AI is making my head spin

It’s been over three years since I posted on my site about some interesting but nascent new technology called generative AI. It was blowing our collective minds by writing a few paragraphs on our behalf, and generating images for us.

Things have changed since then. Simple image generation has graduated to video. AI can handle writing longer documents. It summarizes our emails, takes notes for us, is connected real time to the web, and helps us write code. Still, for a while there, everyone in my professional world was talking about AI as if it was life changing stuff, and fundamentally transforming our work. But has that really been the average persons experience? Many of us use it regularly, but “transforming” our work feels like a stretch.

In my own role, I struggled to get it to do much for me other than summarizing documents, generating images, taking notes, performing market research, etc. All useful to a degree, but not transformational. The constant drum beat of “it’s world changing” and “you need to use this every day” and “you will be left behind If you don’t” was making me feel a bit like Mr. Affleck here.

But things are changing again. It feels like we are in another one of those “moments”, like we were a few years ago when ChatGPT showed up and sparked our collective imaginations.

These “moments” raise plenty of questions, and since this is new and unprecedented technology, its hard to see around the corners of its implications. Hence the head spinning reference in this posts title.

Stopping to take the time to write helps me think clearly. So this post is more for my benefit than any reader. Still, I hope the below is helpful, and thought provoking for anyone who does stumble upon this. There are three angles I want to tackle in this post.

1. The new “moment”. Generative AI → agentic execution.
2. The meaning and benefits of work.
3. Unbounded technological progression.

Generative AI → agentic execution

The major AI companies are all racing to release and gain market share on some powerful new technology. We’re moving past the state of simple generative AI, into what I’m calling agentic execution. As in, AI is executing tasks on our behalf.

These products are generally not in mass market yet. Most are in beta, only available on MacOS, and often require a subscription to use.

I’m fortunate to work in both an industry, and a company that strongly encourages us to be on the cutting edge of this tech, so I have been able to play around with these products. I’m particularly using something called Claude Code, and Claude Cowork (made by Anthropic), which are the products that much of the chatter and excitement is aimed around today. Though competitors like OpenAI are releasing their own iterations.

So what is it about these products that is causing my head to spin? First, I have to give you context on the technology itself.

The new technology:

Claude Code is a tool that was originally targeted to developers. Developers could run Claude locally on their computers terminal, and use natural language to perform coding tasks. Recently, Claude has moved Claude Code into a desktop app. So now you don’t need to open your computers terminal to chat with Claude, you can just do it on their app. Recently, Claude released Claude Cowork. It’s effectively the same product, but geared towards non-developers. So when you use your desktop app now, you can flip between Claude the chatbot, Claude Cowork, or Claude Code.

The big change between these products (code and cowork), and the traditional AI chatbot we are used to is Claude Cowork can work with your files and actually perform tasks. You can give it access to folders on your computer, and it can organize files, create new files, or edit existing files. Beyond working locally on your computer, you can also run Claude in your browsers and navigate the web, click links, and even have it buy stuff for you.

I’ve been using it for a couple weeks now, and my entire worldview with respect to the sort of tasks and work I can get done has changed. I’d struggled to get major value out of AI chat bots with respect to my job. I’d dump them large datasets, and ask them to create spreadsheets and files capable of cleaning, analyzing, and reporting on that data, like I’ve done myself faithfully for years. But the chatbots really struggle. This is because they have relatively small context windows. They can’t store all the data in their limited memory and keep it there while they perform tasks. 

So the approach to working with Claude Cowork is a little different. Successful outputs mean spending a bit more time with your prompts (how you ask AI to do something for you). For large complex tasks, you often need to use an “agentic” approach to prompting. This is where, instead of simply asking it to do something for you, you assign “agents” to work on chunks of what you need at a time. You still provide Claude with all the context of what you want, but you divide the work up to various “expert agents” who focus on specific chunks of your goal. This makes the context windows much smaller for AI, so it truly can hold what it needs in its memory and perform tasks. You can create permanent agents that you can call on again in various tasks. You can define skills for those agents. It also has its own out of the box skills such as “excel” that it can call upon if you are asking it to work in a spreadsheet. You can organize your work into projects. Those projects might have defined agents and skills. This means every time you jump back into a given category or project, you have a very tailored team of agents ready to help. And by the way, creating agents is done just like talking to an AI chatbot. You simply chat to AI, and ask to create an agent. There is nothing technical or complicated about it.

