AI Agent Newsletter. - Week of 7.7.25

How cognitive science can inform how we think about AI.

When I started this newsletter recently I said:

Humans are deeply complex creatures and have different ways of knowing and being in the world but most of our work requires a narrow set of capabilities i.e. our ability to applying routines and skills to numbers and facts.

I wanted to lay out a little more about what informs me or helps me think about how we all should think about AI and how it will impact us.

One of my primary inspirations for this statement comes from one of my constant sources of insight, the cognitive scientist John Vervaeke. Cog sci has gone through a revolution in recent years, and he espouses a field of growing influence in cog sci called the 4E model. The very short definition of this model is that the mind and cognition is embodied, embedded, extended, and enacted. It's in our bodies, tools, cultures and between us all. That we are not, as espoused by the computationalist or representationalist model, just bodies with information processing capabilities somewhere up in our heads.

John's more user friendly (and slightly more philosophical) version of this is something he calls the 4P model, which I explain in a little more here on slide 4. This model says we have 4 ways of knowing and being in the world and making contact with reality:

• Propositional i.e. knowing facts. Accessed via our semantic memory
• Procedural i.e. knowing how. Accessed via procedural memory
• Perspectival i.e. knowing what it’s like to be. Access via episodic memory, and
• Participatory i.e. knowing consciously. Accessed via our sense of self.

This model has all sorts of implications for how we think about ourselves however for our purposes, the primary implications and concern as relates to AI is that:

• The majority of most people's work is propositional and procedural.
• AI is very rapidly becoming really good at both of these and even mimicking perspectival knowing also (this is where prompt and context engineering play a very important role).

For anyone who works with LLMs and AI Agent development, it's not hard to map these Ps onto how we approach and the function of things like memory stores, prompt and context engineering, few shot prompting etc.

But for all of us, relationships (commercial and personal), the ability to tell stories and weave narrative, self awareness and empathy, these capabilities will have even more value now as AI takes on more of the drudgery work. AI also represents however a kind of accelerant of the 4E model. More and more we will rely on AI as an extension of ourselves to get more and more done. I think this commentary is right however:

• That we'll just do more "perspectival" work, that we previously spread over along period as AI does more of the procedural and propositional work.

I think AI won't just take our work and we'll do the same. We'll just lean more and more the non-operational stuff.

Anyway, hope this lays out some of how I both think about my self while also helping me frame where we're heading with AI.

Signal versus Noise section...

• Signal
  • More signal of AI eating entry level and other propositional & procedural work:
    • Big 4 reducing entry level roles - link
• More job lose predictions from the CEO of Ford and others industry leaders- link
• Using the front end of the foundational model companies isn't probably best for data privacy (anecdotal but interesting):
  • ChatGPT regurgitating board minutes. This could be probabilistically but that doesn't make it any less invasive.
• AI is smart but it doesn't reason exactly. Good podcast on one of the main benchmarks out there - ARC - for simply testing whether AI is actually "thinking". Check it out.
• Finally, AI's ability to mimic our perspectival selves can easily go off the rails, as Grok (and the rest of the world) learnt this week. This video is a tour de force of why getting the basics right means everything.
• Noise:
  • Much as I would love to revel is this article rain schadenfreude on my old industry - because it deserves it & much of what is said here about their dubious ROI is true - the big 4 / management consulting firms I believe won't, at least in the short to medium and probably even in the long term, get smashed by AI. Anyone who works with AI knows statements like - “It used to take two weeks to do a SWOT analysis with all the people engaged in doing research,” said Soren Kaplan, an innovation expert, “Now it takes two minutes with AI" - are wildly misleading. Major factors for their perpetual survival will be:
    • They'll make $billions from AI advisory and implementations
    • The entry level jobs will get hit initially which will only feed margins (see signal link above)
    • The economy isn't some stagnant beast where we'll just swap out like for like jobs for AI and
    • Long term, companies will still need accounting & regulatory related projects as well as project plausible deniability.

People / teams I follow constantly for insight:

• Andrej Karpathy - Technical genius
• Nate Jones - Product manager / AI Agent deep diver
• All-in pod - Can be nauseating but they are smart and connected
• BG2 pod - Like All-in but less nauseating
• John Vervaeke - The most important public intellectual on matters of the mind, cognition, meaning and what it means to be human.
• More to come...

Ciao for now. Have a great weekend.

Contact details

Email: tom@bundlrco.com

LinkedIn