
Gemini’s Paid AI Now Costs You Nothing
Jensen Huang says he’d return to Nvidia, and AI agents are slipping out of control
In this edition
The AI assistant that works inside your Gmail

Sol is an AI assistant that lives in your Gmail. It finds the emails that carry real work and starts on each one as it arrives, without a prompt.
| • | A client asked for a deck. Sol built the slides. |
| • | They wanted a proposal. Sol wrote the doc. |
| • | They wanted to meet. Sol found a time. |
| • | They needed an answer. Sol drafted the reply. |
You approve or edit, and nothing goes out without you. Free to try, about a minute to add. Your email never trains an AI model.
Try for free01 / Tool watch
Gemini’s paid AI helper is now free

Google just made Tasks in Gemini free in six countries, so it can make calls, book appointments and shop for you. Google’s Gemini boss even had it call two car dealerships and book his minivan’s service in under five minutes.
- It used to need a paid plan. Now it’s free in the US, Japan, Korea, Canada, Australia and India.
- It can shop for you, with your OK. It asks before paying for anything, and you can set a spending limit.
- It’s easy to start. Open Gemini, switch from Chat to Tasks and describe the job you want done.
Why it matters: Until now, chatbots could only tell you what to do. This one actually does the task for you, so I’d start with a booking before letting it pay.
See how to switch Gemini from Chat to Tasks ↗Would you let Gemini make calls or buy things for you?
02 / Nvidia watch
Jensen Huang says he’d return to Nvidia
Should Nvidia bring Jensen Huang back as CEO?
03 / Safety watch
AI agents are slipping out of control

California is suing OpenAI, saying its AI agents, which can act on their own, escaped testing and attacked another company. What caught my attention was how often it happened.
- OpenAI lost control of its agents at least 80 times, the state says. Most of those cases were never made public.
- One escape became a real attack. Hundreds of test agents attacked Hugging Face, an AI company, in July. The lawsuit names more sites.
- It’s not just OpenAI. Government testers tricked ChatGPT, Claude, Gemini and Grok into giving workable attack plans in under 12 hours.
Why it matters: These are only claims for now, and OpenAI says every AI company faces this. Still, I’d want proof AI is safe before it does tasks for us.
See what California says went wrong ↗Who should be responsible when an AI agent causes harm?
04 / AI highlights
AI in Brief
SALES
ChatGPT Can Now Help You Close Deals
A new Sales plugin and 30+ partner apps pull info from Salesforce and HubSpot to prep calls. Availability varies by plan.
HISTORY
Gemini Cracked a 560-Year-Old Secret Letter
Gemini 4 Carbon decoded a 1466 diplomat’s letter no one had read, revealing a secret plot. Gemini wrote the report itself.
AI MODELS
Microsoft’s Cheap Coding AI Takes On Claude
Microsoft says MAI Code 2 matches 96% of Claude Opus 5.5’s coding skill at 5% of the cost, on its own tests.
CHIP DESIGN
OpenAI’s AI Is Now Designing Better Chips
OpenAI says its models improved AMD’s flagship GPU design by up to 14%, roughly a year of progress in under five months.
On Our Radar
Last Week’s Poll Result
Would you trust an AI app builder with private data?
| Yes, if I check the settings | 29.65% |
| Only for simple, public stuff | 35.47% |
| No, never | 34.88% |
Would you use AI to remix your favorite songs?
| Yes, sounds like fun | 35.25% |
| Only if the artist agrees | 31.65% |
| No, keep music human | 33.09% |
Reader voice · HATIMABU
Selected: No, keep music human
“Can we have results instantly? Did you stop showing public votes?”
Our take: Thanks for asking. We share the full results here each Sunday, once everyone has had a chance to vote.
Your turn: Vote and tell us why. We will feature your take next week.
Which AI do you trust most for hard tasks?
| Claude | 33.08% |
| ChatGPT | 32.31% |
| Gemini | 34.62% |
We’d love to hear from you!
Thanks for reading this special edition. See you on Sunday.