I’ve also found it helpful to ask Claude to ask me clarifying questions before it gets started. This way, it can seek to clarify anything you may not have thought of in the original prompt, and tighten up the work it is about to do. Cowork has a really nice interface for this. It often pops up a nice multiple choice box for you to respond to when it asks something like “what is your top priority, A, B, C or D” and it usually gives a free form field in case none of those options work for you.

This is what the interface looks like.

As an example of a prompt I have used recently that uses agents, see below:

---

Rather than struggling like traditional chatbots did, this technology, along with specifying agents, was able to create an excel spreadsheet in a file on my computer that re-created a spreadsheet I had made a couple years ago. It took me 20+ hours to create the comprehensive spreadsheet myself, and Cowork had it done in 5-10 minutes.

It’s quite amazing to watch. As you prompt It, Cowork creates a to-do list, that as each agent is activated, it checks off its respective item of the to-do list. When one agent checks off its item, the next agent is activated. Claude shows the file format of the outputs it will create, and it shows you the “skills” it is leveraging to do the work. For example, it has an out of the box skill of knowing how to use Excel. But you can also define or teach it skills. There is even products out there today where AI can be pointed to a webpage, and it can learn skills itself to be used at a later date.

I watched real-time as AI downloaded python files so it could perform calculations. Before it makes changes on your computer, it always stops and asks for your permission.

I also asked it to create a long document for me, using a similar agentic style prompt. The doc ended up being 30 pages long (way longer than a traditional chatbot can typically produce), full of charts, tables, visuals, and commentary. At one point, I was bumping up against my token limits (your plans are capped at certain limits that you get back over time as it costs Claude money to perform all these tasks) and rather than simply stopping, Claude paused, broke the report into four chunks and in four separate documents, and as it was doing that, I was thinking to myself “oh, I’m going to have to combine all these, not a big deal I guess, but kind of annoying”, but seconds later, Claude took all four of the separate docs it created to save me token space, and combined them into one. Amazing. Then I opened the folder in my desktop, and there was a completely AI-generated 30-page word doc on my computer for me to review. 

You can run multiple jobs or projects at once. I’ve seen people discuss letting Claude run overnight while they sleep, testing their code, fixing issues, and preparing a summary report on what changed for you when you wake up.

You can also connect Claude to a whole suite of applications through something called “MCP” (model context protocol) which is an open-source standard created by Anthropic that allows AI to interact with third party applications. So I can connect Claude to my Google calendar, GitHub, Gmail, Spotify, Microsoft, Tableau, Stripe, and a host of other popular applications, and have it take actions for me like manage my calendar, control Spotify, pay bills, etc.

The pace at which this technology is moving is nearly impossible to keep up with. To give you an idea, I started drafting this post on Feb 4, 2026, and on Feb 5th, I saw this.

And, on that very same day, they released an improved model, that it so happens is better at advanced finance tasks.

Claude Opus 4.6 marks a step forward in AI for finance. It can be used to help professionals make decisions based on accurate information and clear analysis, and it produces deliverables with real polish. The model is substantially better than others in the market at financial reasoning, multitasking, and maintaining focus over longer multi-step tasks. 

Consider that this technology is not yet primarily designed with me in mind (Not a developer). The things that developers are outputting with Claude Code/Cowork far surpass the successes I have had.

