It’s been a crazy few weeks in AI land. There is so much to digest and talk about that it’s hard to know where to start. Also, I’ve been busy in the lab with my new project (code name: Handrail AI) and I am about to sign up my first implementation partner.
Since the last article…
Anthropic released a first look at their new Mythos model, which they claim has found security vulnerabilities across every major operating system. Depending on who you believe, this is either the end of the world as we know it or a form of guerrilla marketing, but either way, I think we are about to reach an inflection point.
Of course, in what has become a tit-for-tat, OpenAI could not be outdone and released their own cybersecurity model. There is a funny video online where a guy explores the 12 different types of Dawn Ultra Plus ++ soap. This is what AI feels like today.
Both companies are raising a gazillion dollars. Cybersecurity and SaaS stocks are in free fall. The cost and supply of energy (oil) worldwide changes by the minute.
Speaking of inflection points, Elon recently said we are in the singularity already. Keep in mind that Elon says a lot of bold statements. Most of the time, he is directionally correct, but his timing is sometimes a bit off. One of the few things I share with a guy like Musk is that we both believe that the universe is always listening, and if you speak up enough, it often listens. He has had more success with this than I have to date. His singularity predictions could be him trying to will it into existence.
There is so much to talk about that I wish I could write it all now. However, here is what is coming and some early thoughts on what the above things mean (tangibly).
No matter what you think about Mythos, the reality of it is that you need to understand what this implies. A lot of apps are security by obscurity alone, and until now, the ROI of a bad actor hacking your little company was not there. When the cost of trying to exploit drops near zero, everything changes. If you have legacy tech and a large surface area (lots of apps), this is what they will exploit first. I’ll dive deeper into what this means for mid-size companies.
Using AI to create software has had its first watershed moment. I’ll go into the specifics in a future article, but the future is becoming clearer on what is going to work and what will not. If you are a business owner with dreams of tech innovation or process improvement, today is the cheapest and most cost-effective it will ever be.
AI model quality is getting worse, expectations are rising, and tensions between China and the US over oil are changing. Model providers (OpenAI, Anthropic) are going to get squeezed, and you are going to pay the price. Right now, investors are giving you money (subsidizing) to grow your company; that will not last much longer.
The biggest mistake companies can make right now is assuming this is all still theoretical and waiting for some big moment. Right now, playing catch-up is still very affordable, but that gap is widening quickly.
This newsletter is about action; here are some practical “what to do now” items.
Find your soft spots.
If your business runs on older apps, patched-together workflows, shared inboxes, spreadsheets, or tribal knowledge, assume those are the first places pressure will show up. Know what systems you have, who touches them, and where the weak links are.Stop thinking AI is just a chat.
If your entire AI strategy is “we bought a seat license for a few people,” that is not a strategy. The real opportunity is not chat. It is throughput. Look for work that gets stuck in queues, inboxes, approvals, hand-offs, and repetitive decision trees.Move while it is still cheap.
If you have been thinking about building an internal tool, workflow assistant, or automation layer, now is the time. This will not get cheaper from here.Assume obscurity is no longer protection.
A lot of smaller companies have been too small to bother with. That math is changing. Clean up old systems, reduce exposure, and tighten security now.Build for the real world.
Pretty demos are easy. Reliable systems are hard. Focus on tools that are useful, repeatable, and survive contact with an actual business.Do not get used to subsidized AI.
Today’s pricing is not permanent. Build with the assumption that model and token costs will rise over time.
The window to play catch-up is still open, but it is not going to stay open for long.