Google Can't Spell, Agents Can Trade, and the Pope Has Notes

Welcome to Briefly AI. Today: Google's image generator can't spell its own name, AI agents are getting brokerage accounts, and new research shows AI adoption comes with a psychological price tag most companies aren't measuring.

Right. Let's start with the slightly embarrassing one.

Google's latest image generation model — part of the Gemini family — has a spelling problem. And not just any spelling problem. It struggles to spell "Google." Also "artificial," "intelligence," and more or less any word you ask it to render as text in an image. Now, this isn't new — text rendering has been a known weak spot for diffusion models since the start. But it's 2026, Google's supposedly at the frontier, and yet if you ask Imagen to generate a sign that says "Welcome to the AI Conference," you'll get something closer to "Welcmoe to teh IA Confrence."

Why does this matter? Two reasons. First, it's a reminder that even the most advanced AI systems have absurdly basic gaps — the kind that make you wonder what else they're confidently getting wrong. And second, it's a trust issue. If a model can't spell, users notice. And once they notice, they start questioning everything else it produces. Google's betting billions on AI search and AI-generated content. Spelling is table stakes.

Now, on a completely different note: AI agents can now lose your money for you.

Robinhood announced this week that it's opening its trading platform to AI agents. You can now create a separate account, fund it with a specific amount, and let an AI agent buy and sell stocks autonomously across the market. Robinhood's pitch is that this lets traders experiment with algorithmic strategies without needing to code or manage infrastructure. The agent does the trading. You just watch.

Here's the thing. This is either brilliant or terrifying, depending on your risk tolerance. On one hand, it democratises access to automated trading — something previously reserved for quant funds and people with serious technical chops. On the other hand, it's handing financial decision-making to software that, as we've just established, sometimes can't spell its own name. The question isn't whether an AI agent can execute trades. It's whether it can consistently make good ones. And whether the average Robinhood user understands the difference between "my AI lost 15% this month" and "my AI made a catastrophically bad bet I didn't understand."

What to watch: whether Robinhood builds in circuit breakers or loss limits by default, whether other brokerages follow suit, and — most importantly — whether the first high-profile AI trading disaster happens on Robinhood's platform or someone else's.

Right, this next one's more practical.

Remote — the payroll and HR startup — just announced it hit $300 million in annual recurring revenue and became cash-flow positive. Nothing shocking there. Except for this bit: they grew revenue per employee by 50% without adding headcount. How? AI adoption. Remote deployed AI across customer support, compliance workflows, onboarding, and document processing. The work still got done. They just didn't need to hire more people to do it.

Now, Remote's framing this as an efficiency win, which it is. But it's also a preview of what's coming for a lot of service businesses. If you can grow revenue 50% per employee using AI, the calculus around hiring changes completely. You're not asking "should we hire another support agent?" You're asking "should we buy another AI seat license?" And that shift — from headcount to software subscriptions — has massive implications for employment, wage growth, and how entire industries are structured.

The unresolved question is whether this is a productivity unlock or just cost-cutting dressed up in AI language. Remote's revenue grew, so this wasn't pure redundancy. But if every SaaS company follows this playbook, we're looking at an economy where growth no longer reliably translates into jobs.

And finally, a bit of research that's worth sitting with.

Harvard Business Review published new findings this week on what they're calling "psychological debt" — the hidden mental health cost of adopting AI at work. The research identifies six specific effects: cognitive offloading, where people stop thinking through problems themselves; reduced autonomy, where decision-making shifts to the machine; diminished competence, where skills atrophy from disuse; weakened social connection, as AI handles tasks that used to involve collaboration; credibility loss, when colleagues start questioning whether your work is yours; and identity threat, where people feel their professional value eroding.

Now, none of this is shocking if you've been paying attention. But it's the first time it's been formalised into a framework that companies can actually measure. And that matters, because right now most organisations are tracking AI adoption like it's a pure upside story — faster outputs, lower costs, higher productivity. What they're not tracking is whether their people feel more anxious, less competent, or disconnected from their work.

Here's the bit that should worry anyone rolling out AI tools at scale: psychological debt compounds. It's not just "people feel weird about AI for a few months and then adjust." It's "people stop trusting their own judgment, stop investing in their skills, and eventually stop feeling like their work matters." If that sounds dramatic, go read the research. The data's there.

What to watch: whether any large employer starts measuring psychological debt alongside productivity gains, whether this framework gets picked up by HR or stays in academic journals, and whether the companies currently sprinting toward full AI adoption hit a morale wall in 12 months.

That's your lot for today. AI that can't spell, AI that can trade stocks, AI that boosts revenue without hiring, and new evidence that AI adoption has a psychological cost most companies aren't accounting for. If any of that was useful, pass it on. If not, blame the machine. See you next time.