I was recently shared on an email update on all things tech that JP Morgan sends out which said:

Another though tape for software with SaaS baskets down ~15% YTD (worst start since ’22) and names like Intuit, Adobe, Hubspot, Salesforce and Sigma down 11-31% this week alone. Several reasons were cited for the carnage (and neither are really new) but top contender was the recent launch of Anthropic’s AI agent, Cowork. The tool reiterated the power of AI and just how easily software can be displaced; thus threatening the moat of legacy software and seat based SAAS revenue models (several investors have noted that Claude has essentially recreated NetSuite in a matter of weeks). This is the crux of the terminal value comment referenced in my CIO conversations above. Cowork also bolstered excitement around Claude Code which is a truly impressive AI coding tool (albeit, launched back in October) that allows everyday users to develop robust applications & software using natural language prompts.

For context, NetSuite is a company that has existed since 1998. It is enterprise accounting software used by almost all major tech companies. We use it at my company currently, and I used it at my last as well. They generate over $1B in revenue/quarter. Imagine NetSuite’s terror as Claude recreates their software they have been working on for 25 years in a matter of weeks.

In an interview with TBPN (a high octane Business news podcast) Sequoia Capital partners talked about the end of “product led growth” as a go-to-market strategy, saying:

You see this most clearly if you’re using Claude Code actively. It says, ‘Hey, for a database, you should use Supabase. For hosting, use Vercel.’ It’s choosing for you, the stuff you should be using. Product-led growth brought us closer to the vision of ‘best product wins,’ but ultimately people are still lazy. They can’t read all the reviews, and they kind of default to what looks cool on the website. Whereas your agent has infinite time to go and make these choices for you. It can go and read all the documentation, read all the user comments, and figure out [what you need] for your use case.”

So this is our current reality. Software companies moats are being destroyed. Products that took years, and billions of dollars to develop are being recreated in weeks. Why would any new software startup prioritize designing software with humans as the front end user? Why spend all the overhead on design, and marketing, and creating useable interfaces and flows for dumb humans to work through and understand? 

More and more people are using AI, and more and more are going to be using agents to accomplish tasks on their behalf. AI isn’t swayed by flashy marketing, or algorithmic feeds or designs meant to capture our attention. They’ll work on the backend. Reading documentation, reading reviews, and spending their infinite time finding the right solution for us. They are completely platform agnostic, and only care about accomplishing your goals for you.

This may also cause the cost of software to come down since we could imagine a world where most tech companies don’t need front-ends designed for humans, landing pages, marketing programs, a design organization, etc. They just need a good product, and good documentation so AI can evaluate and learn it.

It’s good for the consumer I think. The best product wins. Not the best marketers. And the ability for products to lock you into walled gardens will be reduced, as there should theoretically be many alternatives for accomplishing the same task.

The head scratchers:

So let’s talk about all the things that are making my head spin.

How does a company trying to solve a problem with software even proceed in this new agentic environment? Most consumers are not using these advanced tools yet, so there is a space in the market where we still need human centric software. But how long will that last?

Traditional SaaS (software as a service) companies have thrived on defensible barriers that made it hard for competitors to encroach. Salesforce’s ecosystem lock-in, or Adobe’s creative tool integrations for example. These moats were built on years of R&D, data accumulation, and customer entrenchment. But AI is turning software creation into a commodity.

If AI can democratize software creation, why would businesses or investors pour millions of dollars into proprietary codebases?

As I mentioned earlier, this isn’t just theoretical. The markets are already reacting and software stocks have tanked amidst these disruptive new technologies coming out.

