How AI Evolves

🧠 Insights You Won’t Forget

Today's insights are inspired by a recent episode of AI and I on Intentional Technology

1. Chatbots are a feature, not a paradigm

   Chat interfaces are just one early manifestation of LLMs, useful for initiating unstructured tasks, but inadequate for structured, long-lived work. The real power lies in co-active systems that collaborate with users across persistent digital substrates.

2. The future hinges on intentional technology

   We must choose between engagement-maximizing hyper-aggregation (like social media algorithms) and tech that aligns with stated intentions (e.g., spending time with family, learning, challenging beliefs). LLMs can enable a renaissance in human agency if built the right way.

3. The four pillars of intentional tech

   A new AI-powered computing model should be:

   • Human-centered (not corporate-centered)

   • Private by design

   • Pro-social and integrative

   • Open-ended and modifiable

   These pillars ensure systems empower users rather than exploit them.

4. LLMs as cognitive exoskeletons (exocortex)

   LLMs should serve as personalized, private cognitive extensions of ourselves. If not aligned with your intentions or misused by corporations, they become dystopian surveillance tools.

5. Why ChatGPT is AOL

   Just as AOL introduced many to the internet but was ultimately surpassed by the open web, ChatGPT is seen as an early, closed interface to LLMs, useful but not the end state. The open LLM ecosystem will eventually dominate due to combinatorial innovation.

6. The Same-Origin Paradigm is holding us back

   This legacy security model isolates data by app or domain, creating “data gravity” that leads to aggregation and silos. Replacing it with contextual flow control could unlock richer, more integrated user experiences and AI collaboration.

7. Prompt injection is the next AI security crisis

   Prompt injection exploits the gullibility of LLMs and threatens integrated systems. Unlike SQL injection, it is harder to defend against and may require foundational architectural changes, especially in systems with irreversible actions (e.g., agents).

8. LLMs transfer tacit knowledge like no tool before

   They can encode and transmit “know-how” that humans can’t easily explain. This creates opportunities for massive learning acceleration, richer communication, and more personalized interactions.

9. Smuggled infinities & the danger of perfect assumptions

   Assuming perfect AGI or agent behavior (“they’ll never make a mistake”) is flawed. Even one high-stakes failure (e.g., buying $5,000 eggs) invalidates the system. We must build resilient systems that account for imperfections.

10. New system architecture is already possible

    Confidential compute and open attested runtimes offer a middle ground: cloud-based, encrypted VMs that run trusted, verifiable software, allowing private, user-aligned AI experiences without sacrificing usability.

Recall from last week

1. Demographics are quietly rewriting global growth assumptions

   The global worker pool is shrinking, China’s working-age population is projected to decline 1% per year to 2050. This demographic reversal erodes the foundational labor-driven growth model that fueled decades of prosperity.

2. Debt sustainability is a long fuse, not a spark

   Despite debt levels exceeding WWII-era highs, demand for safe, liquid sovereign assets remains strong. The real risk lies not in default but in the slow erosion of the dollar’s role as a truly risk-free asset. The U.S. now spends more servicing debt than on defense, a historically destabilizing threshold.

💡 Eko Worth Remembering

⚡ Active Recall – Test Yourself 

🛤️ Off the Record

As mentioned on Friday, here are the full insights from the episode!

As someone who works full time in crypto I wanted to relate the Same Origin Paradigm to how I think it affects the Crypto AI space:

As AI and crypto systems increasingly converge, their intersection is revealing more than just technological synergy, it exposes a shared commitment to values like transparency, sovereignty, and verifiability. This convergence isn’t accidental. It reflects a deliberate push for systems that prioritize user agency and secure collaboration. Central to this vision is the Same-Origin Paradigm (SOP), a browser-era security model that enforces strict boundaries between different web origins. SOP’s principle of isolation-by-default and permission-by-design offers more than just security hygiene; it provides a conceptual framework that can inform how we build intentional, trust-aware Crypto x AI architectures.

At its core, SOP enforces separation: different web origins cannot interact unless explicitly permitted. In Crypto x AI ecosystems, this maps well onto the way agent identities like wallets and smart contracts interact under cryptographic rules. Each agent, like a web origin, operates within bounded authority, with explicit permission required for cross-agent communication or data access. Architectures like model enclaves, wallet-bound agents, and onchain authorization of inference jobs mirror SOP’s boundary enforcement. Both paradigms embody a commitment to making system boundaries legible, enforceable, and auditable, key traits for building systems where trust cannot be assumed but must be cryptographically verified.

This design choice of default separation and explicit consent carries ethical and functional weight. Just as SOP uses mechanisms like CORS or postMessage to enable intentional cross-origin interactions, Crypto x AI systems must encode consent through signed messages, scoped permissions, and verifiable collaborations. By embracing SOP as both metaphor and architectural compass, we move toward building a better future where these new systems are not only decentralized, but also composable, secure, and designed with intentionality from the ground up.

Eko’s Top Pods

Answer:

It ensures web safety by isolating app data but limits integration across tools, reinforcing aggregation and silos. A more flexible model like contextual flow control could enable secure data interoperability and richer, user-centric AI systems.

Enjoyed these insights? Forward this newsletter to a friend. Let’s grow smarter, together.