound unmistakably like you break through. The owners here edit their AI drafts hard to protect that voice, the habit that separates brand-building from adding to the pile.
📡 Watch For: "Made with taste" only counts if buyers reward it, and that's the number to watch. Adobe sells the editing tools, so trust your own: see whether your most hand-shaped posts still out-engage your fastest AI ones.
Meta Built a Free Agent Model Small Enough for Your Laptop
Most capable AI phones home to a data center to answer. Meta's new Muse Glimmer skips the trip... a 30-billion-parameter agent model, free under Apache 2.0, that runs on a Mac or PC with one consumer graphics card and no internet once loaded.
📋 The Details: Meta squeezed it to about 4-bit precision so it fits under 20GB, with headroom on a 24GB card, and drafts whole blocks of words at once for speed. It's trained for the messy parts of agent work: calling tools right, holding a plan across steps, and retrying a failed call instead of quitting. The weights are on Hugging Face today.
🎯 Why You Need to Know: An agent on your own machine keeps your prompts and files off outside servers and drops the running cost to nothing, so any always-on task you'd hesitated to send to the cloud gets easier to keep in-house. Meta says it holds up against similar-sized models from Google and Alibaba.
📡 Watch For: The one-click apps due in coming days. Meta says Ollama and LM Studio support lands shortly, and that's when running this locally stops needing a developer.
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]
