Tilman's Blog on AI at Work in IO

Warning: Humans in the loop

Mon, 02 Nov 2026 00:00:00 +0000

By the time AI has the capabilities to be a threat to humanity on its own, it may be our only salvation.

As the velocity of AI abilities increases, so too does the fear that it will go rogue and potentially have the “inclination” to take out humanity—either intentionally or unintentionally (see the paperclip maximizer problem [1]). I’ve read a bit of science fiction and understand the apocalyptic premise of many popular stories. However, the current estimates of a doomsday scenario are all over the place, and almost all seem to place the fear at the doorstep of the AI itself. What they fail to explicitly state, though often imply, is that it’s not the AI that is dangerous; it’s the human in the loop.

New Job Skill: Agent Orchestration and Coordination

Fri, 30 Oct 2026 00:00:00 +0000

More than a year before the OpenClaw explosion, I told a group of about 40 I-O psychologists that agents would proliferate. You wouldn’t use one agent to tackle a problem, you’d use many, and their numbers would grow exponentially. I explained that they would evolve into agents designed with narrow skill focus that would function something like the brain, with an orchestrator (prefrontal cortex) controlling the communication between agents to some extent. Well, here we are. Welcome to the event horizon of the singularity?

What is intelligence?

Thu, 29 Oct 2026 00:00:00 +0000

You’ll get the most confident answer to this question from a recent doctoral graduate. From someone who has studied it for a lifetime, you’ll get either a complete mish-mash or a noncommittal shrug. The theoretical landscape of intelligence has a rich and interesting history, but our current models are shifting toward broader, more neurologically based ideas as constructs are mapped to physical phenomena (e.g., brain activity). I call this the inward approach to understanding human behavior.

Muse and the demand for neocloud?

Fri, 23 Oct 2026 00:00:00 +0000

The new opiate of the masses - Jarvis with an Oracle for a brain.

My take on the current state of AI, etc.: Meta’s new Muse agent(s) is setting records and has Meta scrabbling to forecast their need for compute - aka neocloud (GPU-as-a-service providers). This is just the beginning of the era of the agents. As they become more and more reliable, people will begin to trust them with more and more of their personal data and daily life (i.e., real-time, real-world data). In business school they called this “boiling the frog.” The adage goes, don’t put a frog in boiling water, he’ll jump right out. Instead, give him a warm bath and turn the fire on low… The point is, we’re are in the warm-bath stage with AI and the fire is just now starting to heat up. What happens when compute availability becomes scarce. What is the expected level of scarcity? – Better yet, who controls the compute when compute becomes scarce? – Answer: A handful of companies competing to stay ahead of OpenSource competition, that’s who. That aside, we are at the event horizon of an exponential arc due to agent proliferation (singularity?). The demand curve will create incredible g-force on the world’s economies over the coming years - 2027 we’ll see the first real acceleration. The build out of compute cannot keep up with the demand for at least 3 years out (that’s an optimistic estimate relying on a technology boost/change within the next 18 months - e.g., better/cheaper robotics, new technologies). Sadly, it will occur in the midst of world turmoil both politically and economically. While literally anything could happen, I think it’s relatively safe to say the 3 year forecast is for a chance of chaos with moderate to heavy crazy.

Building sentiment

Wed, 21 Oct 2026 00:00:00 +0000

Once I have a general sense that the podcast and article summaries are relatively stable, I plan to build a sentiment extractor using the data collected over time to try to understand changes. Stay tuned, this should show up on the homepage within a week or two.

How this started

Wed, 21 Oct 2026 00:00:00 +0000

This whole AI journey started last spring when I was planning to retire and knew that I wanted to play around with AI. And at that time what was really hot was a new AI harness called OpenClaw.

I took OpenClaw for a test drive using a cloud server and I played around with it a few days. At that time people were purchasing the Apple Mini to run OpenClaw. However, by the time I looked into it they were really hard to find. Instead, I asked AI for an alternative and ended up finding my Asus GX10 system currently using a Nvidia version of Qwen 3.6 (and Sonnet 5.5 when needed).