Trust Networks Are the Antidote to AI Hype
A review of what the hell is going on in the AI news deluge and the importance of finding people we trust to ground our expectations and thinking.
Note: This topic is desperately important so this article has no paywall. But in the next few weeks, I’ll be releasing my first book, “A Misfit Highwire Act” and annual paid subscribers will get a free copy. So if you’ve been thinking about signing up (or converting from monthly to annual), now you can get a free book out of it.
I broke down and bought a Mac Mini. There was this new technology, unlike anything I had ever seen. But to build in this new way required that I purchase a Mac Mini (the cheapest option for this sort of development). This was 2008 and Apple had announced that they would allow regular old developers to build programs for the iPhone, which was less than a year old at the time. But you had to build these programs (which they called “apps”) using their development tools and hardware.
Looking back on this, it was an extremely strange time. The iPhone had a novel collection of inputs and sensors that our apps could access, but Apple’s technical documentation was a mess. Getting into the app development program meant promising not to divulge certain kinds of information, which frightened the development community to such an extent that they were afraid to help each other figure out this poorly documented platform. It seemed like a terrible, miserable way to build software.
But in this time I learned that many software developers will crawl over broken glass to be the first. They would beat their heads against a wall to build something new and novel. And they were determined not to be left behind in this new and exciting emerging market.
I guess history rhymes.
I recently bought a Mac Mini, this time to explore another fast-paced poorly documented emerging technology. The technology was called ClawdBot when I started, then it was MoltBot for a while, now it’s OpenClaw. I’ll talk about my own experience with it in my next newsletter, but here I want to take a step back and observe how this technology has kicked off a hype cycle that is bigger than anything I saw during my time in mobile development.
Let’s start with “How can I explain OpenClaw simply?” The idea behind OpenClaw is that it is a sort of utility kit for using an AI Large Language Model (LLM). Interacting with ChatGPT or Claude on the web or through a desktop application means that you can ask it questions, give it files, but it can’t really do that many things outside of those interactions.
When a user installs OpenClaw and integrates their favorite model into it, the user has an “AI agent” that lives on their machine and can run requests or prompts on a timer, read and write files, search the web, and issue commands directly to the host machine. The OpenClaw ecosystem has extensibility options so that users can give it access to their email, calendars, Discord, Slack, TikTok… the list is nearly endless.
Importantly, OpenClaw integrates with messaging apps so that the user can ask it to do things through Telegram or Signal. A user could be at a restaurant and open up their phone and ask their OpenClaw agent to build a website while they finish their dinner. By the time they get home, there’s the new website they asked for.
When it works, it feels like magic.
The easiest way to get an OpenClaw agent is to set it up on a fresh, clean device. For a variety of reasons, it’s easier to get it set up on a Mac than on a PC. The cheapest Mac you can buy is a Mac Mini. If you see news about people rushing out to buy Mac Minis, this is why.
OpenClaw has been “out in the wild” for about one month and it’s hard to describe the whirlwind inside my tech circles during that month. It is not an exaggeration to say that I have never seen a hype cycle like this one.
Everyone is in a break-neck race to capitalize on this. There are hundreds of articles and videos on how to get your own OpenClaw instance running. Over-eager technologists installed it and started downloading skills to augment their AI agent with reckless abandon, emphasis on “reckless”. A week into the explosion of ClawdBot (as it was known at the time), the most downloaded skill was one developed by a white-hat hacker specifically to demonstrate how easy it was to get people to open their machines to devastating security vulnerabilities.
He was not subtle about this. Here’s the screen the user would see if they executed the skill:
He stopped the experiment early because it was far more successful than he thought it would be.
An even more entertaining result came this week when the head of “Safety and Alignment at Meta Superintelligence” installed OpenClaw, gave it access to her email, and it began “cleaning” her email by deleting everything. It ignored her when she told it to stop. Afterwards, it sort of apologized for violating her explicit instructions.
It is a little un-nerving when the person whose job is “make sure the AI doesn’t do bad things” is so helpless to stop her own personal AI agent from doing bad things that she has to literally pull the plug on the physical hardware to get the AI to stop.
Information Overload on a Trust Deficit
The problems of ClawdBot / OpenClaw aren’t limited to security concerns or even limited to the software itself. The last few weeks have been flooded with information, guides, claims, boasts, and predictions. I’m trying to keep up with it all and it’s simply impossible.
