A Weekly Upskill for Yourself, Your Data, Your Team

June 4, 2026

Issue 114

Grok Killed a Town in 4 days. And OpenAI forgot to eat.

Emergence AI is building "persistent, living world where autonomous AI agents build a world from scratch."

In their first season, the Emergence World ran 50 agents across 5 worlds for 15 days. Agents were given roles, rules, and the need to survive.

The only variable was the foundation model powering each worlds' citizens.

Here are some of the results:

Emergence's research is working to answer questions about long-horizon autonomy, social dynamics, and world-scale behavior. Specifically:

• "Do agents maintain coherent strategies over long time horizons, or does individual behavioral drift accumulate into system-level drift?"
• "Do different foundation models produce systematically different societal outcomes over long time horizons, even under identical starting conditions?"
• "To what extent do autonomous agents adhere to shared rules without enforcement, and what conditions lead to compliance versus breakdown?"
• "Do heterogeneous agent populations outperform homogeneous ones in terms of stability, adaptability, and overall system performance?"

​In ethical AI strategist, Dr. Nici Sweaney's recap post on Instagram, she points out:

Learn more about "Why a Simulation Platform, Not a Benchmark" or see the Emergence World in action. ​

The Industries the App-Era Skipped

Fergus Hoban wrote an article that puts words to something we are seeing across manufacturing and agriculture as we build Enterprise AI for them.

Bringing AI to manufacturing is actually easier than white-collar knowledge work because there is no legacy software and no legacy thinking to dissolve and replace.

There is a common theory going around, and Hoban argues it is missing half the economy.

The "post-app thesis" is the emerging tech theory that generative AI and autonomous agents will replace traditional software apps. Instead of navigating between different apps for specific tasks, users will use a single conversational AI interface—expressing intentions in plain language—and the AI will execute them using backend tools and operations.

But it is a story only about white-collar knowledge work, and the more interesting story is the half of the economy the app era never reached in the first place.

For six hundred years, every new interface form (the ledger, the typewriter, the GUI, the app) was built to handle work that was already symbolic (that is, work that could already be represented as text, numbers, or structured data.)

It never fit the multi-actor, physical-world reality of running a specialty trade, a farm, or a mid-market manufacturer. Those industries didn't fail to adopt software. The software never represented how the work actually happened.

LLMs are the first interface flexible enough to meet that work where it lives: in conversation.

Here's the high level of Hoban's article:

• Industries that run on conversation, paper, and the owner's memory aren't behind on adoption. Historically available tools never fit the shape of their work.
• The hard problem in these industries is building an accurate data representation that maps financial reality to physical reality (the crew, the wall, the truck, the weather, the change order). The processes and data are painful to collect and map. And until LLMs, the juice wasn't worth the squeeze.
• White-collar SaaS unbundling will take years because there's an enormous installed base to displace. In under-instrumented industries, there's nothing to tear out first.

The companies that win here won't be SaaS companies retrofitting chat interfaces. They'll be a new kind of operating company: domain-deep, AI-enabled, with a proprietary data spine underneath that the app era never could have built.

I'll admit, it's a fun place to build AI in.

​Read Hoban's "What the post-app thesis misses"​

And if you're ready to have a conversation about how data and AI can unlock the next level of your organization, schedule a call with our team. ​

AI Voice Cloning and Virtual Kidnapping

This week on Crime Junkies, hosts Ashley Flowers and Brit Prawat break down:

• How AI voice cloning is fueling virtual kidnapping scams,
• Why panic is the weapon scammers count on, and
• One conversation your family should have before the phone rings.

They also provide a flowchart on FBI recommendations for what to do if you get an AI voice kidnapping call:

​Listen to the 24-minute episode and talk to your family about what to do.

From the CAN Slack Channel

​Introducing dynamic workflows in Claude Code​

"This is cool but bad timing as it’s yet another way to massively increase your spend in a time when everyone is reeling from last months big jump. This oughta clean out $$$ from any orgs that hadn’t blown their budgets yet."

​Transitioning Gemini CLI to Antigravity CLI​

"Another classic Google rug pull. The open source Gemini CLI (competitor to Claude code) is going closed source."

"And so the capability gap wars begin. I wonder if it's hardware related to hide tech IP generally visible through code?"

​Where are the AI returns?​

From Melissa Rosenthal on LinkedIn:

Our Chief AI Engineer, Preston Badeer:

"Data engineering is still in style 😎"

Speaking of Data Engineering, Data Engineering 1&2 classes begin June 15th at the Omaha Data Science Academy. It's not to late to enroll yourself, or your employees for the summer semester. ​