Ok, so AI is weakening traditional software companies. That’s just what happens when new technology emerges. But this technology is still new, and not largely adopted yet in the general public. What are companies supposed to do right now? Ignore it and hope it goes away? Keep on doing what they are doing for now (and risk falling behind), or make big bets on this emerging technology? But:

• What if your AI bet flops? Which agent platforms will dominate? OpenAI’s GPT series, Anthropic’s Claude, Google’s Gemini, X’s Grok, Perplexity, or something else? Which protocols and standards for accessing data, executing commerce, etc will stick? AI firms burned billions in 2025 alone on infrastructure and talent. Consolidation is inevitable, and some of these companies will collapse like past dot-com darlings. So if a business builds product for a failing ecosystem, the product becomes obsolete overnight.
• Human-AI hybrid limbo: Do companies optimize for today’s human users (risking irrelevance) or tomorrow’s agents (risking premature optimization)? What about ethical pitfalls, like AI hallucinations in critical software (e.g., finance or healthcare)?
• Talent crunch: Software teams must now include AI specialists, but the supply is limited, and top talent is unbelievably expensive (Meta Just Paid $250M To Lure This 24-year-old AI Whiz Kid). How do you upskill without derailing current roadmaps?

Even bigger questions emerge. What happens to legacy investments and human capital?

SaaS companies today represent trillions in market cap. Salesforce alone is worth hundreds of billions, with the broader sector employing millions. If AI unbundles these empires, what happens?

AI could render much of the software investment moot, leading to a “great unbundling” where specialized AI tools chip away at all-in-one suites. Or, perhaps somewhat counterintuitively, a “great consolidation” where companies with deep pockets (e.g., Microsoft) absorb AI startups and fortify their positions.

Beyond that, millions of developers, salespeople, and support staff are tied to these “legacy” ecosystems. AI could automate 30-40% of these jobs, forcing a reskilling wave. But where do they go? Into AI companies that might not survive?

I’m also wondering about knowledge workers young in their career trying to enter the tech industry. Junior accountants, analysts, content marketers, developers, designers, recruiters, HR, researchers. They’ve likely just come out of college, spending years picking up the baseline skills for these careers, and now more and more companies aren’t touching them, because AI is capable of essentially everything a junior level human would do. Maybe they get a job with a company that is not staying up on trends in this area, which allows them time to build some skills, but how long until the very company they work for is leapfrogged by an AI first competitor? I imagine the key for these young workers is to become as fluent as possible with some of these tools, and use them to their advantage. Especially while many (most?) companies have not yet adopted them.

How are traditional websites that rely on human traffic and ad revenue going to work in this new future? Many companies are already seeing alarming declines in traffic across websites, as many of us get our information from AI summaries now, not from web pages themselves. What does the future of the internet look like when the typical website becomes optimized for AI as the primary readers, not humans?

I could keep going, but I am getting dizzy just thinking about all this, so I will stop there. But like I said, I want to approach this dizziness from three angles. We just finished the first. So maybe take a quick break to stretch and grab a snack, get your head back on straight, then lets go to the next section.

Work is for the worker

I stole this headline from an article I read recently, that came right in the middle of my head-spinning, AI induced mania. I found it to be a thought provoking read, and matched a number of thoughts I had on my mind as well. Here is the article if interested: Work is for the worker.

The basic assertion of the author is that work serves two purposes:

1. An objective purpose. To provide value to society.
2. A subjective purpose. To shape and fulfill the worker, and build character through effort, failings, learning, accomplishment, and humility.

He uses the example of monkey bars to explain the subjective value of work.

This “superior ethical ­criterion,” the subjective end of work, is immediately evident to parents. When your daughter is dangling from the monkey bars, if your only concern were the objective end of the work—namely, getting her body from one end of the apparatus to the other—you would just carry her to the other end. But what a stupid idea! We all know that getting across the monkey bars is worthwhile precisely because of the time and difficulty and ­failure—the inefficiencies, if you will—­involved in accomplishing it. As it turns out, time and difficulty and failure are the only way to achieve the subjective end of work—which is also called ­character.

The parallels to AI are evident. In a world where people are increasingly asking robots to do things for them, what happens to the subjective benefits we gain from work?

The point of the article isn’t to decry AI. But it’s to encourage us to keep in mind that we need to use AI responsibility, and our goal in life should not be to outsource work completely.