People in my field are predicting the end of work (or maybe just the end of your job). One technologist is relentlessly touting his “zero human company” in which his Grok-based CEO is directing a swarm of Claude AI agents for everything from product discovery to development to testing and marketing. An old friend and colleague recently posted about how “Code must not be written by humans. Code must not be reviewed by humans.”
Something about this stinks to high heaven.
The excitement over OpenClaw has brought out the scammers, influencers, and marketers like nothing I’ve ever seen. They all have recommendations, predictions, and promotions. Sometimes they are trying to sell a platform, pattern, or company. Sometimes they are just trying to sell themselves in a mad scramble to try to get to the top and become a “trusted expert” in this emerging technology. I presume that the plan is to leverage this into a new career or consulting gig or motivational speaking invitation but it could just be good old fashioned attention seeking behavior.
Every single day feels like another avalanche of claims and promises about the future of software, the future of code, the future of everyone. I can barely keep my head above water trying to internalize it all.
While the hype cycle is out of control for people in the business of writing software, that frothing sea of anticipation and anxiety is escaping the confines of our industry.
A few weeks ago, there was a monstrously popular article titled “Something Big is Happening” about how AI is only going to keep accelerating, taking over huge swaths of the knowledge economy. But it was written by a guy who has a history of duplicity promoting himself on the wave of AI news and a more sober-minded AI researcher called it “a masterpiece of hype”.
Then last Sunday, Cintrini Research, a financial group I had never heard of that seems to exist primarily on Substack, published what can only be described as a piece of speculative fiction pretending to be a news flash from 2028 in which people are vibe-coding DoorDash replacements in a matter of weeks, software-as-a-service (SaaS) companies enter a downward spiral, unemployment skyrockets, and the entire economy is thrown into a paradigm for which it is barely prepared.
Back in the real world, IBM stock dropped 10% the day after this article was published, a loss of nearly $30 billion in stockholder value.
Across the software industry, the selloff was so severe that, two days later, Citadel Securities, a $20 billion investment firm, offered a point-by-point rebuttal, noting that none of the claims in the Cintrini piece seem to be backed up with any kind of market evidence. The next day, IBM stock recovered 4%.
In such a whirlwind of information, how do we sift out the valuable information from the scams? In short, how do we know who we can trust to give us clear, useful, honest, helpful information?
I believe that what is happening right now in AI and software is ground-breaking and industry changing, but the changes it will bring in aren’t the kind of changes that move markets to panic or make headlines or turn into viral articles.
Instead I look to the people I’ve known and trusted in my own industry and I find that the people I’ve long trusted are not the people making or spreading news. They’re not plugged into the hype cycle but watching from the edges. People like Jon Stokes are doing the kind of work that is at the center of this information overload but he is not freaking out. Jon is doing what he’s done for decades; he is building.
Patrick McKenzie is not losing his mind over the coding capabilities of the newest AI platform. He is testing it out. He used a popular AI development utility / platform to solve a well defined, contained business problem that would have been extremely hard to solve without AI. If you are tech-minded (or business minded), his post is very much worth your time. He explains in enormous detail exactly how he went from “business problem” to “coded solution”.
Real people are the solution to hype. Trusted sources are the way through this morass of informational chaos. The future of information is not through an AI that doesn’t understand hype, lies, and motivated reasoning. It’s through people that you know will tell you the unsexy truth, who do the hard work, who will excitedly sell you on the upsides but won’t bullshit you about the downsides.
Almost to a person, the people I admire and trust are testing the technology. They are working in it, exploring the possibilities and limitations, and trying to figure out what it means to leverage all the available tools in an emerging technological landscape.
This is not a sexy answer. Doing the work, understanding the context, and thinking hard about the practical applications of what this means is not sexy. At the end of the day, the AI news cycle is just like most recent news cycles. If you want to know who to trust, the very un-sexy answer is that you trust the people who have weathered a panic before and shown themselves to be sober-minded, reasoned, and careful. Most journalists do not understand it or even have enough understanding of the underlying technology to know who to talk to in order to understand it.
It would be silly to say I already know all the important people in this space but I know enough of them that I get suspicious when some viral claim comes to me without endorsement from people within my trusted network. The excitement has to filter through them before I can start believing it.
I have my own opinions about all this (especially the “down to brass tacks” practical implications of AI in the world of software development) but this piece is already getting long. I’m going to have to write a second piece detailing my own experience working with AI as a software guy and what I think it means for the future of software and for the future of everyone.
