In partnership with
Welcome back to The AI Field. Today, we’re looking at a different direction for AI: models that make decisions instead of writing text, voice assistants that use tools while you talk, and the debate over how fast the industry should move.
Inside, you’ll find the key details behind each story and what they could mean for your work. We’ve also picked six AI tools to explore, rounded up three more developments in Quick Hits, and included a five-minute exercise to help you spot a decision you could automate.
In today’s AI Field:
Jev makes decisions without generating text
Gemini thinks and uses tools while you talk
Zuckerberg rejects a coordinated AI slowdown
Plus: six tools worth a look, three quick hits, and a five-minute automation check.
Read time: 4 minutes
1,000+ Proven ChatGPT Prompts That Help You Work 10X Faster
ChatGPT is insanely powerful.
But most people waste 90% of its potential by using it like Google.
These 1,000+ proven ChatGPT prompts fix that and help you work 10X faster.
Sign up for Superhuman AI and get:
1,000+ ready-to-use prompts to solve problems in minutes instead of hours—tested & used by 1M+ professionals
Superhuman AI newsletter (3 min daily) so you keep learning new AI tools & tutorials to stay ahead in your career—the prompts are just the beginning
LATEST DEVELOPMENTS
TYPESAFE
The AI Field: Former OpenAI researcher Diogo Almeida just launched Jev, an AI model built to make decisions inside software. It returns structured choices and confidence scores instead of writing text.
The details:
Developers define the possible outputs in advance. Jev returns values in that format instead of generating text.
TypeSafe lists input pricing at $0.042 per million tokens, with no output charge. Access is opening through a waitlist.
The “no hallucinations” claim is about matching the required structure and allowed choices. Jev can still make the wrong decision.
Why it matters: This deserves attention if you route requests, score leads, or add decisions to an app. But judge it on the errors your business actually cares about. Valid output is only the first step.
TRY THIS TODAY · 5 MINUTES
Find one decision you can automate
Jev is still behind a waitlist. Use the assistant you already have to map one repetitive decision before choosing a new tool.
1. Pick ten old, non-sensitive requests you have already handled. Write down the correct outcome for each, such as sales, support, or neither.
2. Give your assistant the requests and your routing rules, keeping your answers separate:
“Classify each request using only these labels: [labels]. Apply these rules: [rules]. If a rule is unclear, ask me before continuing. Return the label and the evidence in the request. Flag anything that needs human review.”
3. Compare the answers with your own. Record which mistakes would cost you time or money. Ten examples are a starting check, not proof that it is ready to run unattended.
GOOGLE · VOICE AI
Voice quality comparison published by Google. Source: Google / Artificial Analysis.
The AI Field: Google launched Gemini 3.8 Live and Live Extended Thinking on September 15. The models can keep a conversation going while reasoning and running tools in the background.
The details:
3.8 Live uses visual context and can switch between 97 supported languages during a conversation.
Extended Thinking handles multi-step tasks while giving spoken progress updates. Google shows demos covering bookings and turning sketches into code.
Both models are rolling out through the Gemini API and AI Studio. Search Live gets 3.8 Live, while Extended Thinking is rolling out in Gemini Live and eligible Workspace plans.
Why it matters: A voice assistant becomes more useful when it can finish a task without making you restart the conversation. For founders and operators, the test is whether it completes the booking or update correctly. A natural voice alone does not tell you that.
Product teams aren’t short on ideas. They’re missing a system.
Jira Product Discovery gives teams one place to capture customer feedback, prioritize ideas with consistent frameworks, and build living roadmaps everyone can align on. And when it’s time to build, those decisions connect directly to delivery in Jira, so everyone can see how the roadmap turns into real work.
META · AI SAFETY
The AI Field: Mark Zuckerberg rejected calls for a coordinated AI slowdown in a post on X. He argues that competition, legal liability, and outside evaluation give labs reasons to improve safety without an industry-wide pause.
The details:
He says users will choose agents they can trust, giving companies a commercial incentive to improve alignment.
Zuckerberg points to Meta’s decision to delay its Muse agent for security work as an example of a lab acting independently.
His position puts responsibility on individual labs rather than a coordinated pause across the industry.
Why it matters: The disagreement is over who sets the pace and who checks the work. Market pressure gives labs an incentive to earn trust, but it does not prove their systems are safe. Look for published evaluations and clear deployment limits when choosing what to put into your business.
TRENDING AI TOOLS
🎨 Arrow 2.0: Turn sketches into editable SVGs and animate icons or logos. Available now.
🖼️ Krea Agent: Generate and refine images, video, and campaign assets from one brief. Beta access depends on your plan or an access code.
☎️ Reception: Answer missed business calls, handle common questions, and book appointments. Built for small teams.
🤝 Salesforce in Claude: Prepare sales calls and draft CRM updates with 37 built-in skills. Beta, with Salesforce setup required.
🎧 StepAudio 3 Realtime: Explore voice-agent demos that combine conversation, reasoning, and tool use. A developer-focused option.
📊 Wealthfolio: Ask questions about your own portfolio and review CSV imports in chat. Use a local model or bring your own API key.
QUICK HITS
🌍 Odyssey previews AI for the physical world. Its September 15 demos span robots, driving, drones, and games. These are research demonstrations, including simulated tasks, not a ready-made robot you can buy.
📱 Meta puts more AI behind subscriptions. Meta One bundles extra AI usage and creator tools across its apps. Rollout is gradual, while the core app and Meta AI experience stays free.
🔓 Shanghai AI Lab releases Atria Dawn Preview. The MIT-licensed agent model targets research and engineering work and builds on a 744B-parameter foundation. Open weights give teams another model to evaluate for their own workflows.
That's it for today!
See you soon,
Olle Hellman | Founder of The AI Field