The article also mentions studies coming out that show concern for humans ability to think critically as our reliance on this technology increases. An MIT study came out mid-last year, and I remember it causing quite a stir. One of the takeaways from the study:

83% of AI Users Can’t Remember Their Own Writing

Other articles such as this one came out in similar time frame and showed that in their research, developers were actually becoming less productive when using these tools.

I’m sure that in reality, we’ll have a mix of both. Some heavy AI users will unlock really cool “superpowers” as they work with this new technology, and they will work with it in such a way as to not turn their brain to sludge. But we know many others won’t have that same fate.

Unbounded technological progression

Would you say that our societies (thinking North America/Western Europe) have an agreed upon morale framework that acts as an “invisible hand” to guide how we approach (or not approach) some of our more revolutionary technological breakthroughs and the moral/ethical questions that arise out of them? I wouldn’t.

The cultural imagination towards morality is best described as “what is right for you, might not be right for me, and there is no universal standard”. And if someone were to suggest there is a universal standard, they would be brushed off as closed minded, and not very inclusive.

What is the implication of this deeply embedded way of thinking?

While we can use AI as an example, it might be helpful to take a more extreme case to illustrate the point I hope to make.

Since we’ve already talked about MIT, let’s reference another article from them, an opinion piece published last year in their MIT technology review.

These “bodyoids” would be genetically engineered, brainless human bodies grown from stem cells, to be used for medical research, drug testing, and organ harvesting. They would somehow be “ethically sourced” (not quite sure what that means), and the article says they would lack consciousness, pain, or moral personhood. So therefore, they could revolutionize medicine by providing humans with an unlimited amount of “human-like material” to draw from for their own needs.

Obviously, this article has faced plenty of criticism and outrage. One critique I read was very severe in its response to this:

However, the authors go on to dismiss this concern. They argue that, like bodyoids, a sufficiently detailed mannequin would look much like us; that does not make it human. But nobody is proposing scientific experiments on mannequins, and for good reason. However realistic they might appear, they are not human, and thus, unlike a bodyoid, they have no value for ­science and medicine. A bodyoid’s value for science and medicine lies precisely in what it would be, which is not a zombie, not a dead person, not a mannequin that mimics the human form. It would be a profoundly disabled human being, designed and created to be profoundly ­disabled—a ­vulnerable human being so ­totally defenceless and voiceless that it could be exploited with ­impunity. If this is the case, we would endorse this macabre project only if we ourselves had become, so to speak, moral zombies.

Let’s look at one more example.

There is this guy named Bryan Johnson. He has such goofy beliefs, that he is almost a caricature, but he is serious about them, which I suppose makes him simply a peak “tech-bro”.

Bryan Johnson wants to live forever. And he believes we might be the first generation to actually accomplish this. Bryan spends all of his time trying to optimize his body for perfect health and longevity. And he owns companies that sell the products/vitamins/supplements/sensors/fitness programs so we can do the same.

Here is a quote from Bryan from an interview with him I saw on Youtube.

My body’s aging extremely slowly. I have 50 biomarkers, the things you would normally expect like cholesterol and triglycerides and things like that that are in the perfect optimal clinical outcome range. I have 100 biomarkers that are less than my chronological age, several fitness tests where I test out as an elite 18 year old, my body runs 3 degrees Fahrenheit, cooler than normal. So no matter how you’re looking at it, whether it’s my DNA methylation, my fitness test, my biomarkers, my phenotypic markers, whatever you’re looking at, the data says the same thing. I’m in near perfect health.

And he’s humble too!

That quote is also just asking for the guy to accidentally get hit by a bus or something, and tragically end his desire to live forever.

In another interview with Bryan, he said the following.

I’ve been testing the hypothesis that if I get a whole bunch of data about my body, and I give it to an algorithm, and feed that algorithm updates with scientific evidence, then it would eventually do a better job than a doctor. So I gave myself over to an algorithm. It really is in my best interest to let it tell me what to eat, tell me when to sleep and exercise, because it would do a better job of making me happy. Instead of my mind haphazardly deciding what it wants to eat based on how it feels in the moment, the body is elevated to a position of authority. AI is going to be omnipresent and built into our everyday activities. Just like it autocompletes our texts, it will be able to autocomplete our thoughts.

