AI Agent newsletter - week ending 8.1.25
AI consciousness? And AI solutions as category versus deployed intelligence.
Before I get into category v deployment, a brief detour into my sojourn this week for a talk at the California Institute for Machine Consciousness. The name sounds a little dramatic but the organization is full of serious people from our most prestigious institutions. The talk was about AI and Future of Humanity. Personally, I have zero expectation that machines will become conscious but there's not doubt people will experience them as conscious the more they act, sound and look like people. And this has deep implications in how we integrate these artificial beings into our workflows and businesses.
The talk itself was a bit jagged and self indulgent but there are very bright, albeit totally techno optimistic, people out there who think we can some how create aware digital beings. This is not the space to delve into why I don't think that's true and why it relies on a very different idea of consciousness but if the emergence theory is correct i.e. consciousness emerges from sufficiently complex brain states, then I could be proven wrong.
Now back to AI product categories versus treating it as intelligence to be deployed. The latter to me is what we primarily focus on and what I will focus on this week. We ARE working on some cloud based deployable solutions also i.e. category products but our broad thesis and contention is that companies need to reimagine workflows and embrace the idea that AI can be deployed in a more pervasive manner, as agents. That means they can perform tasks, sit across multiple systems and data sets, with various tools at their disposal and integrate into broad workflows. Layer over this an ability to orchestrate while using the UIs already in place e.g. Teams, Slack, email etc, and firms can integrate intelligence that people can interface with in a repeatable, scalable and structured way but which is not bound by a single system.
A useful framework I saw recently was this idea of agents as not end-to-end solutions but middle-to-middle solutions. AI needs prompting and it needs validation or review. To execute this, they need abilities that span systems, data sets etc. A very good example is an ambient agent for a sales or operational rep. It monitors emails, understands vendors and customers, has hooks into the company CRM and maybe financials, and can then instantly draft a response to an incoming customer request and as well any external parties e.g. vendors, set reminders or meeting invites, and request user approval of out bound communications. The human here acts more as a reviewer, and spending more time focused on high end and more complex research or customer support.
Some good further elaboration on these idea are captured below:
- Tts increasingly about orchestration - link. full video here
- Embrace the middle to middle approach - link
Signal:
- Working with AI: Measuring the Occupational Implications of Generative AI
People / teams I follow constantly for insight:
- Aaron Levie - CEO of Box. Saying a lot of smart things about how AI in enterprise will play out.
- 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
