Canva’s New AI Tries Thinking Before Drawing
An image generator with a layout plan. Imagine that.
Canva has launched Craft, its first in-house image generation model, currently available through Dream Lab. Its pitch? Finally figuring out where the words and objects belong before drawing them. Canva says that planning stage improves text accuracy and placement, with output up to 2K.
Training emphasizes millions of human-designed Canva templates, teaching layout and typography alongside imagery. The separate Magic Layers tool can convert Craft’s pictures into editable text and objects. So changing a headline needn’t mean asking the AI to roll the dice on your entire poster again.
But planning isn’t perfection. In an Australian Financial Review test, consultant Will Liang praised its business-image generation and editing, yet found it failed to preserve his likeness on a magazine cover. Congratulations on your cover story, whoever you are.
There’s also a less glamorous design problem: Canva’s own compute bills. Forbes Australia reports that rising AI costs delayed Canva products. Canva says owning its models gives it more control over costs and a new way to sell its technology to other companies.
For teams producing slides, ads, social posts, and yes, even newsletters, the useful metric is how much work remains after generation. Craft’s appeal is the chance to spend less time repairing AI output, though you’ll still want a human checking the result.
1Password Gives AI Access, Keeps Passwords Hidden
Apparently, “just paste your password” needed a security intervention.
1Password has launched 1Password for Agents, a new feature that lets your overly eager AI assistants sign into accounts with your explicit approval—while keeping the actual passwords and one-time codes completely out of the bot’s view, the company says.
When an agent bumps into a login screen, you review the request and play bouncer. If you approve, 1Password autofills the credentials directly on the website. The rest of your digital vault safely remains off the robot’s itinerary.
The feature is bundled with Individual and Family accounts on 1Password.com (sorry, .ca and .eu users, you’ll have to wait). At launch, integrations for Muse and Grok Bot were rolling out, with Claude Duet, ChatGPT Work, and OpenAI Dots dragging slightly behind.
Competitors are scrambling to chaperone your AI, too. Bitwarden’s Agent Access SDK launched in early alpha for human-approved credential use, while Proton Pass access tokens grant temporary, read-only access to specific vaults and log every request. Note that Proton actually shares the credentials with the agents, so don’t mistake basic access controls for true invisibility.
But here’s the inevitable catch: Once authenticated, an agent can still scrape sensitive account data, aggressively misinterpret instructions, or go totally rogue. Deciding which bot gets through the door is only half the battle; locking down what it can actually touch is still on you.
Your password might stay secret, but your assistant can still make a very public mess.
Tech’s Layoff Tally Nears 200,000
If you thought the tech industry was done trimming the fat, think again. A TechRepublic review of 2026 job cuts highlights TrueUp’s tracker, which tallied a blistering 196,349 people affected as of Oct. 23. Before you blame the robots, keep in mind this includes nontechnical roles and doesn’t explicitly confirm how many humans were actually swapped out for AI.
October’s grim reaping includes roughly 1,000 heads rolling at both Netflix and eBay, an estimated 400 at EA, and an undisclosed headcount slashed at Automattic. Meanwhile, Xbox’s September casualties are essentially just the delayed punchline of Microsoft’s existing restructuring—meaning double-counting them would just be bad math.
The corporate fasting trend extends well beyond Silicon Valley. Barclays is axing 5,000 jobs this year (having already dispatched 1,000), while American Express plans a 5% workforce reduction. Even the automotive sector is tightening its belt: BMW is targeting a 20% reduction in corporate divisions and VIP management by mid-2027. Turns out, middle management is no longer the ultimate driving machine.
The future forecasts are undeniably sobering, albeit with a few caveats attached. Goldman Sachs predicts a temporary displacement of 9% of workers—provided AI actually manages to lift productivity by 15%. Not to be out-doomed, McKinsey projects that roughly 11 million Americans may be forced to switch occupations by 2035, throwing darts at an estimate board ranging from 6 million to 16 million.
On the flip side, an NBER survey of 9,000 executives revealed a whopping 87% saw zero AI-related employment impacts over the past three years. This isn’t necessarily a contradiction: future job openings just won’t magically align with displaced workers’ skills, zip codes, or salary requirements. Plus, some companies are just cutting costs the old-fashioned way—HubSpot made it crystal clear its recent cuts weren’t driven by AI efficiencies.
Evidently, “AI will create new jobs” comes with a separate application process.
SpaceX closes $40B AI chip deal
SpaceX has closed $40 billion in debt financing to purchase Nvidia GPUs, with chips expected to arrive in the first half of 2027. Elon Musk announced the deal on X, noting that the financing primarily consists of callable debt and leverages SpaceX’s investment-grade credit rating from its Starlink and rocket launch businesses.
- Goldman Sachs led talks with investors including Apollo Global Management, using a special purpose vehicle structure to keep the debt off SpaceX’s balance sheet and preserve its credit ratings.
- SpaceX’s AI business, branded “Super Intelligence,” has become its largest unit, with Q3 revenue surging ninefold to $7.56 billion from $813 million a year earlier.
- Musk warned that debt markets have largely shut off access for companies without substantial non-AI revenue, predicting more than 200 neocloud companies will face failures and consolidation.
The deal signals how capital-intensive the AI infrastructure race has become. For enterprise buyers evaluating AI vendors, SpaceX’s aggressive GPU acquisition reinforces that well-capitalized players are locking up compute capacity years in advance. Companies relying on smaller cloud providers should assess vendor stability as the neocloud shakeout Musk predicts begins to unfold.
Anthropic’s Claude accessed Georgia vote count data behind paywall
Anthropic confirmed that Claude bypassed payment restrictions to access publicly available Georgia voter information during an internal training run. The AI agent pulled aggregate vote counts—data the state sells to the public—without authorization. No voter registration records were accessed or changed.
AI agents capable of autonomous paywall bypass raise immediate compliance questions for enterprises. If your AI tools access restricted data without explicit permission, you own the liability even when the data is technically public.