This is from an article where Bryan is talking about wanting to start a new religion, where the body is God. He wants it to be a religion only because he sees religious institutions as helpful mediums for accomplishing something. But his religion elevates the body to God in that, we give our mind over to AI and algorithms, so that we can optimize everything towards longevity of our body. In other words, our highest ideal is simply existing.

This technology is obviously not fully here, but its not difficult to imagine a not too distant future where we have AI capable of telling us what to do, and overriding our emotional impulses to do things that don’t optimize for longevity (like eating a cookie). We already have brain-computer interfaces like Elon’s Neuralink. And I already wear a Garmin watch that constantly monitors and tracks a fairly large number of health metrics.

Are people who are optimizing their lives towards maximum longevity really living? I imagine an extremely useful way to exist for a long time is to go into one of those cryogenic chambers we read about in our sci-fi novels, and exist in a frozen coma for the rest of your life. That is certainly more tightly optimized for longevity than having a conversation with a friend. That so called “friend” might accidentally cough on you, and set your biomarkers back eight years. Or they might say something that slightly rubs you the wrong way, which elevates your heart unnecessarily. No, no. Better to freeze yourself.

So if Bryan Johnson, an influential and wealthy Silicon Valley tech bro has his way, he’ll be able to help people be much happier and more human, by providing them the technology to live exist as long as possible.

So, to re-ask my question, what is the implication of not having a largely agreed upon moral framework embedded in our culture to guide these rapidly progressing and unprecedented ethical dilemmas that are springing up?

It’s unbounded technological progression. Technology advances simply because it can. Human flourishing is not a prerequisite. Or, perhaps it claims to be on the surface, but as we can see, people like Bryan, or the MIT authors have some pretty unique views on technology that is good for humans. You have to wonder if some technological advancement is really aimed at improving humans lives, or if that’s just what the developers say, so they can win our trust, and our business. It’s kind of like when you read the original mission statement of Facebook, which was:

to give people the power to build community and bring the world closer together

All you can really do is laugh when you read that. That may be what Facebook says their mission is, but we all know their actual mission is to capture as much of your attention as possible, for as long as possible, so they can run as many eyeballs over ads as possible so they can increase their wealth and influence. And the impact Facebook/Instagram has had on our brains, attention spans, mental health, and relationships certainly doesn’t seem commensurate with that mission statement.

All of this reminds me of a book by C.S. Lewis called The Abolition of Man. Lewis argues in the book that modern society’s rejection of objective moral values, leaves humanity adrift in subjectivism and scientism. And in this void, science and technology aren’t tools for human flourishing but ends in themselves, which enable us to “conquer nature”.  And there is nothing, especially not a shared objective morality, to guide it. Technology advances simply because it can, not because it’s for the good of humanity.  He points out that the final stage of this conquest of nature isn’t liberation, but subjugation. As in, future generations become influenced and reshaped by a elite class (like Bryan Johnson) who wield power without any transcendent ethical framework to guide them. So now draw the dotted line to the tech bro ethos, where biohacking and longevity experiments don’t see humans as sacred, but as raw material just waiting to be optimized. And this points to the very abolition of man that Lewis wrote about. It reduces people to programmable entities devoid of true humanity, all simply because we have the means to do so with our scientific and technological “progress”.

---

This has been some light reading, I know. If you’ve made it this far, you are probably, like me, exhausted. Don’t get me wrong, when it comes to the advancements in AI I talked about, I am mostly excited. Change is inevitable, we can adapt. The “superpowers” this tech can bring us could lead to many exciting new changes to how we live and work. While much of the vision and direction of this technology is out of our control, and instead controlled by a relatively small set of business people and developers, its still up to us to decide how we use it to ensure we preserve our humanity.