Special highlight from our network
Hybrid pricing is reshaping AI economics
AI agents don’t follow legacy SaaS models. 92.4% of AI companies now use hybrid pricing to better reflect agent value. Orb’s latest report analyzes 66 products to uncover:
Which pricing structures are gaining traction
How teams protect margins at scale
Why billing infrastructure is a hidden growth lever
These insights reveal how AI-native companies are monetizing more effectively.
How OpenAI’s AgentKit Teaches Machines to Work Together
Image: OpenAI
At Dev Day 2025, OpenAI unveiled AgentKit, a toolkit that lets developers build and deploy intelligent agents without heavy infrastructure. During demo, Christina Huang created a working agent in just eight minutes, showing what OpenAI described as a “production-ready agent workflow.”
Here’s what’s inside:
AgentBuilder offers a visual workspace where developers design an agent’s reasoning through nodes and logic flows.
ChatKit embeds conversational agents directly into apps, turning any product into an interactive interface.
Evals tracks reliability and performance through tracing, dataset testing, and prompt optimization.
Connector Registry links agents with APIs and enterprise systems, enabling real-world action within live workflows.
The release hints at a future where software is less about code and more about cognition systems that think, coordinate, and build on each other’s output. Sam Altman called it “a bridge from prototype to production.” In practice, it looks like a blueprint for turning intelligence into a raw material of work.
Google’s PASTA Learns to Read Your Taste in Images
Image Credit: Google
Google Research has developed PASTA, a reinforcement learning system that studies how people refine AI-generated images through feedback. Instead of guessing through endless prompts, users make simple choices, and the model learns what they actually mean by “good.”
How it works:
Learning through interaction: PASTA shows four image options per turn and updates its understanding based on what you pick.
Grounded in data: It was trained on 7,000 real user sessions and 30,000 simulated ones, building a map of how people express visual preference.
Reinforcement logic: The agent experiments with new prompts while tracking what consistently earns approval.
Proof in numbers: Testers preferred PASTA’s results in 85% of comparisons against baseline models.
Instead of acting like a paintbrush, PASTA behaves like a collaborator that studies your instincts and adapts in real time. Google has released its dataset and simulator for other researchers, turning this small experiment in preference learning into a shift toward AI that understands style, not just syntax.
The AI Workforce Paradox: Overcapacity and Scarcity
Image: BearingPoint
The World Economic Forum paints a restless picture of work in the age of AI. Automation is freeing up time faster than people can fill it, and at the same moment, companies are begging for skills that hardly exist yet. Millions are in motion, learning, unlearning, and trying to stay relevant in a world that won’t slow down.
Here’s what the data shows:
92% of executives say automation has already created up to 20% workforce overcapacity, leaving parts of their teams idle.
By 2028, nearly half expect overcapacity to exceed 30%.
94% of leaders say they can’t find enough people trained in AI-critical skills.
The most in-demand roles now include AI governance, agentic workflow design, and human–AI collaboration specialists.
Work is being rewritten, and the human side of that story is still unfinished. Companies are being urged to retrain workers instead of replacing them, to rebuild roles around curiosity and creative oversight, and to treat reskilling as infrastructure. Those who manage that balance will create organizations that feel alive rather than automated.
Google Puts Jules in the Command Line to Rival AI Coding Agents
Image Credit: Google
Google is pushing its AI coding agent Jules deeper into developer ecosystems with new integrations across terminals, CI/CD systems, and Slack. The update includes a command-line interface (CLI) and a public API, signaling Google’s intent to make Jules part of everyday engineering work rather than an experimental assistant.
Here’s what’s changing:
Jules Tools launches: The new CLI brings Jules directly into the terminal, allowing developers to assign and validate tasks without switching windows.
API goes public: Developers can now connect Jules to their IDEs or build custom workflows. Google says dedicated plug-ins for VS Code and others are in the works.
Scoped autonomy: Jules executes tasks independently once approved, unlike the more collaborative Gemini CLI.
Memory and mobility: The agent now retains user corrections and preferences, with mobile notifications planned for rollout.
With Anthropic, OpenAI, and Cursor all chasing the same goal, Jules’ entry into developers’ command lines makes the race feel less theoretical and more like the new normal of software work.
🚀 Boost your business with us. Advertise where 13M+ AI leaders engage!
🌟 Sign up for the first AI Hub in the world.
