The Next Six Months

Six Months

Today, I noticed a post by Logan Kilpatrick:

The next 6 months of AI are likely to be the most wild we will have seen so far

This got me thinking about where we are with artificial intelligence.

1. Many newfound AI enthusiasts and AI gurus appear to have grown bored. Producing value with computers still takes work. AI still is not magic.
2. In AI, so much happens so quickly. It can seem exhausting to keep up.
3. The gap has widened between what AI can (currently) do for ML researchers and experienced coders on the one hand---and common computer users on the other.
4. The ecosystem around agents evolves quicker than your average SARS genome. MCP (and A2A) are incredibly impactful.

AI is not Magic

People with an actual background in Machine Learning (pre 2022) are not usually overenthusiastic about new models coming out, but appropriately enthusiastic. Not every release needs to be the game changer. Progress can just be good enough.

But many newcomers to AI have had a different experience and are now disappointed. What, Large Language Models cannot do calculus (well)? Tokenization stands in the way of counting letters? Video generation still does not produce feature-length films?

Great papers are published all the time. The tooling keeps improving. Hardware is insane (and I understand none of the progress). And we keep gaining experience with these models. To me, Machine Learning feels incredible at the moment, but I see that the pace of progress has created an interesting dynamic for anyone trying to keep up.

Constant Progress

Since AI hit the mainstream, being able to quickly absorb and integrate a lot of information has felt like a huge unfair advantage. Ironic, given that (according to some) LLMs should do all the reading now.

I remember the first review I wrote as a grad student. It took the better part of a week. Towards the end of my PhD, I could do several before lunch. (On average, of course. Long papers with a lot of substance and without obvious flaws would take much, much longer---unless they were beyond me and I excused myself due to my lack of “expertise”.)

I pity any poor soul getting all excited about AI in 2022 or 2023, discovering in 2025 that now they need to read, learn, practice, and constantly change their mental model.

So, who is really enjoying all that artificial intelligence at the moment?

Machine Learning is still for Machine People

Current tools still require significant technical expertise to use effectively.

What we can do with what is currently available is amazing. As long as we have been coding for years and want to work with a terminal. See Armin Ronacher, e.g., his recommendations for agentic coding. He often shares what he does and how he works, and it is incredible to witness the rate of change.

I can currently work with computers the way I had dreamed about 30 years ago. And I am loving it. I built my own MCP IMAP Server, hooked it up to an agent, and am finally achieving “zero inbox”---the kind of assisted workflow I’ve wanted for decades but could never quite get.

But forging this raw power into tools that small and medium enterprises would and could use? Sure, there is a ton of value in that space, but it’s not where I see the excitement at the moment.

Part of what’s needed is better infrastructure for this translation to happen. And it is coming.

An Evolving Ecosystem

Just a few months ago, I wrote about how networks might (and should) move away from visual-first implementations to API-first approaches (with interfaces for machines and humans).

Now we have MCP being adopted rapidly. Sure, it’s not perfect, and it might not stick around, but it is a protocol and it helps developers and users form an ecosystem that can grow. It’s exciting and not without danger. I am just waiting for story after story to come up of agents with too much free reign properly screwing things up.

We’re playing with fire here, but it burns oh-so-bright. The next six months are going to be wild---the question is, who is going to be in control.